Dissertations / Theses: 'Classifications des images'

1

Sonoda, Luke Ienari. "Classifications of lesions in magnetic resonance images of the breast." Thesis, King's College London (University of London), 2003. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.406934.

Full text

APA, Harvard, Vancouver, ISO, and other styles

2

Thompson, J. Paul. "Classifications of gross morphologic and magnetic resonance images of human intervertebral discs." Thesis, University of British Columbia, 1987. http://hdl.handle.net/2429/26647.

Full text

Abstract:

The pathogenesis of low back pain is complex but likely involves the intervertebral disc (Nachemson, 1976). Direct evidence for its importance is lacking because an accurate in vivo method of imaging the lumbar intervertebral disc has not been established. The objective of this research was to develop classifications of gross morphologic appearance and magnetic resonance image (MRI) of the disc that describe the changes associated with aging and degeneration, thereby permitting interpretation of the MRI in terms of gross morphology and allowing correlation of morphologic, chemical, mechanical, radiologic and epidemiologic data with a standard reference of disc aging and degeneration. The classifications were developed on the basis of literature review, detailed examination of 55 discs and expert advice. Two sets of three observers, one for the morphologic classification and one for the MRI classification evaluated 68 life size randomized duplicates of discs making detailed observations about overall category and 17 regional morphologic parameters and 11 regional MRI parameters. The data was tested to demonstrate the validity of the classifications using established criteria (Tugwell & Bombardier, 1982; Guyatt 4 Kirschner, 1985; Feinstein, 1985). The consistency with which the classifications could be applied was evaluated by calculating weighted kappa, a statistical test of agreement that corrects for agreement by chance; the ability of the classifications to distinguish stages in the process of ageing and degeneration by stepwise discriminant analysis; their conformity with other measures by comparisons within and between classifications and, comparisons with histologic and chemical data. The degree of agreement for all six intra-observer pairs was 'almost perfect' (weighted kappa > 0.80); for 5 interobserver pairs 'substantial' (weighted kappa > 0.60) and for one MRI interobserver pair 'moderate' (weighted kappa > 0.50). This represented a satisfactory level of agreement and indicated the classifications could be applied consistently (Feinstein, 1981). The linear regression model developed by stepwise discriminant analysis clearly demonstrated the ability of the classifications to distinguish distinct stages in disc aging and degeneration. Wilk's lambda, a likelihood ratio statistic reflecting discriminatory function, approached zero in both the morphologic (0.0408) and MRI (0.0H80) classifications. In both models, parameters pertaining to the nucleus pulposus of the disc accounted for the majority of the variance (morphologic partial R² 0.8598 and MRI partial R² 0.8811) suggesting nuclear parameters are the most important in distinguishing overall category. The correlation table generated by principal component analysis demonstrated that the categories assigned to regional parameters correlated significantly (p > 0.0001) with each other and with the overall category. From the linear combinations of parameters (principal components) generated the weighting of the nucleus pulposus behaved independently attesting to its importance. Comparisons of the morphologic and MRI classifications yielded high indices of trend (Pearson correlation coefficient of 0.81) and concordance (kappa of 0.62). Trends in the histologic and chemical data were consistent with the classifications but could not be evaluated statistically because only 15 specimens were studied. This research suggests that the classifications are valid and will form a basis for the interpretation of MRI. Preliminary evidence suggested MRI is sensitive to early changes in extracellular matrix composition not apparent in gross morphology.
Medicine, Faculty of
Graduate

APA, Harvard, Vancouver, ISO, and other styles

3

Arshad, Irshad Ahmad. "Using statistical methods for automatic classifications of clouds in ground-based photographs of the sky." Thesis, University of Essex, 2003. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.250129.

Full text

APA, Harvard, Vancouver, ISO, and other styles

4

Ngo, Ho Anh Khoi. "Méthodes de classifications dynamiques et incrémentales : application à la numérisation cognitive d'images de documents." Thesis, Tours, 2015. http://www.theses.fr/2015TOUR4006/document.

Full text

Abstract:

Cette thèse s’intéresse à la problématique de la classification dynamique en environnements stationnaires et non stationnaires, tolérante aux variations de quantités des données d’apprentissage et capable d’ajuster ses modèles selon la variabilité des données entrantes. Pour cela, nous proposons une solution faisant cohabiter des classificateurs one-class SVM indépendants ayant chacun leur propre procédure d’apprentissage incrémentale et par conséquent, ne subissant pas d’influences croisées pouvant émaner de la configuration des modèles des autres classificateurs. L’originalité de notre proposition repose sur l’exploitation des anciennes connaissances conservées dans les modèles de SVM (historique propre à chaque SVM représenté par l’ensemble des vecteurs supports trouvés) et leur combinaison avec les connaissances apportées par les nouvelles données au moment de leur arrivée. Le modèle de classification proposé (mOC-iSVM) sera exploité à travers trois variations exploitant chacune différemment l’historique des modèles. Notre contribution s’inscrit dans un état de l’art ne proposant pas à ce jour de solutions permettant de traiter à la fois la dérive de concepts, l’ajout ou la suppression de concepts, la fusion ou division de concepts, tout en offrant un cadre privilégié d’interactions avec l’utilisateur. Dans le cadre du projet ANR DIGIDOC, notre approche a été appliquée sur plusieurs scénarios de classification de flux d’images pouvant survenir dans des cas réels lors de campagnes de numérisation. Ces scénarios ont permis de valider une exploitation interactive de notre solution de classification incrémentale pour classifier des images arrivant en flux afin d’améliorer la qualité des images numérisées
This research contributes to the field of dynamic learning and classification in case of stationary and non-stationary environments. The goal of this PhD is to define a new classification framework able to deal with very small learning dataset at the beginning of the process and with abilities to adjust itself according to the variability of the incoming data inside a stream. For that purpose, we propose a solution based on a combination of independent one-class SVM classifiers having each one their own incremental learning procedure. Consequently, each classifier is not sensitive to crossed influences which can emanate from the configuration of the models of the other classifiers. The originality of our proposal comes from the use of the former knowledge kept in the SVM models (represented by all the found support vectors) and its combination with the new data coming incrementally from the stream. The proposed classification model (mOC-iSVM) is exploited through three variations in the way of using the existing models at each step of time. Our contribution states in a state of the art where no solution is proposed today to handle at the same time, the concept drift, the addition or the deletion of concepts, the fusion or division of concepts while offering a privileged solution for interaction with the user. Inside the DIGIDOC project, our approach was applied to several scenarios of classification of images streams which can correspond to real cases in digitalization projects. These different scenarios allow validating an interactive exploitation of our solution of incremental classification to classify images coming in a stream in order to improve the quality of the digitized images

APA, Harvard, Vancouver, ISO, and other styles

5

Alchicha, Élie. "Confidentialité Différentielle et Blowfish appliquées sur des bases de données graphiques, transactionnelles et images." Thesis, Pau, 2021. http://www.theses.fr/2021PAUU3067.

Full text

Abstract:

Les données numériques jouent un rôle crucial dans notre vie quotidienne en communiquant, en enregistrant des informations, en exprimant nos pensées et nos opinions et en capturant nos moments précieux sous forme d'images et de vidéos numériques. Les données numériques présentent d'énormes avantages dans tous les aspects de la vie moderne, mais constituent également une menace pour notre vie privée. Dans cette thèse, nous considérons trois types de données numériques en ligne générées par les utilisateurs des médias sociaux et les clients du commerce électronique : les graphiques, les transactions et les images. Les graphiques sont des enregistrements des interactions entre les utilisateurs qui aident les entreprises à comprendre qui sont les utilisateurs influents dans leur environnement. Les photos postées sur les réseaux sociaux sont une source importante de données qui nécessitent des efforts d'extraction. Les ensembles de données transactionnelles représentent les opérations qui ont eu lieu sur les services de commerce électronique.Nous nous appuyons sur une technique de préservation de la vie privée appelée Differential Privacy (DP) et sa généralisation Blowfish Privacy (BP) pour proposer plusieurs solutions permettant aux propriétaires de données de bénéficier de leurs ensembles de données sans risque de violation de la vie privée pouvant entraîner des problèmes juridiques. Ces techniques sont basées sur l'idée de récupérer l'existence ou la non-existence de tout élément dans l'ensemble de données (tuple, ligne, bord, nœud, image, vecteur, ...) en ajoutant respectivement un petit bruit sur la sortie pour fournir un bon équilibre entre intimité et utilité.Dans le premier cas d'utilisation, nous nous concentrons sur les graphes en proposant trois mécanismes différents pour protéger les données personnelles des utilisateurs avant d'analyser les jeux de données. Pour le premier mécanisme, nous présentons un scénario pour protéger les connexions entre les utilisateurs avec une nouvelle approche où les utilisateurs ont des privilèges différents : les utilisateurs VIP ont besoin d'un niveau de confidentialité plus élevé que les utilisateurs standard. Le scénario du deuxième mécanisme est centré sur la protection d'un groupe de personnes (sous-graphes) au lieu de nœuds ou d'arêtes dans un type de graphes plus avancé appelé graphes dynamiques où les nœuds et les arêtes peuvent changer à chaque intervalle de temps. Dans le troisième scénario, nous continuons à nous concentrer sur les graphiques dynamiques, mais cette fois, les adversaires sont plus agressifs que les deux derniers scénarios car ils plantent de faux comptes dans les graphiques dynamiques pour se connecter à des utilisateurs honnêtes et essayer de révéler leurs nœuds représentatifs dans le graphique.Dans le deuxième cas d'utilisation, nous contribuons dans le domaine des données transactionnelles en présentant un mécanisme existant appelé Safe Grouping. Il repose sur le regroupement des tuples de manière à masquer les corrélations entre eux que l'adversaire pourrait utiliser pour violer la vie privée des utilisateurs. D'un autre côté, ces corrélations sont importantes pour les propriétaires de données dans l'analyse des données pour comprendre qui pourrait être intéressé par des produits, biens ou services similaires. Pour cette raison, nous proposons un nouveau mécanisme qui expose ces corrélations dans de tels ensembles de données, et nous prouvons que le niveau de confidentialité est similaire au niveau fourni par Safe Grouping.Le troisième cas d'usage concerne les images postées par les utilisateurs sur les réseaux sociaux. Nous proposons un mécanisme de préservation de la confidentialité qui permet aux propriétaires des données de classer les éléments des photos sans révéler d'informations sensibles. Nous présentons un scénario d'extraction des sentiments sur les visages en interdisant aux adversaires de reconnaître l'identité des personnes
Digital data is playing crucial role in our daily life in communicating, saving information, expressing our thoughts and opinions and capturing our precious moments as digital pictures and videos. Digital data has enormous benefits in all the aspects of modern life but forms also a threat to our privacy. In this thesis, we consider three types of online digital data generated by users of social media and e-commerce customers: graphs, transactional, and images. The graphs are records of the interactions between users that help the companies understand who are the influential users in their surroundings. The photos posted on social networks are an important source of data that need efforts to extract. The transactional datasets represent the operations that occurred on e-commerce services.We rely on a privacy-preserving technique called Differential Privacy (DP) and its generalization Blowfish Privacy (BP) to propose several solutions for the data owners to benefit from their datasets without the risk of privacy breach that could lead to legal issues. These techniques are based on the idea of recovering the existence or non-existence of any element in the dataset (tuple, row, edge, node, image, vector, ...) by adding respectively small noise on the output to provide a good balance between privacy and utility.In the first use case, we focus on the graphs by proposing three different mechanisms to protect the users' personal data before analyzing the datasets. For the first mechanism, we present a scenario to protect the connections between users (the edges in the graph) with a new approach where the users have different privileges: the VIP users need a higher level of privacy than standard users. The scenario for the second mechanism is centered on protecting a group of people (subgraphs) instead of nodes or edges in a more advanced type of graphs called dynamic graphs where the nodes and the edges might change in each time interval. In the third scenario, we keep focusing on dynamic graphs, but this time the adversaries are more aggressive than the past two scenarios as they are planting fake accounts in the dynamic graphs to connect to honest users and try to reveal their representative nodes in the graph. In the second use case, we contribute in the domain of transactional data by presenting an existed mechanism called Safe Grouping. It relies on grouping the tuples in such a way that hides the correlations between them that the adversary could use to breach the privacy of the users. On the other side, these correlations are important for the data owners in analyzing the data to understand who might be interested in similar products, goods or services. For this reason, we propose a new mechanism that exposes these correlations in such datasets, and we prove that the level of privacy is similar to the level provided by Safe Grouping.The third use-case concerns the images posted by users on social networks. We propose a privacy-preserving mechanism that allows the data owners to classify the elements in the photos without revealing sensitive information. We present a scenario of extracting the sentiments on the faces with forbidding the adversaries from recognizing the identity of the persons. For each use-case, we present the results of the experiments that prove that our algorithms can provide a good balance between privacy and utility and that they outperform existing solutions at least in one of these two concepts

APA, Harvard, Vancouver, ISO, and other styles

6

Thornström, Johan. "Domain Adaptation of Unreal Images for Image Classification." Thesis, Linköpings universitet, Datorseende, 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-165758.

Full text

Abstract:

Deep learning has been intensively researched in computer vision tasks like im-age classification. Collecting and labeling images that these neural networks aretrained on is labor-intensive, which is why alternative methods of collecting im-ages are of interest. Virtual environments allow rendering images and automaticlabeling, which could speed up the process of generating training data and re-duce costs.This thesis studies the problem of transfer learning in image classificationwhen the classifier has been trained on rendered images using a game engine andtested on real images. The goal is to render images using a game engine to createa classifier that can separate images depicting people wearing civilian clothingor camouflage. The thesis also studies how domain adaptation techniques usinggenerative adversarial networks could be used to improve the performance ofthe classifier. Experiments show that it is possible to generate images that canbe used for training a classifier capable of separating the two classes. However,the experiments with domain adaptation were unsuccessful. It is instead recom-mended to improve the quality of the rendered images in terms of features usedin the target domain to achieve better results.

APA, Harvard, Vancouver, ISO, and other styles

7

Pavez, Ojeda Jorge. "Africanismes à Cuba (1812-1917) : textes, images et classes." Paris, EHESS, 2007. http://www.theses.fr/2007EHES0097.

Full text

Abstract:

Cette thèse aborde la constitution du champ des études afro-cubaines au début du XXè siècle par l'œuvre de Fernando Ortiz, avocat, ethnologue, historien et folkloriste. On y verra la tension entre les logiques disciplinaires européennes et les formes de co-production des savoirs avec les sujets afro-cubains qui participent du processus ethnographique. Il est ainsi proposé une déconstruction des principaux sujets et notions sur lesquels s'instituent un regard « scientifique » sur l'Afrique à Cuba : sorcellerie, dégénérescence, « pègre », classements ethniques, écritures afro-cubaines (tatouages, symbolisme, musique, cultes et rites). L'accent mis sur les classes et les systèmes de classification des disciplines médicales et sociales mènera à une généalogie des conceptions de classe et de race noire adoptées par les Afro-cubains. Pour ce faire, on proposera l'analyse d'un corpus d'archives sur l'intellectuel et artiste afro-cubain José Antonio Aponte, exécuté en 1812 comme conspirateur
This dissertation analyzes the constitution of the field of Afro-Cuban Studies at the beginnings of the XXth century in the work of Fernando Ortiz, criminal lawyer, ethnologist, historian and folklorist. We will find in it the tension between the European logics of disciplines and the forms of Afro Cuban agency in the co-production of ethnographical knowledge. In that way, we propose a deconstruction of the principals subjects and concepts on which is instituted a vision of Africa in Cuba: witchcraft, degeneration, "mob", ethnic classifications, Afro-Cubans' writings (tattoos, symbolisms, music, cults and rites). The accent on the classes and the classifications systems of social and medical disciplines will lead to a genealogy of the conceptions of black class and race adopted by the Afro-Cubans. For this, we will propose the analysis of a corpus of archives about the Afro Cuban artist and intellectual Jose Antonio Aponte, accused and executed in 1812 as conspirator and rebel

APA, Harvard, Vancouver, ISO, and other styles

8

BATISTA, LEONARDO VIDAL. "COMPARING AUTOMATIC IMAGE CLASSIFICATION TECHNIQUES OF REMOTE SENSING IMAGES." PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO, 1993. http://www.maxwell.vrac.puc-rio.br/Busca_etds.php?strSecao=resultado&nrSeq=8870@1.

Full text

Abstract:

CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO
Neste trabalho, diversas técnicas de classificação automática de imagens de sensoriamento remoto são investigadas. Na análise, incluem-se um método não- paramétrico, denominado K-Médias. Adaptativos Hierárquico (KMAH), e seis paramétricos: o Classificador de Máxima Verossimilhança (MV), o de Máxima Probabilidade a Posteriori (MAP), o MAP Adaptativo (MAPA), por Subimagens (MAPSI), o Contextual Tilton-Swain (CXTS) e o Contextual por Subimagens (CXSI). O treinamento necessário à implementação das técnicas paramétricas foi realizado de forma não-supervisionada, usando-se para tanto a classificação efetuada pelo KMAH. Considerações a respeito das vantagens e desvantagens dos classificadores, de acordo com a observação das taxas de erros e dos tempos de processamento, apontaram as técnicas MAPA e MAPSI com as mais convenientes
In this thesis, several techniques of automatic classfication of remote sensing impeages are investigated. Included in the analysis are ane non-parametric method, known as Adaptative hierarchical K-means (KMAH), and six parametric ones: the Maximum Likelihood (MV), the Maximum a Posteriori Probability (MAP), the Adaptative MAP (MAPA), the Subimages MAP (MAPSI), the tilton-Swain Contextual, (CXTS) and the Subimages Contextual (CXSI) classifiers. The necessary training for the parametric case was done in a non-supervised form, by using the KMAH classification. Considerations about the advantages and disadvantages of the classifiers were made and, based on the observation of the error rates and processing time, the MAPA and MAPSI have shown the best performances.

APA, Harvard, Vancouver, ISO, and other styles

9

Råhlén, Oskar, and Sacharias Sjöqvist. "Image Classification of Real Estate Images with Transfer Learning." Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-259759.

Full text

Abstract:

Each minute, over 2 000 searches are made on Sweden’s largest real estate website. The site has over 20 000 apartments for sale in the Stockholm region alone. This makes the search-function a vital tool for the users to find their dream apartment, and thus the quality of the search-function is of significance. As of today, it’s only possible to filter and sort by meta-data such as number of rooms, living area, price, and location, but not on more complex attributes, such as balcony or fireplace. To prevent the need for manual categorization of objects on the market, one option could be to use images of the apartments as data-points in deep neural networks to automatically add rich attributes. This thesis aims to investigate if a high rate of success when classifying apartment images can be achieved using deep neural networks, specifically looking at the categories and attributes balcony, fireplace, as well as type of room. Different types of architectures was compared amongst each other and feature extraction was compared against fine-tuning, in order to exhaustively investigate the thesis. The investigation showed that the balcony model could determine if a balcony exists in an image, with a certainty of 98.1%. For fireplaces, the maximum certainty reached was 85.5%, which is significantly lower. The type-of-room classification reached a certainty of 97,9%. This all proves the possibility of using deep neural networks in order to classify and attribute real estate images.
Varje minut görs 2000 sökningar på Sveriges största webbplats för bostadsannonser som har 20 000 bostadsrätter till salu bara i Stockholm. Detta ställer höga krav på sökfunktionen för att ge användarna en chans att hitta sin drömbostad. Idag finns det möjlighet att filtrera på attribut såsom antal rum, boarea, pris och område men inte på attribut som balkong och eldstad. För att inte behöva kategorisera objekt manuellt för attribut såsom balkong och eldstad finns det möjlighet att använda sig av mäklarbilder samt djupa neurala nätverk för att klassificera objekten automatiskt. Denna uppsats syftar till att utreda om det med hög sannolikhet går att klassificera mäklarbilder efter attributen balkong, eldstad samt typ av rum, med hjälp av djupa neurala nätverk. För att undersöka detta på ett utförligt sätt jämfördes olika arkitekturer med varandra samt feature extraction mot fine-tuning. Testerna visade att balkongmodellen med 98,1% sannolikhet kan avgöra om det finns en balkong på någon av bilderna eller inte. För eldstäder nåddes ett maximum på 85,5% vilket är väsentligt sämre än för balkonger. Under sista klassificeringen, den för rum, nåddes ett resultat på 97,9%.Sammanfattningsvis påvisar detta att det är fullt möjligt att använda djupa neurala nätverk för att klassificera mäklarbilder.

APA, Harvard, Vancouver, ISO, and other styles

10

Vargas, Muñoz John Edgar 1991. "Contextual superpixel-based active learning for remote sensing image classification = Aprendizado ativo baseado em atributos contextuais de superpixel para classificação de imagem de sensoriamento remoto." [s.n.], 2015. http://repositorio.unicamp.br/jspui/handle/REPOSIP/275555.

Full text

Abstract:

Orientadores: Alexandre Xavier Falcão, Jefersson Alex dos Santos
Dissertação (mestrado) - Universidade Estadual de Campinas, Instituto de Computação
Made available in DSpace on 2018-08-27T14:43:51Z (GMT). No. of bitstreams: 1 VargasMunoz_JohnEdgar_M.pdf: 9138091 bytes, checksum: bdb40e3a5655df0e10a137f2d08f0d8d (MD5) Previous issue date: 2015
Resumo: Recentemente, técnicas de aprendizado de máquina têm sido propostas para criar mapas temáticos a partir de imagens de sensoriamento remoto. Estas técnicas podem ser divididas em métodos de classificação baseados em pixels ou regiões. Este trabalho concentra-se na segunda abordagem, uma vez que estamos interessados em imagens com milhões de pixels e a segmentação da imagem em regiões (superpixels) pode reduzir consideravelmente o número de amostras a serem classificadas. Porém, mesmo utilizando superpixels, o número de amostras ainda é grande para anotá-las manualmente e treinar o classificador. As técnicas de aprendizado ativo propostas resolvem este problema começando pela seleção de um conjunto pequeno de amostras selecionadas aleatoriamente. Tais amostras são anotadas manualmente e utilizadas para treinar a primeira instância do classificador. Em cada iteração do ciclo de aprendizagem, o classificador atribui rótulos e seleciona as amostras mais informativas para a correção/confirmação pelo usuário, aumentando o tamanho do conjunto de treinamento. A instância do classificador é melhorada no final de cada iteração pelo seu treinamento e utilizada na iteração seguinte até que o usuário esteja satisfeito com o classificador. Observamos que a maior parte dos métodos reclassificam o conjunto inteiro de dados em cada iteração do ciclo de aprendizagem, tornando este processo inviável para interação com o usuário. Portanto, enderaçamos dois problemas importantes em classificação baseada em regiões de imagens de sensoriamento remoto: (a) a descrição efetiva de superpixels e (b) a redução do tempo requerido para seleção de amostras em aprendizado ativo. Primeiro, propusemos um descritor contextual de superpixels baseado na técnica de sacola de palavras, que melhora o resultado de descritores de cor e textura amplamente utilizados. Posteriormente, propusemos um método supervisionado de redução do conjunto de dados que é baseado em um método do estado da arte em aprendizado ativo chamado Multi-Class Level Uncertainty (MCLU). Nosso método mostrou-se tão eficaz quanto o MCLU e ao mesmo tempo consideravelmente mais eficiente. Adicionalmente, melhoramos seu desempenho por meio da aplicação de um processo de relaxação no mapa de classificação, utilizando Campos Aleatórios de Markov
Abstract: In recent years, machine learning techniques have been proposed to create classification maps from remote sensing images. These techniques can be divided into pixel- and region-based image classification methods. This work concentrates on the second approach, since we are interested in images with millions of pixels and the segmentation of the image into regions (superpixels) can considerably reduce the number of samples for classification. However, even using superpixels the number of samples is still large for manual annotation of samples to train the classifier. Active learning techniques have been proposed to address the problem by starting from a small set of randomly selected samples, which are manually labeled and used to train a first instance of the classifier. At each learning iteration, the classifier assigns labels and selects the most informative samples for user correction/confirmation, increasing the size of the training set. An improved instance of the classifier is created by training, after each iteration, and used in the next iteration until the user is satisfied with the classifier. We observed that most methods reclassify the entire pool of unlabeled samples at every learning iteration, making the process unfeasible for user interaction. Therefore, we address two important problems in region-based classification of remote sensing images: (a) the effective superpixel description and (b) the reduction of the time required for sample selection in active learning. First, we propose a contextual superpixel descriptor, based on bag of visual words, that outperforms widely used color and texture descriptors. Second, we propose a supervised method for dataset reduction that is based on a state-of-art active learning technique, called Multi-Class Level Uncertainty (MCLU). Our method has shown to be as effective as MCLU, while being considerably more efficient. Additionally, we further improve its performance by applying a relaxation process on the classification map by using Markov Random Fields
Mestrado
Ciência da Computação
Mestre em Ciência da Computação

APA, Harvard, Vancouver, ISO, and other styles

11

Udas, Swati. "Classification algorithms for finding the eye fixation from digital images /." free to MU campus, to others for purchase, 2003. http://wwwlib.umi.com/cr/mo/fullcit?p1418072.

Full text

APA, Harvard, Vancouver, ISO, and other styles

12

Masse, Antoine. "Développement et automatisation de méthodes de classification à partir de séries temporelles d'images de télédétection - Application aux changements d'occupation des sols et à l'estimation du bilan carbone." Phd thesis, Université Paul Sabatier - Toulouse III, 2013. http://tel.archives-ouvertes.fr/tel-00921853.

Full text

Abstract:

La quantité de données de télédétection archivées est de plus en plus importante et grâce aux nouveaux et futurs satellites, ces données offriront une plus grande diversité de caractéristiques : spectrale, temporelle, résolution spatiale et superficie de l'emprise du satellite. Cependant, il n'existe pas de méthode universelle qui maximise la performance des traitements pour tous les types de caractéristiques citées précédemment; chaque méthode ayant ses avantages et ses inconvénients. Les travaux de cette thèse se sont articulés autour de deux grands axes que sont l'amélioration et l'automatisation de la classification d'images de télédétection, dans le but d'obtenir une carte d'occupation des sols la plus fiable possible. En particulier, les travaux ont portés sur la la sélection automatique de données pour la classification supervisée, la fusion automatique d'images issues de classifications supervisées afin de tirer avantage de la complémentarité des données multi-sources et multi-temporelles et la classification automatique basée sur des séries temporelles et spectrales de référence, ce qui permettra la classification de larges zones sans référence spatiale. Les méthodes ont été testées et validées sur un panel de données très variées de : capteurs : optique (Formosat-2, Spot 2/4/5, Landsat 5/7, Worldview-2, Pleiades) et radar (Radarsat,Terrasar-X), résolutions spatiales : de haute à très haute résolution (de 30 mètres à 0.5 mètre), répétitivités temporelles (jusqu'à 46 images par an) et zones d'étude : agricoles (Toulouse, Marne), montagneuses (Pyrénées), arides (Maroc, Algérie). Deux applications majeures ont été possibles grâce à ces nouveaux outils : l'obtention d'un bilan carbone à partir des rotations culturales obtenues sur plusieurs années et la cartographie de la trame verte (espaces écologiques) dans le but d'étudier l'impact du choix du capteur sur la détection de ces éléments.

APA, Harvard, Vancouver, ISO, and other styles

13

Cheriyadat, Anil Meerasa. "Limitations of principal component analysis for dimensionality-reduction for classification of hyperspectral data." Master's thesis, Mississippi State : Mississippi State University, 2003. http://library.msstate.edu/etd/show.asp?etd=etd-11072003-133109.

Full text

APA, Harvard, Vancouver, ISO, and other styles

14

Gormus, Esra Tunc. "Improved classification of remote sensing imagery using image fusion techniques." Thesis, University of Bristol, 2013. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.601185.

Full text

Abstract:

Remote sensing is a quick and inexpensive way of gathering information about the Earth. It enables one to constantly get updated information from satellite images for real-time local and global mapping of environmental changes. Current classification methods used for extracting relevant knowledge from this huge information pool are not very efficient because of the limited training samples and high dimensionality of the images. Information fusion is often used in order to improve the classification accuracy prior or after performing classification. However, these techniques cannot always successfully overcome the aforementioned issues. Therefore, in this thesis, new methods are introduced in order to increase the classification accuracy of remotely sensed data by means of information fusion techniques. This thesis is structured in three parts. In the first part, a novel pixel based image fusion technique is introduced to fuse optical and SAR image data in order to increase classification accuracy. Fused images obtained via conventional fusion methods may not contain enough information for subsequent processing such as classification or feature extraction. The proposed method aims to keep the maximum contextual and spatial information from the source data by exploiting the relationship between spatial domain cumulants and wavelet domain cumulants. The novelty of the method consists in integrating the relationship between spatial and wavelet domain cumulants of the source images into an image fusion process as well as in employing these wavelet cumulants for optimisation of weights in a Cauchy convolution based image fusion scheme. In the second part, a novel feature based image fusion method is proposed in order to increase the classification accuracy of hyperspectral images. An application of Empirical Mode Decomposition (EMD) to wavelet based dimensionality reduction is presented with an aim to generate the smallest set I of features that leads to better classification accuracy compared to single tech! niques. Useful spectral information for hyperspectral image classi6cation can be oj:>tained by applying the Wavelet Transform (WT) to each hyperspectral signature. As EMD has the ability to describe short term spatial changes in frequencies, it helps to get a better understanding of the spatial information of the signal. In order to take advantage of both spectral and spatial information, a novel dimensionality reduction method is introduced, which relies on using the wavelet transform of EMD features. This leads to better class separability and hence to better classification.

APA, Harvard, Vancouver, ISO, and other styles

15

Nyman, Joakim. "Pixel classification of hyperspectral images." Thesis, Uppsala universitet, Institutionen för informationsteknologi, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-353175.

Full text

Abstract:

For sugar producers, it is a major problem to detect contamination of sugar. Doing it manually would not be feasible because of the high demand and would require toomuch labor. This report evaluates if the problem can be solved by using a hyperspectral camera operating in a wavelength range of 400-1000 nm and a spectralresolution of 224. Using the machine learning algorithms Artificial Neural Networkand Support Vector Machine, models were trained on pixels labeled as sugar or different materials of contamination. An autonomous system could be developed to analyze the sugar in real time and remove the contaminated sugar. This paper presents the results from using both Artificial Neural Networks as well as SupportVector Machine. It also addresses the impact of the pre-processing techniques filtering and maximum normalization when applying machine learning algorithms. The results showed that the accuracy can be significantly increased by using a hyperspectral camera instead of a normal camera, especially for plastic materials where using anormal camera gave a precision and recall score of 0.0 while using the hyperspectral camera gave above 0.9. Support Vector Machine performed slightly better than using Artificial Neural Network, especially for plastic material. The filtering and themaximum normalization did not increase the accuracy and could therefore be omitted in favor for performance.

APA, Harvard, Vancouver, ISO, and other styles

16

NETO, ALEXANDRE HENRIQUE LEAL. "UNSUPERVISED CLASSIFICATION OF SATELLITE IMAGES." PONTIFÍCIA UNIVERSIDADE CATÓLICA DO RIO DE JANEIRO, 1994. http://www.maxwell.vrac.puc-rio.br/Busca_etds.php?strSecao=resultado&nrSeq=8497@1.

Full text

Abstract:

CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO
A classificação e segmentação não-supervisionadas de imagens de sensoriamento remoto são examinadas neste trabalho. A classificação é realizada tomando-se como base o critério de Bayes, que busca minimizar o valor esperado do erro de classificação. Os algoritmos desenvolvidos foram propostos pressupondo-se que a estrutura das classes presentes na imagem podem ser bem modeladas por vetores aleatórios guassianos. Os classificadores convencionais, que só levam em conta a informação dos pixels de forma isolada, forma tratados sob a ótica da quantização vetorial. Em particular, foi proposto um algoritmo de classificação com base na quantização vetorial com restrição de entropia. O desempenho das técnicas de classificação é analisado obsevando-se a discrepância entre classificações, comparando-se as imagens classificadas com imagens referencia e classificando-se imagens sintéticas. A taxa de acerto, entre 80% e 95%. Este bom desempenho dos classificadores é limitado pelo fato de, em suas estruturas, levarem em conta a informação dos pixels de forma isolada. Buscamos, através da classificação de segmentos, incorporar informações de contexto em nossos classificadores. A classificação de segmentos levou a taxas de erros inferiores àquelas alcançadas por classificadores baseados em pixels isolados. Um algoritmo de segmentação, que incorpora ao modelo de classificação por pixels a influencia de sua vizinhança através de uma abordagem markoviana, é apresentado.
Unsupervised classification and segmentation of satellite images are examined in this work. The classification is based on Bayes` criterion, which tries to minimize the expected value of the classification error. The algorthms developed were proposed postulating that the classes in the image are well modeled by gaussian random vectors. Conventional classifiers, which take into account only pixelwise information, were treated as vector quantizers. Specifically, it was proposed a classification algorithm based on entropy constrained vector. The behaviour of the classifiers is examined observing the discrepancy between classifications, comparing classified images with reference-images and classifyng sinthetic images. The percentage of pixels whitch are assigned to the same class as in the reference-images ranged from 80,0% to 95,0%. This good behaviour of the classidiers is limited by the fact that, in theirs structures, are taken into account only isolated pixel information. We have sought, by classifying segments, to introduce contextual information into the classifiers structure. The segments classidiers. A segmentation algorithm, which introduces contextual information into pixelwise classifier by a markovian approach, is presented.

APA, Harvard, Vancouver, ISO, and other styles

17

Santos, Jefersson Alex dos 1984. "Semi-automatic classification of remote sensing images = Classificação semi-automática de imagens de sensorimento remoto." [s.n.], 2013. http://repositorio.unicamp.br/jspui/handle/REPOSIP/275630.

Full text

Abstract:

Orientadores: Ricardo da Silva Torres, Alexandre Xavier Falcão
Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Computação
Made available in DSpace on 2018-08-23T15:18:27Z (GMT). No. of bitstreams: 1 Santos_JeferssonAlexdos_D.pdf: 18672412 bytes, checksum: 58ac60d8b5342ab705a78d5c82265ab8 (MD5) Previous issue date: 2013
Resumo: Um grande esforço tem sido feito para desenvolver sistemas de classificação de imagens capazes de criar mapas temáticos de alta qualidade e estabelecer inventários precisos sobre o uso do solo. As peculiaridades das imagens de sensoriamento remoto (ISR), combinados com os desafios tradicionais de classificação de imagens, tornam a classificação de ISRs uma tarefa difícil. Grande parte dos desafios de pesquisa estão relacionados à escala de representação dos dados e, ao mesmo tempo, à dimensão e à representatividade do conjunto de treinamento utilizado. O principal foco desse trabalho está nos problemas relacionados à representação dos dados e à extração de características. O objetivo é desenvolver soluções efetivas para classificação interativa de imagens de sensoriamento remoto. Esse objetivo foi alcançado a partir do desenvolvimento de quatro linhas de pesquisa. A primeira linha de pesquisa está relacionada ao fato de embora descritores de imagens propostos na literatura obterem bons resultados em várias aplicações, muitos deles nunca foram usados para classificação de imagens de sensoriamento remoto. Nessa tese, foram testados doze descritores que codificam propriedades espectrais e sete descritores de textura. Também foi proposta uma metodologia baseada no classificador K-Vizinhos mais Próximos (K-nearest neighbors - KNN) para avaliação de descritores no contexto de classificação. Os descritores Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID) e Quantized Compound Change Histogram (QCCH), apresentaram os melhores resultados experimentais na identificação de alvos de café e pastagem. A segunda linha de pesquisa se refere ao problema de seleção de escalas de segmentação para classificação de imagens de sensoriamento baseada em objetos. Métodos propostos recentemente exploram características extraídas de objetos segmentados para melhorar a classificação de imagens de alta resolução. Entretanto, definir uma escala de segmentação adequada é uma tarefa desafiadora. Nessa tese, foram propostas duas abordagens de classificação multiescala baseadas no algoritmo Adaboost. A primeira abordagem, Multiscale Classifier (MSC), constrói um classificador forte que combina características extraídas de múltiplas escalas de segmentação. A outra, Hierarchical Multiscale Classifier (HMSC), explora a relação hierárquica das regiões segmentadas para melhorar a eficiência sem reduzir a qualidade da classificação xi quando comparada à abordagem MSC. Os experimentos realizados mostram que é melhor usar múltiplas escalas do que utilizar apenas uma escala de segmentação. A correlação entre os descritores e as escalas de segmentação também é analisada e discutida. A terceira linha de pesquisa trata da seleção de amostras de treinamento e do refinamento dos resultados da classificação utilizando segmentação multiescala. Para isso, foi proposto um método interativo para classificação multiescala de imagens de sensoriamento remoto. Esse método utiliza uma estratégia baseada em aprendizado ativo que permite o refinamento dos resultados de classificação pelo usuário ao longo de interações. Os resultados experimentais mostraram que a combinação de escalas produzem melhores resultados do que a utilização de escalas isoladas em um processo de realimentação de relevância. Além disso, o método interativo obtém bons resultados com poucas interações. O método proposto necessita apenas de uma pequena porção do conjunto de treinamento para construir classificadores tão fortes quanto os gerados por um método supervisionado utilizando todo o conjunto de treinamento disponível. A quarta linha de pesquisa se refere à extração de características de uma hierarquia de regiões para classificação multiescala. Assim, foi proposta uma abordagem que explora as relações existentes entre as regiões da hierarquia. Essa abordagem, chamada BoW-Propagation, utiliza o modelo bag-of-visual-word para propagar características ao longo de múltiplas escalas. Essa ideia foi estendida para propagar descritores globais baseados em histogramas, a abordagem H-Propagation. As abordagens propostas aceleram o processo de extração e obtém bons resultados quando comparadas a descritores globais
Abstract: A huge effort has been made in the development of image classification systems with the objective of creating high-quality thematic maps and to establish precise inventories about land cover use. The peculiarities of Remote Sensing Images (RSIs) combined with the traditional image classification challenges make RSI classification a hard task. Many of the problems are related to the representation scale of the data, and to both the size and the representativeness of used training set. In this work, we addressed four research issues in order to develop effective solutions for interactive classification of remote sensing images. The first research issue concerns the fact that image descriptors proposed in the literature achieve good results in various applications, but many of them have never been used in remote sensing classification tasks. We have tested twelve descriptors that encode spectral/color properties and seven texture descriptors. We have also proposed a methodology based on the K-Nearest Neighbor (KNN) classifier for evaluation of descriptors in classification context. Experiments demonstrate that Joint Auto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition (SID), and Quantized Compound Change Histogram (QCCH) yield the best results in coffee and pasture recognition tasks. The second research issue refers to the problem of selecting the scale of segmentation for object-based remote sensing classification. Recently proposed methods exploit features extracted from segmented objects to improve high-resolution image classification. However, the definition of the scale of segmentation is a challenging task. We have proposed two multiscale classification approaches based on boosting of weak classifiers. The first approach, Multiscale Classifier (MSC), builds a strong classifier that combines features extracted from multiple scales of segmentation. The other, Hierarchical Multiscale Classifier (HMSC), exploits the hierarchical topology of segmented regions to improve training efficiency without accuracy loss when compared to the MSC. Experiments show that it is better to use multiple scales than use only one segmentation scale result. We have also analyzed and discussed about the correlation among the used descriptors and the scales of segmentation. The third research issue concerns the selection of training examples and the refinement of classification results through multiscale segmentation. We have proposed an approach for xix interactive multiscale classification of remote sensing images. It is an active learning strategy that allows the classification result refinement by the user along iterations. Experimental results show that the combination of scales produces better results than isolated scales in a relevance feedback process. Furthermore, the interactive method achieves good results with few user interactions. The proposed method needs only a small portion of the training set to build classifiers that are as strong as the ones generated by a supervised method that uses the whole available training set. The fourth research issue refers to the problem of extracting features of a hierarchy of regions for multiscale classification. We have proposed a strategy that exploits the existing relationships among regions in a hierarchy. This approach, called BoW-Propagation, exploits the bag-of-visual-word model to propagate features along multiple scales. We also extend this idea to propagate histogram-based global descriptors, the H-Propagation method. The proposed methods speed up the feature extraction process and yield good results when compared with global low-level extraction approaches
Doutorado
Ciência da Computação
Doutor em Ciência da Computação

APA, Harvard, Vancouver, ISO, and other styles

18

De, Hoedt Amanda Marie. "Clubfoot Image Classification." Thesis, University of Iowa, 2013. https://ir.uiowa.edu/etd/4836.

Full text

Abstract:

Clubfoot is a congenital foot disorder that, left untreated, can limit a person's mobility by making it difficult and painful to walk. Although inexpensive and reliable treatment exists, clubfoot often goes untreated in the developing world, where 80% of cases occur. Many nonprofit and non-governmental organizations are partnering with hospitals and clinics in the developing world to provide treatment for patients with clubfoot, and to train medical personnel in the use of these treatment methods. As a component of these partnerships, clinics and hospitals are collecting patient records. Some of this patient information, such as photographs, requires expert quality assessment. Such assessment may occur at a later date by a staff member in the hospital, or it may occur in a completely different location through the web interface. Photographs capture the state of a patient at a specific point in time. If a photograph is not taken correctly, and as a result, has no clinical utility, the photograph cannot be recreated because that moment in time has passed. These observations have motivated the desire to perform real-time classification of clubfoot images as they are being captured in a possibly remote and challenging environment. In the short term, successful classification could provide immediate feedback to those taking patient photos, helping to ensure that the image is of good quality and the foot is oriented correctly at the time of image capture. In the long term, this classification could be the basis for automated image analysis that could reduce the workload of a busy staff, and enable broader provision of treatment.

APA, Harvard, Vancouver, ISO, and other styles

19

Borges, Vinicius Ruela Pereira. "A computer-assisted approach to supporting taxonomical classification of freshwater green microalga images." Universidade de São Paulo, 2016. http://www.teses.usp.br/teses/disponiveis/55/55134/tde-07022017-163412/.

Full text

Abstract:

The taxonomical identification of freshwater green microalgae is highly relevant problem in Phycology. In particular, the taxonomical identification of samples from the Selenastraceae family of algae is considered particularly problematic with many known inconsistencies. Biologists manually inspect and analyze microscope images of alga strains, and typically carry out several complex and time-consuming procedures that demand considerable expert knowledge. Such practical limitations motivated this investigation on the applicability of image processing, pattern recognition and visual data mining techniques to support the biologists in tasks of species identification. This thesis describes methodologies for the classification of green alga images, considering both traditional automated classification processes and also a user-assisted incremental classification process supported by Neighbor Joining tree visualizations. In this process, users can interact with the visualizations to introduce their knowledge into the classification process, e.g. by selecting suitable training sets and evaluate the results, thus steering the classification process. In order for visualization and classification to be feasible, accurate features must be obtained from the images capable of distinguishing between the different species of algae. As morphological shape properties are a fundamental property in identifying species, suitable segmentation and shape feature extraction strategies have been developed. This was particularly challenging, as different alga species share common morphological characteristics. Two segmentation methodologies are introduced, in which one relies on the level set method and the other is based on the region growing principle. Although the contour-based approach is capable of handling the uneven conditions of green alga images, its computation is time-consuming and not suitable for real time applications. A specialized formulation of the region-based methodology is proposed that considers the specific characteristics of the green alga images handled. This second formulation was shown to be more efficient than the level set approach and generates highly accurate segmentations. Once accurate alga segmentation is achieved, two descriptors are proposed that capture alga shape properties, and also an effective general shape descriptor that computes quantitative measures from two signatures associated to the shape properties. Experimental results are described that indicate that the proposed solutions can be useful to biologists conducting alga identification tasks once it reduces their effort and attains satisfactory discrimination among species.
A identificação taxonômica de algas verdes de água doce é um problema de extrema relevância na Ficologia. Identificar espécies de algas da família Selenastraceae é uma tarefa complexa devido às inconsistências existentes em sua taxonomia, reconhecida como problemática. Os biólogos analisam manualmente imagens de microscópio de cepas de algas e realizam diversos procedimentos demorados que necessitamde conhecimento sólido. Tais limitaçõesmotivaramo estudo da aplicabilidade de técnicas de processamento de imagens, reconhecimento de padrões e mineração visual de dados para apoiar os biólogos em tarefas de identificação de espécies de algas. Esta tese descreve metodologias computacionais para a classificação de imagens de algas verdes, nas abordagens tradicional e baseada em classificação visual incremental com participação do usuário. Nesta última, os usuários interagem com visualizações baseadas em árvores filogenéticas para utilizar seu conhecimento no processo de classificação, como por exemplo, na seleção de instâncias relevantes para o conjunto de treinamento de um classificador, como também na avaliação dos resultados. De forma a viabilizar o uso de classificadores e técnicas de visualização, vetores de características devem ser obtidos das imagens de algas verdes. Neste trabalho, utiliza-se extração de características de forma, uma vez que a taxonomia da família Selenastraceae considera primordialmente as características morfológicas na identificação das espécies. No entanto, a obtenção de características representativas requer que as algas sejam precisamente segmentadas das imagens. Esta é, de fato, uma tarefa altamente desafiadora considerando a baixa qualidade das imagens e a maneira pelas quais as algas se organizam nas imagens. Duas metodologias de segmentação foram introduzidas: uma baseada no método Level Set e outra baseada no algoritmo de crescimento de regiões. A primeira se mostrou robusta e consegue identificar com alta precisão as algas nas imagens, mas seu tempo de execução é alto. A outra apresenta maior precisão e é mais rápida, uma vez que as técnicas de pré-processamento são especializadas para as imagens de algas verdes. Uma vez segmentadas as algas, dois descritores para caracterizar as imagens foram propostos: um baseado em características geométricas básicas e outro que utiliza medidas quantitativas calculadas a partir das assinaturas de forma. Resultados experimentais indicaram que as soluções propostas têm um bom potencial para serem utilizadas em tarefas de identificação taxonômica de algas verdes, uma vez que reduz o esforço nos procedimentos manuais e obtém-se classificações satisfatórias.

APA, Harvard, Vancouver, ISO, and other styles

20

Maggiori, Emmanuel. "Approches d'apprentissage pour la classification à large échelle d'images de télédétection." Thesis, Université Côte d'Azur (ComUE), 2017. http://www.theses.fr/2017AZUR4041/document.

Full text

Abstract:

L’analyse des images satellite et aériennes figure parmi les sujets fondamentaux du domaine de la télédétection. Ces dernières années, les avancées technologiques ont permis d’augmenter la disponibilité à large échelle des images, en comprenant parfois de larges étendues de terre à haute résolution spatiale. En plus des questions évidentes de complexité calculatoire qui en surgissent, un de plus importants défis est l’énorme variabilité des objets dans les différentes régions de la terre. Pour aborder cela, il est nécessaire de concevoir des méthodes de classification qui dépassent l’analyse du spectre individuel de chaque pixel, en introduisant de l’information contextuelle de haut niveau. Dans cette thèse, nous proposons d’abord une méthode pour la classification avec des contraintes de forme, basée sur l’optimisation d’une structure de subdivision hiérarchique des images. Nous explorons ensuite l’utilisation des réseaux de neurones convolutionnels (CNN), qui nous permettent d’apprendre des descripteurs hiérarchiques profonds. Nous étudions les CNN depuis de nombreux points de vue, ce qui nous permettra de les adapter à notre objectif. Parmi les sujets abordés, nous proposons différentes solutions pour générer des cartes de classification à haute résolution et nous étudions aussi la récolte des données d’entrainement. Nous avons également créé une base de données d’images aériennes sur des zones variées, pour évaluer la capacité de généralisation des CNN. Finalement, nous proposons une méthode pour polygonaliser les cartes de classification issues des réseaux de neurones, afin de pouvoir les intégrer dans des systèmes d’information géographique. Au long de la thèse, nous conduisons des expériences sur des images hyperspectrales, satellites et aériennes, toujours avec l’intention de proposer des méthodes applicables, généralisables et qui passent à l’échelle
The analysis of airborne and satellite images is one of the core subjects in remote sensing. In recent years, technological developments have facilitated the availability of large-scale sources of data, which cover significant extents of the earth’s surface, often at impressive spatial resolutions. In addition to the evident computational complexity issues that arise, one of the current challenges is to handle the variability in the appearance of the objects across different geographic regions. For this, it is necessary to design classification methods that go beyond the analysis of individual pixel spectra, introducing higher-level contextual information in the process. In this thesis, we first propose a method to perform classification with shape priors, based on the optimization of a hierarchical subdivision data structure. We then delve into the use of the increasingly popular convolutional neural networks (CNNs) to learn deep hierarchical contextual features. We investigate CNNs from multiple angles, in order to address the different points required to adapt them to our problem. Among other subjects, we propose different solutions to output high-resolution classification maps and we study the acquisition of training data. We also created a dataset of aerial images over dissimilar locations, and assess the generalization capabilities of CNNs. Finally, we propose a technique to polygonize the output classification maps, so as to integrate them into operational geographic information systems, thus completing the typical processing pipeline observed in a wide number of applications. Throughout this thesis, we experiment on hyperspectral, atellite and aerial images, with scalability, generalization and applicability goals in mind

APA, Harvard, Vancouver, ISO, and other styles

21

Hagedorn, Michael. "Classification of synthetic aperture radar images." Thesis, University of Canterbury. Electrical and Computer Engineering, 2004. http://hdl.handle.net/10092/5966.

Full text

Abstract:

In this thesis the maximum a posteriori (MAP) approach to synthetic aperture radar (SAR) analysis is reviewed. The MAP model consists of two probability density functions (PDFs): the likelihood function and the prior model. Contributions related to both models are made. As the first contribution a new likelihood function describing the multilook three-polarisation intensity SAR speckle process, which is equivalent to the averaged squared amplitude samples from a three-dimensional complex zero-mean circular Gaussian density, has been derived. This PDF is a correlated three-dimensional chi-square density in the form of an infinite series of modified Bessel functions with seven independent parameters. Details concerning the PDF such as the estimation of the PDF parameters from sample data and the moments of the PDF are described. The new likelihood function is tested against simulated and measured SAR data. The second contribution is a novel parameter estimation method for discrete Gibbs random field (GRF) prior models. Given a quantity of sample data, the parameters of the GRF model, which comprise the values of the potential functions of individual cliques, are estimated. The method uses an error function describing the difference between the local model PDF and the equivalent estimated from sample data. The concept of "equivalencies" is introduced to simplify the process. The new parameter estimation method is validated and compared to Besag's parameter estimation method (coding method) using GRF realisations and other sample data.

APA, Harvard, Vancouver, ISO, and other styles

22

Langdon, Matthew James. "Classification of images and censored data." Thesis, University of Leeds, 2005. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.434618.

Full text

APA, Harvard, Vancouver, ISO, and other styles

23

McGuire, Peter Frederick. "Image classification using eigenpaxels." Thesis, National Library of Canada = Bibliothèque nationale du Canada, 1999. http://www.collectionscanada.ca/obj/s4/f2/dsk1/tape8/PQDD_0002/NQ41239.pdf.

Full text

APA, Harvard, Vancouver, ISO, and other styles

24

Long, Yang. "Zero-shot image classification." Thesis, University of Sheffield, 2017. http://etheses.whiterose.ac.uk/18613/.

Full text

Abstract:

Image classification is one of the essential tasks for the intelligent visual system. Conventional image classification techniques rely on a large number of labelled images for supervised learning, which requires expensive human annotations. Towards real intelligent systems, a more favourable way is to teach the machine how to make classification using prior knowledge like humans. For example, a palaeontologist could recognise an extinct species purely based on the textual descriptions. To this end, Zero-Shot Image Classification (ZIC) is proposed, which aims to make machines that can learn to classify unseen images like humans. The problem can be viewed from two different levels. Low-level technical issues are concerned by the general Zero-shot Learning (ZSL) problem which considers how to train a classifier on the unseen visual domain using prior knowledge. High-level issues incorporate how to design and organise visual knowledge representation to construct a systematic ontology that could be an ultimate knowledge base for machines to learn. This thesis aims to provide a thorough study of the ZIC problem, regarding models, challenges, possible applications, etc. Besides, each main chapter demonstrates an innovative contribution that is creatively made during my study. The first is to solve the problem of Visual-Semantic Ambiguity. Namely, the same semantic concepts (e.g. attributes) can refer to a huge variety of visual features, and vice versa. Conventional ZSL methods usually adopt a one-way embedding that maps such high-variance visual features into the semantic space, which may lead to degraded performance. As a solution, a dual-graph regularised embedding algorithm named Visual-Semantic Ambiguity Removal (VSAR) is proposed, which can capture the intrinsic local structure of both visual and semantic spaces. In the intermediate embedding space, the structural difference is reconciled to remove the ambiguity. The second contribution aims to circumvent costly visual data collection for conventional supervised classification using ZSL techniques. The key idea is to synthesise visual features from the semantic information, just like humans can imagine features of an unseen class from the semantic description of prior knowledge. Hereafter, new objects from unseen classes can be classified in a conventional supervised framework using the inferred visual features. To overcome the correlation problem, we propose an intermediate Orthogonal Semantic-Visual Embedding (OSVE) space to remove the correlated redundancy. The proposed method achieves promising performance on fine-grained datasets. In the third contribution, the graph constraint of VSAR is incorporated to synthesise improved visual features. The orthogonal embedding is reconsidered as an Information Diffusion problem. Through an orthogonal rotation, the synthesised visual features become more discriminative. On four benchmarks, the proposed method demonstrates the advantages of synthesised visual features, which significantly outperforms state-of-the-art results. Since most of ZSL approaches highly rely on expensive attributes, the fourth contribution of this thesis explores a more feasible but more effective Semantic Simile model to describe unseen classes. From a group of similes, e.g. an unknown animal has the same parts of a wolf, and the colour looks like a bobcat, implicit attributes are discovered by a graph-cut algorithm. Comprehensive experimental results suggest the simile-based implicit attributes can significantly boost the performance. To maximumly reduce the cost of building ontologies for ZIC, the final chapter introduces a novel scheme, using which ZIC can be achieved by only a few similes of each unseen class. No annotations of seen classes are needed. Such an approach finally sets ZIC attribute-free, which significantly improve the feasibility of ZIC. Unseen classes can be recognised using a conventional setting without expensive attribute ontology. It can be concluded that the methods introduced in this thesis provide fundamental components of a zero-shot image classification system. The thesis also points out four core directions for future ZIC research.

APA, Harvard, Vancouver, ISO, and other styles

25

Wakade, Shruti Vijay. "Classification of Image Spam." University of Akron / OhioLINK, 2011. http://rave.ohiolink.edu/etdc/view?acc_num=akron1311113808.

Full text

APA, Harvard, Vancouver, ISO, and other styles

26

Bittencourt, Helio Radke. "Detecção de mudanças a partir de imagens de fração." reponame:Biblioteca Digital de Teses e Dissertações da UFRGS, 2011. http://hdl.handle.net/10183/36053.

Full text

Abstract:

A detecção de mudanças na superfície terrestre é o principal objetivo em aplicações de sensoriamento remoto multitemporal. Sabe-se que imagens adquiridas em datas distintas tendem a ser altamente influenciadas por problemas radiométricos e de registro. Utilizando imagens de fração, obtidas a partir do modelo linear de mistura espectral (MLME), problemas radiométricos podem ser minimizados e a interpretação dos tipos de mudança na superfície terrestre é facilitada, pois as frações têm um significado físico direto. Além disso, interpretações ao nível de subpixel são possíveis. Esta tese propõe três algoritmos – rígido, suave e fuzzy – para a detecção de mudanças entre um par de imagens de fração, gerando mapas de mudança como produtos finais. As propostas requerem a suposição de normalidade multivariada para as diferenças de fração e necessitam de pouca intervenção por parte do analista. A proposta rígida cria mapas de mudança binários seguindo a mesma metodologia de um teste de hipóteses, baseando-se no fato de que os contornos de densidade constante na distribuição normal multivariada são definidos por valores da distribuição qui-quadrado, de acordo com a escolha do nível de confiança. O classificador suave permite gerar estimativas da probabilidade do pixel pertencer à classe de mudança, a partir de um modelo de regressão logística. Essas probabilidades são usadas para criar um mapa de probabilidades de mudança. A abordagem fuzzy é aquela que melhor se adapta ao conceito de pixel mistura, visto que as mudanças no uso e cobertura do solo podem ocorrer em nível de subpixel. Com base nisso, mapas dos graus de pertinência à classe de mudança foram criados. Outras ferramentas matemáticas e estatísticas foram utilizadas, tais como operações morfológicas, curvas ROC e algoritmos de clustering. As três propostas foram testadas utilizando-se imagens sintéticas e reais (Landsat-TM) e avaliadas qualitativa e quantitativamente. Os resultados indicam a viabilidade da utilização de imagens de fração em estudos de detecção de mudanças por meio dos algoritmos propostos.
Land cover change detection is a major goal in multitemporal remote sensing applications. It is well known that images acquired on different dates tend to be highly influenced by radiometric differences and registration problems. Using fraction images, obtained from the linear model of spectral mixing (LMSM), radiometric problems can be minimized and the interpretation of changes in land cover is facilitated because the fractions have a physical meaning. Furthermore, interpretations at the subpixel level are possible. This thesis presents three algorithms – hard, soft and fuzzy – for detecting changes between a pair of fraction images. The algorithms require multivariate normality for the differences among fractions and very little intervention by the analyst. The hard algorithm creates binary change maps following the same methodology of hypothesis testing, based on the fact that the contours of constant density are defined by chi-square values, according to the choice of the probability level. The soft one allows for the generation of estimates of the probability of each pixel belonging to the change class by using a logistic regression model. These probabilities are used to create a map of change probabilities. The fuzzy approach is the one that best fits the concept behind the fraction images because the changes in land cover can occurr at a subpixel level. Based on these algorithms, maps of membership degrees were created. Other mathematical and statistical techniques were also used, such as morphological operations, ROC curves and a clustering algorithm. The algorithms were tested using synthetic and real images (Landsat-TM) and the results were analyzed qualitatively and quantitatively. The results indicate that fraction images can be used in change detection studies by using the proposed algorithms.

APA, Harvard, Vancouver, ISO, and other styles

27

Shin, Jiwon. "Parts-based object classification for range images." Zürich : Swiss Federal Institute of Technology, Autonomous Systems Lab, 2008. http://e-collection.ethbib.ethz.ch/show?type=dipl&nr=384.

Full text

APA, Harvard, Vancouver, ISO, and other styles

28

Dos, santos Jefersson Alex. "Semi-automatic Classification of Remote Sensing Images." Phd thesis, Université de Cergy Pontoise, 2013. http://tel.archives-ouvertes.fr/tel-00878612.

Full text

Abstract:

A huge effort has been made in the development of image classification systemswith the objective of creating high-quality thematic maps and to establishprecise inventories about land cover use. The peculiarities of Remote SensingImages (RSIs) combined with the traditional image classification challengesmake RSI classification a hard task. Many of the problems are related to therepresentation scale of the data, and to both the size and therepresentativeness of used training set.In this work, we addressed four research issues in order to develop effectivesolutions for interactive classification of remote sensing images.The first research issue concerns the fact that image descriptorsproposed in the literature achieve good results in various applications, butmany of them have never been used in remote sensing classification tasks.We have tested twelve descriptors that encodespectral/color properties and seven texture descriptors. We have also proposeda methodology based on the K-Nearest Neighbor (KNN) classifier for evaluationof descriptors in classification context. Experiments demonstrate that JointAuto-Correlogram (JAC), Color Bitmap, Invariant Steerable Pyramid Decomposition(SID), and Quantized Compound Change Histogram (QCCH) yield the best results incoffee and pasture recognition tasks.The second research issue refers to the problem of selecting the scaleof segmentation for object-based remote sensing classification. Recentlyproposed methods exploit features extracted from segmented objects to improvehigh-resolution image classification. However, the definition of the scale ofsegmentation is a challenging task. We have proposedtwo multiscale classification approaches based on boosting of weak classifiers.The first approach, Multiscale Classifier (MSC), builds a strongclassifier that combines features extracted from multiple scales ofsegmentation. The other, Hierarchical Multiscale Classifier (HMSC), exploits thehierarchical topology of segmented regions to improve training efficiencywithout accuracy loss when compared to the MSC. Experiments show that it isbetter to use multiple scales than use only one segmentation scale result. Wehave also analyzed and discussed about the correlation among the useddescriptors and the scales of segmentation.The third research issue concerns the selection of training examples and therefinement of classification results through multiscale segmentation. We have proposed an approach forinteractive multiscale classification of remote sensing images.It is an active learning strategy that allows the classification resultrefinement by the user along iterations. Experimentalresults show that the combination of scales produces better results thanisolated scales in a relevance feedback process. Furthermore, the interactivemethod achieves good results with few user interactions. The proposed methodneeds only a small portion of the training set to build classifiers that are asstrong as the ones generated by a supervised method that uses the whole availabletraining set.The fourth research issue refers to the problem of extracting features of ahierarchy of regions for multiscale classification. We have proposed a strategythat exploits the existing relationships among regions in a hierarchy. Thisapproach, called BoW-Propagation, exploits the bag-of-visual-word model topropagate features along multiple scales. We also extend this idea topropagate histogram-based global descriptors, the H-Propagation method. The proposedmethods speed up the feature extraction process and yield good results when compared with globallow-level extraction approaches.

APA, Harvard, Vancouver, ISO, and other styles

29

Pop, David. "Classification of Heart Views in Ultrasound Images." Thesis, Linköpings universitet, Datorseende, 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-165276.

Full text

Abstract:

In today’s society, we experience an increasing challenge to provide healthcare to everyone in need due to the increasing number of patients and the shortage of medical staff. Computers have contributed to mitigating this challenge by offloading the medical staff from some of the tasks. With the rise of deep learning, countless new possibilities have opened to help the medical staff even further. One domain where deep learning can be applied is analysis of ultrasound images. In this thesis we investigate the problem of classifying standard views of the heart in ultrasound images with the help of deep learning. We conduct mainly three experiments. First, we use NasNet mobile, InceptionV3, VGG16 and MobileNet, pre-trained on ImageNet, and finetune them to ultrasound heart images. We compare the accuracy of these networks to each other and to the baselinemodel, a CNN that was proposed in [23]. Then we assess a neural network’s capability to generalize to images from ultrasound machines that the network is not trained on. Lastly, we test how the performance of the networks degrades with decreasing amount of training data. Our first experiment shows that all networks considered in this study have very similar performance in terms of accuracy with Inception V3 being slightly better than the rest. The best performance is achieved when the whole network is finetuned to our problem instead of finetuning only apart of it, while gradually unlocking more layers for training. The generalization experiment shows that neural networks have the potential to generalize to images from ultrasound machines that they are not trained on. It also shows that having a mix of multiple ultrasound machines in the training data increases generalization performance. In our last experiment we compare the performance of the CNN proposed in [23] with MobileNet pre-trained on ImageNet and MobileNet randomly initialized. This shows that the performance of the baseline model suffers the least with decreasing amount of training data and that pre-training helps the performance drastically on smaller training datasets.

APA, Harvard, Vancouver, ISO, and other styles

30

Augereau, Olivier. "Reconnaissance et classification d’images de documents." Thesis, Bordeaux 1, 2013. http://www.theses.fr/2013BOR14764/document.

Full text

Abstract:

Ces travaux de recherche ont pour ambition de contribuer à la problématique de la classification d’images de documents. Plus précisément, ces travaux tendent à répondre aux problèmes rencontrés par des sociétés de numérisation dont l’objectif est de mettre à disposition de leurs clients une version numérique des documents papiers accompagnés d’informations qui leurs sont relatives. Face à la diversité des documents à numériser, l’extraction d’informations peut s’avérer parfois complexe. C’est pourquoi la classification et l’indexation des documents sont très souvent réalisées manuellement. Ces travaux de recherche ont permis de fournir différentes solutions en fonction des connaissances relatives aux images que possède l’utilisateur ayant en charge l’annotation des documents.Le premier apport de cette thèse est la mise en place d’une méthode permettant, de manière interactive, à un utilisateur de classer des images de documents dont la nature est inconnue. Le second apport de ces travaux est la proposition d’une technique de recherche d’images de documents par l’exemple basée sur l’extraction et la mise en correspondance de points d’intérêts. Le dernier apport de cette thèse est l’élaboration d’une méthode de classification d’images de documents utilisant les techniques de sacs de mots visuels
The aim of this research is to contribute to the document image classification problem. More specifically, these studies address digitizing company issues which objective is to provide the digital version of paper document with information relating to them. Given the diversity of documents, information extraction can be complex. This is why the classification and the indexing of documents are often performed manually. This research provides several solutions based on knowledge of the images that the user has. The first contribution of this thesis is a method for classifying interactively document images, where the content of documents and classes are unknown. The second contribution of this work is a new technique for document image retrieval by giving one example of researched document. This technique is based on the extraction and matching of interest points. The last contribution of this thesis is a method for classifying document images by using bags of visual words techniques

APA, Harvard, Vancouver, ISO, and other styles

31

Lu, Ying. "Transfer Learning for Image Classification." Thesis, Lyon, 2017. http://www.theses.fr/2017LYSEC045/document.

Full text

Abstract:

Lors de l’apprentissage d’un modèle de classification pour un nouveau domaine cible avec seulement une petite quantité d’échantillons de formation, l’application des algorithmes d’apprentissage automatiques conduit généralement à des classifieurs surdimensionnés avec de mauvaises compétences de généralisation. D’autre part, recueillir un nombre suffisant d’échantillons de formation étiquetés manuellement peut s’avérer très coûteux. Les méthodes de transfert d’apprentissage visent à résoudre ce type de problèmes en transférant des connaissances provenant d’un domaine source associé qui contient beaucoup plus de données pour faciliter la classification dans le domaine cible. Selon les différentes hypothèses sur le domaine cible et le domaine source, l’apprentissage par transfert peut être classé en trois catégories: apprentissage par transfert inductif, apprentissage par transfert transducteur (adaptation du domaine) et apprentissage par transfert non surveillé. Nous nous concentrons sur le premier qui suppose que la tâche cible et la tâche source sont différentes mais liées. Plus précisément, nous supposons que la tâche cible et la tâche source sont des tâches de classification, tandis que les catégories cible et les catégories source sont différentes mais liées. Nous proposons deux méthodes différentes pour aborder ce problème. Dans le premier travail, nous proposons une nouvelle méthode d’apprentissage par transfert discriminatif, à savoir DTL(Discriminative Transfer Learning), combinant une série d’hypothèses faites à la fois par le modèle appris avec les échantillons de cible et les modèles supplémentaires appris avec des échantillons des catégories sources. Plus précisément, nous utilisons le résidu de reconstruction creuse comme discriminant de base et améliore son pouvoir discriminatif en comparant deux résidus d’un dictionnaire positif et d’un dictionnaire négatif. Sur cette base, nous utilisons des similitudes et des dissemblances en choisissant des catégories sources positivement corrélées et négativement corrélées pour former des dictionnaires supplémentaires. Une nouvelle fonction de coût basée sur la statistique de Wilcoxon-Mann-Whitney est proposée pour choisir les dictionnaires supplémentaires avec des données non équilibrées. En outre, deux processus de Boosting parallèles sont appliqués à la fois aux distributions de données positives et négatives pour améliorer encore les performances du classificateur. Sur deux bases de données de classification d’images différentes, la DTL proposée surpasse de manière constante les autres méthodes de l’état de l’art du transfert de connaissances, tout en maintenant un temps d’exécution très efficace. Dans le deuxième travail, nous combinons le pouvoir du transport optimal (OT) et des réseaux de neurones profond (DNN) pour résoudre le problème ITL. Plus précisément, nous proposons une nouvelle méthode pour affiner conjointement un réseau de neurones avec des données source et des données cibles. En ajoutant une fonction de perte du transfert optimal (OT loss) entre les prédictions du classificateur source et cible comme une contrainte sur le classificateur source, le réseau JTLN (Joint Transfer Learning Network) proposé peut effectivement apprendre des connaissances utiles pour la classification cible à partir des données source. En outre, en utilisant différents métriques comme matrice de coût pour la fonction de perte du transfert optimal, JTLN peut intégrer différentes connaissances antérieures sur la relation entre les catégories cibles et les catégories sources. Nous avons effectué des expérimentations avec JTLN basées sur Alexnet sur les jeux de données de classification d’image et les résultats vérifient l’efficacité du JTLN proposé. A notre connaissances, ce JTLN proposé est le premier travail à aborder ITL avec des réseaux de neurones profond (DNN) tout en intégrant des connaissances antérieures sur la relation entre les catégories cible et source
When learning a classification model for a new target domain with only a small amount of training samples, brute force application of machine learning algorithms generally leads to over-fitted classifiers with poor generalization skills. On the other hand, collecting a sufficient number of manually labeled training samples may prove very expensive. Transfer Learning methods aim to solve this kind of problems by transferring knowledge from related source domain which has much more data to help classification in the target domain. Depending on different assumptions about target domain and source domain, transfer learning can be further categorized into three categories: Inductive Transfer Learning, Transductive Transfer Learning (Domain Adaptation) and Unsupervised Transfer Learning. We focus on the first one which assumes that the target task and source task are different but related. More specifically, we assume that both target task and source task are classification tasks, while the target categories and source categories are different but related. We propose two different methods to approach this ITL problem. In the first work we propose a new discriminative transfer learning method, namely DTL, combining a series of hypotheses made by both the model learned with target training samples, and the additional models learned with source category samples. Specifically, we use the sparse reconstruction residual as a basic discriminant, and enhance its discriminative power by comparing two residuals from a positive and a negative dictionary. On this basis, we make use of similarities and dissimilarities by choosing both positively correlated and negatively correlated source categories to form additional dictionaries. A new Wilcoxon-Mann-Whitney statistic based cost function is proposed to choose the additional dictionaries with unbalanced training data. Also, two parallel boosting processes are applied to both the positive and negative data distributions to further improve classifier performance. On two different image classification databases, the proposed DTL consistently out performs other state-of-the-art transfer learning methods, while at the same time maintaining very efficient runtime. In the second work we combine the power of Optimal Transport and Deep Neural Networks to tackle the ITL problem. Specifically, we propose a novel method to jointly fine-tune a Deep Neural Network with source data and target data. By adding an Optimal Transport loss (OT loss) between source and target classifier predictions as a constraint on the source classifier, the proposed Joint Transfer Learning Network (JTLN) can effectively learn useful knowledge for target classification from source data. Furthermore, by using different kind of metric as cost matrix for the OT loss, JTLN can incorporate different prior knowledge about the relatedness between target categories and source categories. We carried out experiments with JTLN based on Alexnet on image classification datasets and the results verify the effectiveness of the proposed JTLN in comparison with standard consecutive fine-tuning. To the best of our knowledge, the proposed JTLN is the first work to tackle ITL with Deep Neural Networks while incorporating prior knowledge on relatedness between target and source categories. This Joint Transfer Learning with OT loss is general and can also be applied to other kind of Neural Networks

APA, Harvard, Vancouver, ISO, and other styles

32

Dutt, Anuvabh. "Continual learning for image classification." Thesis, Université Grenoble Alpes (ComUE), 2019. http://www.theses.fr/2019GREAM063.

Full text

Abstract:

Cette thèse traite de l'apprentissage en profondeur appliqu'e aux tâches de classification des images. La principale motivation du travail est de rendre les techniques d’apprentissage en profondeur actuelles plus efficaces et de faire face aux changements dans la distribution des données. Nous travaillons dans le cadre élargi de l’apprentissage continu, dans le but d’avoir 'a l’avenir des modèles d’apprentissage automatique pouvant être améliorés en permanence.Nous examinons d’abord la modification de l’espace étiquette d’un ensemble de données, les échantillons de données restant les mêmes. Nous considérons une hiérarchie d'étiquettes sémantiques à laquelle appartiennent les étiquettes. Nous étudions comment nous pouvons utiliser cette hiérarchie pour obtenir des améliorations dans les modèles formés à différents niveaux de cette hiérarchie.Les deuxième et troisième contributions impliquent un apprentissage continu utilisant un modèle génératif. Nous analysons la facilité d'utilisation des échantillons d'un modèle génératif dans le cas de la formation de bons classificateurs discriminants. Nous proposons des techniques pour améliorer la sélection et la génération d'échantillons à partir d'un modèle génératif. Ensuite, nous observons que les algorithmes d’apprentissage continu subissent certaines pertes de performances lorsqu’ils sont entraînés séquentiellement à plusieurs tâches. Nous analysons la dynamique de la formation dans ce scénario et comparons avec la formation sur plusieurs tâches simultanément. Nous faisons des observations qui indiquent des difficultés potentielles dans l’apprentissage de modèles dans un scénario d’apprentissage continu.Enfin, nous proposons un nouveau modèle de conception pour les réseaux de convolution. Cette architecture permet de former des modèles plus petits sans compromettre les performances. De plus, la conception se prête facilement à la parallélisation, ce qui permet une formation distribuée efficace.En conclusion, nous examinons deux types de scénarios d’apprentissage continu. Nous proposons des méthodes qui conduisent à des améliorations. Notre analyse met 'egalement en évidence des problèmes plus importants, dont nous aurions peut-être besoin de changements dans notre procédure actuelle de formation de réseau neuronal
This thesis deals with deep learning applied to image classification tasks. The primary motivation for the work is to make current deep learning techniques more efficient and to deal with changes in the data distribution. We work in the broad framework of continual learning, with the aim to have in the future machine learning models that can continuously improve.We first look at change in label space of a data set, with the data samples themselves remaining the same. We consider a semantic label hierarchy to which the labels belong. We investigate how we can utilise this hierarchy for obtaining improvements in models which were trained on different levels of this hierarchy.The second and third contribution involve continual learning using a generative model. We analyse the usability of samples from a generative model in the case of training good discriminative classifiers. We propose techniques to improve the selection and generation of samples from a generative model. Following this, we observe that continual learning algorithms do undergo some loss in performance when trained on several tasks sequentially. We analyse the training dynamics in this scenario and compare with training on several tasks simultaneously. We make observations that point to potential difficulties in the learning of models in a continual learning scenario.Finally, we propose a new design template for convolutional networks. This architecture leads to training of smaller models without compromising performance. In addition the design lends itself to easy parallelisation, leading to efficient distributed training.In conclusion, we look at two different types of continual learning scenarios. We propose methods that lead to improvements. Our analysis also points to greater issues, to over come which we might need changes in our current neural network training procedure

APA, Harvard, Vancouver, ISO, and other styles

33

Mohamed, Aamer S. S. "From content-based to semantic image retrieval. Low level feature extraction, classification using image processing and neural networks, content based image retrieval, hybrid low level and high level based image retrieval in the compressed DCT domain." Thesis, University of Bradford, 2010. http://hdl.handle.net/10454/4438.

Full text

Abstract:

Digital image archiving urgently requires advanced techniques for more efficient storage and retrieval methods because of the increasing amount of digital. Although JPEG supply systems to compress image data efficiently, the problems of how to organize the image database structure for efficient indexing and retrieval, how to index and retrieve image data from DCT compressed domain and how to interpret image data semantically are major obstacles for further development of digital image database system. In content-based image, image analysis is the primary step to extract useful information from image databases. The difficulty in content-based image retrieval is how to summarize the low-level features into high-level or semantic descriptors to facilitate the retrieval procedure. Such a shift toward a semantic visual data learning or detection of semantic objects generates an urgent need to link the low level features with semantic understanding of the observed visual information. To solve such a -semantic gap¿ problem, an efficient way is to develop a number of classifiers to identify the presence of semantic image components that can be connected to semantic descriptors. Among various semantic objects, the human face is a very important example, which is usually also the most significant element in many images and photos. The presence of faces can usually be correlated to specific scenes with semantic inference according to a given ontology. Therefore, face detection can be an efficient tool to annotate images for semantic descriptors. In this thesis, a paradigm to process, analyze and interpret digital images is proposed. In order to speed up access to desired images, after accessing image data, image features are presented for analysis. This analysis gives not only a structure for content-based image retrieval but also the basic units ii for high-level semantic image interpretation. Finally, images are interpreted and classified into some semantic categories by semantic object detection categorization algorithm.

APA, Harvard, Vancouver, ISO, and other styles

34

Mohamed, Aamer Saleh Sahel. "From content-based to semantic image retrieval : low level feature extraction, classification using image processing and neural networks, content based image retrieval, hybrid low level and high level based image retrieval in the compressed DCT domain." Thesis, University of Bradford, 2010. http://hdl.handle.net/10454/4438.

Full text

Abstract:

Digital image archiving urgently requires advanced techniques for more efficient storage and retrieval methods because of the increasing amount of digital. Although JPEG supply systems to compress image data efficiently, the problems of how to organize the image database structure for efficient indexing and retrieval, how to index and retrieve image data from DCT compressed domain and how to interpret image data semantically are major obstacles for further development of digital image database system. In content-based image, image analysis is the primary step to extract useful information from image databases. The difficulty in content-based image retrieval is how to summarize the low-level features into high-level or semantic descriptors to facilitate the retrieval procedure. Such a shift toward a semantic visual data learning or detection of semantic objects generates an urgent need to link the low level features with semantic understanding of the observed visual information. To solve such a 'semantic gap' problem, an efficient way is to develop a number of classifiers to identify the presence of semantic image components that can be connected to semantic descriptors. Among various semantic objects, the human face is a very important example, which is usually also the most significant element in many images and photos. The presence of faces can usually be correlated to specific scenes with semantic inference according to a given ontology. Therefore, face detection can be an efficient tool to annotate images for semantic descriptors. In this thesis, a paradigm to process, analyze and interpret digital images is proposed. In order to speed up access to desired images, after accessing image data, image features are presented for analysis. This analysis gives not only a structure for content-based image retrieval but also the basic units ii for high-level semantic image interpretation. Finally, images are interpreted and classified into some semantic categories by semantic object detection categorization algorithm.

APA, Harvard, Vancouver, ISO, and other styles

35

Paulin, Mattis. "De l'apprentissage de représentations visuelles robustes aux invariances pour la classification et la recherche d'images." Thesis, Université Grenoble Alpes (ComUE), 2017. http://www.theses.fr/2017GREAM007/document.

Full text

Abstract:

Ce mémoire de thèse porte sur l’élaboration de systèmes de reconnaissance d’image qui sont robustes à la variabilité géométrique. La compréhension d’une image est un problème difficile, de par le fait qu’elles sont des projections en deux dimensions d’objets 3D. Par ailleurs, des représentations qui doivent appartenir à la même catégorie, par exemple des objets de la même classe en classification, peuvent être visuellement très différentes. Notre but est de rendre ces systèmes robustes à la juste quantité de déformations, celle-ci étant automatiquement déterminée à partir des données. Nos deux contributions sont les suivantes. Nous montrons tout d’abord comment utiliser des exemples virtuels pour rendre les systèmes de classification d’images robustes et nous proposons ensuite une méthodologie pour apprendre des descripteurs de bas niveau robustes, pour la recherche d’image.Nous étudions tout d’abord les exemples virtuels, en tant que transformations de vrais exemples. En représentant une image en tant que sac de descripteurs transformés, nous montrons que l’augmentation de données, c’est-à-dire le fait de les considérer comme de nouveaux exemples iid, est la meilleure manière de les utiliser, pourvu qu’une étape de vote avec les descripteurs transformés soit opérée lors du test. Du fait que les transformations apportent différents niveaux d’information, peuvent être redondants, voire nuire à la performance, nous pro-posons un nouvel algorithme capable de sélectionner un petit nombre d’entre elles,en maximisant la justesse de classification. Nous montrons par ailleurs comment remplacer de vrais exemples par des virtuels, pour alléger les couts d’annotation.Nous rapportons de bons résultats sur des bancs d’essai de classification.Notre seconde contribution vise à améliorer les descripteurs de régions locales utilisés en recherche d’image, et en particulier nous proposons une alternative au populaire descripteur SIFT. Nous proposons un nouveau descripteur, appelé patch-CKN, appris sans supervision. Nous introduisons un nouvel ensemble de données liant les images et les imagettes, construit à partir de reconstruction3D automatique d’images récupérées sur Internet. Nous définissons une méthode pour tester précisément la performance des descripteurs locaux au niveau de l’imagette et de l’image. Notre approche dépasse SIFT et les autres approches à base d’architectures convolutionnelles sur notre banc d’essai, et d’autres couramment utilisés dans la littérature
This dissertation focuses on designing image recognition systems which are robust to geometric variability. Image understanding is a difficult problem, as images are two-dimensional projections of 3D objects, and representations that must fall into the same category, for instance objects of the same class in classification can display significant differences. Our goal is to make systems robust to the right amount of deformations, this amount being automatically determined from data. Our contributions are twofolds. We show how to use virtual examples to enforce robustness in image classification systems and we propose a framework to learn robust low-level descriptors for image retrieval. We first focus on virtual examples, as transformation of real ones. One image generates a set of descriptors –one for each transformation– and we show that data augmentation, ie considering them all as iid samples, is the best performing method to use them, provided a voting stage with the transformed descriptors is conducted at test time. Because transformations have various levels of information, can be redundant, and can even be harmful to performance, we propose a new algorithm able to select a set of transformations, while maximizing classification accuracy. We show that a small amount of transformations is enough to considerably improve performance for this task. We also show how virtual examples can replace real ones for a reduced annotation cost. We report good performance on standard fine-grained classification datasets. In a second part, we aim at improving the local region descriptors used in image retrieval and in particular to propose an alternative to the popular SIFT descriptor. We propose new convolutional descriptors, called patch-CKN, which are learned without supervision. We introduce a linked patch- and image-retrieval dataset based on structure from motion of web-crawled images, and design a method to accurately test the performance of local descriptors at patch and image levels. Our approach outperforms both SIFT and all tested approaches with convolutional architectures on our patch and image benchmarks, as well as several styate-of-theart datasets

APA, Harvard, Vancouver, ISO, and other styles

36

Laouamer, Lamri. "Approche exploratoire sur la classification appliquée aux images /." Trois-Rivières : Université du Québec à Trois-Rivières, 2006. http://www.uqtr.ca/biblio/notice/resume/24710337R.pdf.

Full text

APA, Harvard, Vancouver, ISO, and other styles

37

Tress, Andrew. "Practical classification and segmentation of large textural images." Thesis, Heriot-Watt University, 1996. http://hdl.handle.net/10399/720.

Full text

APA, Harvard, Vancouver, ISO, and other styles

38

Nwoye, Ephraim O. "Fuzzy neural classification of colon cancer cell images." Thesis, University of Newcastle Upon Tyne, 2006. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.432763.

Full text

APA, Harvard, Vancouver, ISO, and other styles

39

Walder, Patrick. "Automated classification of cloud types from satellite images." Thesis, University of the West of Scotland, 1998. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.496612.

Full text

APA, Harvard, Vancouver, ISO, and other styles

40

Trakas, Joannis. "Classification of medical images with small data sets." Thesis, University of Sussex, 2009. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.508988.

Full text

APA, Harvard, Vancouver, ISO, and other styles

41

Bin, Ghaith Alsuwaidi Ali Rashed Saeed A. "Feature analysis of hyperspectral images for plant classification." Thesis, University of Manchester, 2018. https://www.research.manchester.ac.uk/portal/en/theses/feature-analysis-of-hyperspectral-images-for-plant-classification(38a3f58f-d057-4a04-8899-81768c055652).html.

Full text

Abstract:

The role of hyperspectral imaging in precision agriculture has increased recently as a result of numerous financial and environmental benefits. Differentiating plant types and conditions is vital to precision agriculture as it helps detect diseases and stresses and optimise growth control. Techniques using hyperspectral images have been developed to study plant types and conditions. Although such techniques are becoming a common trend in precision agriculture, the task remains challenging, because of high image dimensionality, data volume, and sensitivity of the analysis techniques to seasonal or condition changes. The research described in this thesis is concerned with the analysis of hyperspectral data for plants of different conditions, or crop classification by means of advanced machine learning techniques. The main contribution of this research lies in three new approaches proposed to improve local and global classification performance of plant and crop classification in a controlled environment (i.e. laboratory), in which adaptive feature selection, ensemble learning, novelty detection, feature level, and decision level fusions are used. The first approach integrates feature selection and ensemble learning, while the second approach incorporates feature selection, ensemble learning, and novelty detection and the third approach merges different levels of image properties at both feature extraction and classifying decision stages. In the first approach, a feature-ensemble framework is proposed to enhance robustness and performance of the classification. The input data are divided into a number of pools using jackknife data split. In total, six feature selection algorithms are used in each pool simultaneously to select the best performing algorithm. The decisions of the retained algorithms are then used to form the final decision of the framework using a hard fusion scheme. This approach is particularly suited when all class labels are available. The second approach extends the framework into unbalanced data cases (e.g. missing labels, or samples, or both). A domain-based novelty detection (probabilistic one-class support vector machine) is used with regard to defining the domain of the available class samples (one class only) and how this domain deviated from unseen testing samples. In the third approach, advanced machine learning techniques are employed to identify distinctive features in both spectral and spatial domains of hyperspectral images. Texture properties are explored as spatial features in the sub-band images. These two levels of properties are then integrated into a proposed spectral-texture framework for studying plant conditions. Feature level, decision level, and feature-and-decision level fusion schemes are used in the third approach framework to improve local and overall classification performance. Performances of these three approaches were evaluated using various hyperspectral datasets in addition to benchmark databases. Results of these three approaches show significant improvements in classification performance as compared to empirical spectral indices and conventional analysis methods. The overall improvement rates, using several hyperspectral datasets under different conditions, are greater than 1%, 2.5%, and 2% for the first, second, and third approaches, respectively. The findings also indicate the usefulness of the proposed approaches for precision agriculture analysis. In addition, the experiments across various hyperspectral datasets show the validity and applicability of the proposed approaches to a wide range of condition monitoring applications.

APA, Harvard, Vancouver, ISO, and other styles

42

Laouamer, Lamri. "Approche exploratoire sur la classification appliquée aux images." Thèse, Université du Québec à Trois-Rivières, 2006. http://depot-e.uqtr.ca/1208/1/000133228.pdf.

Full text

APA, Harvard, Vancouver, ISO, and other styles

43

Kalvakolanu, Anjaneya Teja Sarma. "Brain Tumor Detection and Classification from MRI Images." DigitalCommons@CalPoly, 2021. https://digitalcommons.calpoly.edu/theses/2267.

Full text

Abstract:

A brain tumor is detected and classified by biopsy that is conducted after the brain surgery. Advancement in technology and machine learning techniques could help radiologists in the diagnosis of tumors without any invasive measures. We utilized a deep learning-based approach to detect and classify the tumor into Meningioma, Glioma, Pituitary tumors. We used registration and segmentation-based skull stripping mechanism to remove the skull from the MRI images and the grab cut method to verify whether the skull stripped MRI masks retained the features of the tumor for accurate classification. In this research, we proposed a transfer learning based approach in conjunction with discriminative learning rates to perform the classification of brain tumors. The data set used is a 3064 T MRI images dataset that contains T1 flair MRI images. We achieved a classification accuracy of 98.83%, 96.26%, and 95.18% for training, validation, and test sets and an F1 score of 0.96 on the T1 Flair MRI dataset.

APA, Harvard, Vancouver, ISO, and other styles

44

Ouji, Asma. "Segmentation et classification dans les images de documents numérisés." Phd thesis, INSA de Lyon, 2012. http://tel.archives-ouvertes.fr/tel-00749933.

Full text

Abstract:

Les travaux de cette thèse ont été effectués dans le cadre de l'analyse et du traitement d'images de documents imprimés afin d'automatiser la création de revues de presse. Les images en sortie du scanner sont traitées sans aucune information a priori ou intervention humaine. Ainsi, pour les caractériser, nous présentons un système d'analyse de documents composites couleur qui réalise une segmentation en zones colorimétriquement homogènes et qui adapte les algorithmes d'extraction de textes aux caractéristiques locales de chaque zone. Les informations colorimétriques et textuelles fournies par ce système alimentent une méthode de segmentation physique des pages de presse numérisée. Les blocs issus de cette décomposition font l'objet d'une classification permettant, entre autres, de détecter les zones publicitaires. Dans la continuité et l'expansion des travaux de classification effectués dans la première partie, nous présentons un nouveau moteur de classification et de classement générique, rapide et facile à utiliser. Cette approche se distingue de la grande majorité des méthodes existantes qui reposent sur des connaissances a priori sur les données et dépendent de paramètres abstraits et difficiles à déterminer par l'utilisateur. De la caractérisation colorimétrique au suivi des articles en passant par la détection des publicités, l'ensemble des approches présentées ont été combinées afin de mettre au point une application permettant la classification des documents de presse numérisée par le contenu.

APA, Harvard, Vancouver, ISO, and other styles

45

Han, Yiding. "An Autonomous Unmanned Aerial Vehicle-Based Imagery System Development and Remote Sensing Images Classification for Agricultural Applications." DigitalCommons@USU, 2009. https://digitalcommons.usu.edu/etd/513.

Full text

Abstract:

This work concentrates on the topic of remote sensing using a multispectral imag-ing system for water management and agriculture applications. The platform, which is alight-weight inexpensive runway-free unmanned aerial vehicle (UAV), namely, AggieAir, ispresented initially. A major portion of this work focuses on the development of a light-weight multispectral imager payload for the AggieAir platform, called GhostFoto. Theimager is band-recongurable, covering both visual red, green, and blue (RGB) and nearinfrared (NIR) spectrum, and interfaced with UAV on-board computer. The developmentof the image processing techniques, which are based on the collected multispectral aerialimages, is also presented in this work. One application is to perform fully autonomous rivertracking for applications such as river water management. Simulation based on aerial mul-tispectral images is done to demonstrate the feasibility of the developed algorithm. Othereort is made to create a systematic method to generate normalized difference vegetationindex (NDVI) using the airborne imagery. The GhostFoto multispectral imaging systembased on AggieAir architecture is proven to be an innovative and useful tool.

APA, Harvard, Vancouver, ISO, and other styles

46

Rimer, Michael Edwin. "Improving Neural Network Classification Training." Diss., CLICK HERE for online access, 2007. http://contentdm.lib.byu.edu/ETD/image/etd2094.pdf.

Full text

APA, Harvard, Vancouver, ISO, and other styles

47

Colak, Tufan, and Rami S. R. Qahwaji. "Automated McIntosh-Based Classification of Sunspot Groups Using MDI Images." Springer, 2007. http://hdl.handle.net/10454/4091.

Full text

Abstract:

yes
This paper presents a hybrid system for automatic detection and McIntosh-based classification of sunspot groups on SOHO/MDI white-light images using active-region data extracted from SOHO/MDI magnetogram images. After sunspots are detected from MDI white-light images they are grouped/clustered using MDI magnetogram images. By integrating image-processing and neural network techniques, detected sunspot regions are classified automatically according to the McIntosh classification system. Our results show that the automated grouping and classification of sunspots is possible with a high success rate when compared to the existing manually created catalogues. In addition, our system can detect and classify sunspot groups in their early stages, which are usually missed by human observers.
EPSRC

APA, Harvard, Vancouver, ISO, and other styles

48

"Image partial blur detection and classification." 2008. http://library.cuhk.edu.hk/record=b5893527.

Full text

Abstract:

Liu, Renting.
Thesis (M.Phil.)--Chinese University of Hong Kong, 2008.
Includes bibliographical references (leaves 40-46).
Abstracts in English and Chinese.
Chapter 1 --- Introduction --- p.1
Chapter 2 --- Related Work and System Overview --- p.6
Chapter 2.1 --- Previous Work in Blur Analysis --- p.6
Chapter 2.1.1 --- Blur detection and estimation --- p.6
Chapter 2.1.2 --- Image deblurring --- p.8
Chapter 2.1.3 --- Low DoF image auto-segmentation --- p.14
Chapter 2.2 --- System Overview --- p.15
Chapter 3 --- Blur Features and Classification --- p.18
Chapter 3.1 --- Blur Features --- p.18
Chapter 3.1.1 --- Local Power Spectrum Slope --- p.19
Chapter 3.1.2 --- Gradient Histogram Span --- p.21
Chapter 3.1.3 --- Maximum Saturation --- p.24
Chapter 3.1.4 --- Local Autocorrelation Congruency --- p.25
Chapter 3.2 --- Classification --- p.28
Chapter 4 --- Experiments and Results --- p.29
Chapter 4.1 --- Blur Patch Detection --- p.29
Chapter 4.2 --- Blur degree --- p.33
Chapter 4.3 --- Blur Region Segmentation --- p.34
Chapter 5 --- Conclusion and Future Work --- p.38
Bibliography --- p.40
Chapter A --- Blurred Edge Analysis --- p.47

APA, Harvard, Vancouver, ISO, and other styles

49

Chang, Chia-Chin, and 張傢欽. "An Image Spam Classification Method for Rotating and Scaling Images." Thesis, 2015. http://ndltd.ncl.edu.tw/handle/96923222784300373599.

Full text

Abstract:

碩士
朝陽科技大學
資訊工程系
103
This thesis proposes a novel algorithm to effectively identify the image spam for the problems of image rotation, scaling and the background interference. This algorithm uses image color layering to reduce the interference of the background color and adopts seven resistant scaling parameters to identifier, and capture the image objects in oval way. This thesis also proposes a hybrid image spam filter. The first part uses Optical Character Recognition (OCR) to capture the text that embed in the image, and used keyword list to filter the identify the spam. The second part uses the proposed algorithm to filter the rotation and scaling of image spam. The experimental results demonstrated that the proposed algorithm can achieve a good classification results.

APA, Harvard, Vancouver, ISO, and other styles

50

Bijaoui, Jérôme. "Complémentarité des images optiques et radars pour la connaissance des littoraux." Phd thesis, 1995. http://pastel.archives-ouvertes.fr/pastel-00979423.

Full text

Abstract:

En raison du nombre croissant des populations vivant sur les littoraux, il devient capital de caractériser ces milieux ainsi que leurs évolutions. De par sa diversité instrumentale, la télédétection spatiale présente des possibilités exceptionnelles pour analyser les caractéristiques des littoraux. Ce travail porte sur l'étude de certaines propriétés du littoral à l'aide des images de télédétection. Le trait de côte caractérise la position du rivage. Sous certaines conditions (tempêtes, raz de marée ...), ce dernier peut évoluer rapidement et dans des proportions importantes. À l'aide des radars imageurs, il est possible d'observer les littoraux dans de telles situations. Pour extraire le trait de côte, nous proposons une méthode exploitant la texture de ces images. Cette méthode est évaluée en la comparant avec d'autres méthodes sur un site étalonné, ainsi que sur d'autres sites et avec des instruments différents. Ce travail porte aussi sur l'analyse du milieu côtier à l'aide d'instruments optiques. Nous montrons que les méthodes usuelles d'analyse permettent, en général, de ne traiter qu'une seule caractéristique parmi l'ensemble. Cette façon d'opérer nous a conduit à définir une méthode plus générale pour analyser les caractéristiques optiques du milieu marin non plus séparément mais ensemble. Basée sur une modélisation du transfert radiatif dans les eaux côtières, cette méthode offre des informations sur les natures de l'eau, des fonds marins et sur la profondeur. Cette méthode est évaluée à l'aide de simulations numériques. Elle est ensuite appliquée sur des images acquises avec un spectromètre aéroporté et avec un radiomètre spatial. Ces exemples montrent que cette méthode offre des résultats prometteurs pour l'étude des eaux côtières. Cette étude montre la complémentarité des instruments spatiaux pour l'étude du littoral. Pour obtenir des cartes riches et précises, nous montrons qu'il faut tenir compte d'images acquises à des dates distinctes et avec des instruments différents.

APA, Harvard, Vancouver, ISO, and other styles

Dissertations / Theses on the topic 'Classifications des images'

Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles