Dissertations / Theses on the topic 'Splicing detection'

Context: There has been a rise in the usage of digital images these days. Digital images are being used in many areas like in medicine, wars, etc. As the images are being used to make many important decisions, it is necessary to know if the images used are clean or forged. In this thesis, we have considered the area of splicing forgery. In this thesis, we are also considering and analyzing the impact of low-resolution images on the considered algorithms. Objectives. Through this thesis, we try to improve the detection rate of splicing forgery detection. We also examine how the examined splicing forgery detection algorithm works on low-resolution images and considered classification algorithms (classifiers). Methods: The research methods used in this research are Implementation and Experimentation. Implementation was used to answer the first research question i.e., to improve the detection rate in splicing forgery. Experimentation was used to answer the second research question. The results of the experiment were analyzed using statistical analysis to find out how the examined algorithm works on different image resolutions and on the considered classifiers. Results: One-tailed Wilcoxon signed rank test was conducted to compare which algorithm performs better, the T+ value obtained was less than To so the null hypothesis was rejected and the alternative hypothesis which states that Algorithm 2 (our enhanced version of the algorithm) performs better than Algorithm 1 (original algorithm), is accepted. Experiments were conducted and the accuracy of the algorithms in different cases were noted, ROC curves were plotted to obtain the AUC parameter. The accuracy, AUC parameters were used to determine the performance of the algorithms. Conclusions: After the results were analyzed using statistical analysis, we came to the conclusion that Algorithm 2 performs better than Algorithm 1 in detecting the forged images. It was also observed that Algorithm 1 improves its performance on low-resolution images when trained on original images and tested on images of different resolutions but, in the case of Algorithm 2, its performance is improved when trained and tested on images of the same resolution. There was not much variance in the performance of both of the algorithms on images of different resolution. Coming to the classifiers, Algorithm 1 improves its performance on linear SVM whereas Algorithm 2 improves its performance when using the simple tree classifier.

Dans cette thèse, nous proposons trois nouvelles méthodes d’imagerie médico-légale pour détecter l’épissage dans les images numériques en exploitant les statistiques de bruit d’image. Pour ce faire, nous introduisons un nouvel outil, l’histogramme de la densité du bruit, et son dérivé, l’histogramme de la contribution de la densité du bruit. Nos méthodes permettent la détection d’épissage sur les images brutes et JPEG. Bien que l’utilisation de différences de bruit pour détecter l’épissage ait déjà été faite plusieurs fois, la plupart des méthodes existantes ont tendance à mal fonctionner sur la génération actuelle d’images de haute qualité, avec une haute résolution et faible bruit. L’efficacité de nos approches est démontrée sur un grand ensemble d’images de ce genre, avec des splicings générés aléatoirement. Nous présentons également une analyse détaillée de l’évolution du bruit dans un appareil photo numérique, et comment il affecte divers existant
In this dissertation, we offer three new forensics imagery methods to detect splicing in digital images by exploiting image noise statistics. To do so, we introduce a new tool, the noise density histogram, and its derivative, the noise density contribution histogram. Our methods allow splicing detection on both raw and JPEG images. Although the use of noise discrepancies to detect splicing has already been done multiple times, most existing methods tend to perform poorly on the current generation of high quality images, with high resolution and low noise. The effectiveness of our approaches are demonstrated over a large set of such images, with randomly-generated splicings. We also present a detailed analysis of the evolution of the noise in a digital camera, and how it affects various existing forensics approaches. In a final part, we use the tool we developed in a counter-forensics approach, in order to hide the trace left by splicing on the image noise

Alternative RNA splicing is one of the leading mechanisms contributing towards transcript and protein diversity. Several alternative splicing surveys have confirmed the frequent occurrence of exon skipping in human genes. However, the occurrence of exon skipping in mouse genes has not yet been extensively examined. Recent improvements in mouse genome sequencing have permitted the current study to explore the occurrence of exon skipping in mouse genes orthologous to human genes on chromosome 22. A low number (5/72 multi-exon genes) of mouse exon-skipped genes were captured through alignments of mouse ESTs to mouse genomic contigs. Exon-skipping events in two mouse exon-skipped genes (GNB1L, SMARCB1) appear to affect biological processes such as electron and protein transport. All mouse, skipped exons were observed to have ubiquitous tissue expression. Comparison of our mouse exon-skipping events to previously detected human exon-skipping events on chromosome 22 by Hide et al.2001, has revealed that mouse and human exon-skipping events were never observed together within an orthologous gene-pair. Although the transcript identity between mouse and human orthologous transcripts were high (greater than 80% sequence identity), the exon order in these gene-pairs may be different between mouse and human orthologous genes.

Main factors contributing towards the low detection of mouse exon-skipping events include the lack of mouse transcripts matching to mouse genomic sequences and the under-prediction of mouse exons. These factors resulted in a large number (112/269) of mouse transcripts lacking matches to mouse genomic contigs and nearly half (12/25) of the mouse multi-exon genes, which have matching Ensembl transcript identifiers, have under-predicted exons. The low frequency of mouse exon skipping on chromosome 22 cannot be extrapolated to represent a genome-wide estimate due to the small number of observed mouse exon-skipping events. However, when compared to a higher estimate (52/347) of exon skipping in human genes for chromosome 22 produced under similar conditions by Hide et al.2001, it is possible that our mouse exon-skipping frequency may be lower than the human frequency. Our hypothesis contradicts with a previous study by Brett et al.2002, in which the authors claim that mouse and human alternative splicing is comparable. Our conclusion that the mouse exon-skipping frequency may be lower than the human estimate remains to be tested with a larger mouse multi-exon gene set. However, the mouse exon-skipping frequency may represent the highest estimate that can be obtained given that the current number (87) of mouse genes orthologous to chromosome 22 in Ensembl (v1 30th Jan. 2002) does not deviate significantly from our total number (72) of mouse multi-exon genes. The quality of the current mouse genomic data is higher than the one utilized in this study. The capture of mouse exon-skipping events may increase as the quality and quantity of mouse genomic and transcript sequences improves.

Les techniques de séquençage à haut-débit développées ces dernières années ont permis d'identifier des milliers d’ARN non-codants et des événements tels que l’épissage ou l’édition. Cette approche est à l’origine d’une meilleure compréhension des mécanismes régulant l'expression des gènes. Les longs ARN non-codants (lncARN) ont ainsi émergé comme des acteurs clés de la régulation de divers processus développementaux. Ils agissent soit directement sous leur forme longue par des interactions lncARN-protéine(s) soit après une étape de maturation qui génère des siARN ou des miARN régulateurs, menant à l’extinction génique par clivage des ARNm, la répression de la traduction ou en entrainant des modifications épigénétiques (ADN/chromatine) de leurs cibles. L’objectif de cette thèse était d’élucider les mécanismes d'action de lncARNs dans le développement de la plante. J'ai contribué à l'analyse de l'action du lncARN APOLO dans la régulation de la topologie de la chromatine chez Arabidopsis thaliana. Ensuite, j’ai concentré mes efforts sur le lncARN ASCO (Alternative Splicing COmpetitor) qui interagit avec les protéines NSRs (Nuclear Speckles RNA-binding Proteins) et participent au patron d’épissage de certains gènes cibles. Lors d’un traitement par l’auxine, NSRb est induit alors qu’ASCO est réprimé dans les racines. Le même type de traitement, chez le double mutant nsra/b et les lignées surexprimant ASCO, entraine déficience partielle dans la formation des racines latérales. En utilisant un nouvel outil bio-informatique appelé "RNAprof", nous avons détecté 1885 ARN différentiellement maturés entre le mutant nsra/b et la lignée sauvage traités à l’auxine. Parmi ces gènes, nous avons identifié ARF19, un régulateur clé de la voie de signalisation de l’auxine au cours de l'initiation et le développement de la racine. J’ai démontré qu'ARF19 interagit directement avec les NSRs et qu’il est différentiellement polyadénylé dans le double mutant nsra/b, conduisant à une isoforme plus courte du transcrit ARF19. D’autre part, parmi les gènes dérégulés de manière transcriptionnelle chez le mutant des gènes impliqués dans la signalisation par l’éthylène ont été identifiés. J’ai ensuite montré que plusieurs de ces gènes sont aussi dérégulés dans les plantes mutantes arf19-1 et arf19-2 en réponse à l’auxine, soutenant un rôle d'ARF19 dans la réponse croisée entre l’auxine et l’éthylène. Le gène NSRb est induit par l'éthylène et l'inhibition de la synthèse d'éthylène par l'AVG complémente le phénotype de racine latérale du mutant nsra/b en réponse à l’auxine. De plus, l'AVG et la surexpression d’ASCO augmentent l'accumulation de l’isoforme courte d’ARF19. Cette étude met en avant la capacité du lncARN ASCO à moduler l’épissage par le détournement des NSRs et la capacité des ARN non-codants à moduler l’épissage
In the last years, high-throughput sequencing techniques have made possible to identify thousands of noncoding RNAs and a plethora of different mRNA processing events occurring in higher organisms. This led to a better understanding of different regulatory mechanisms controlling gene expression. Long noncoding RNAs (lncRNAs) are emerging as key players in the regulation of varied developmental processes. They can act directly in a long form by lncRNA-protein interactions or be processed into shorter small si/miRNAs, leading to mRNA cleavage, translational repression or epigenetic DNA/chromatin modification of their targets. In this study, we aim to understand the mechanism of action of lncRNAs in plant development. Initially, I contributed to the analysis of the action of the APOLO lncRNA in chromatin topology regulation. Then, I focused my work on the lncRNA ASCO (Alternative Splicing COmpetitor) that interacts with NSRs (Nuclear Speckles RNA-binding Proteins) to modulate the splicing pattern of NSR-regulated mRNA targets. Auxin treatment induces NSRb and represses ASCO expression in roots. The nsra/b double mutant and ASCO overexpressing lines treated with auxin are partially impaired in lateral root formation. Using a new bioinformatic tool called “RNAprof”, we detected 1885 differential RNA processing events genome-wide in auxin-treated nsra/b mutants compared to WT. Among them, we identified ARF19, a key regulator of auxin signaling in lateral root initiation and development. I demonstrated that ARF19 is directly bound by both NSRs and that in the nsra/b double mutant ARF19 is alternatively polyadenylated leading to a short transcript isoform. Furthermore, among the transcriptionally deregulated genes in the nsra/b mutant plants, I identified an important group related to ethylene response. I further showed that several of these genes are also deregulated in the arf19-1 and arf19-2 mutants plants in response to auxin, supporting a role of ARF19 in the auxin-ethylene crosstalk. NSRb is also induced by ethylene and the inhibition of ethylene synthesis by AVG rescues the nsra/b double mutant lateral root phenotype in response to auxin. Moreover, AVG and ASCO overexpression lead to increased accumulation of the ARF19 short isoform. Altogether, this study shed new light on the role of the lncRNA ASCO in the regulation of RNA processing by hijacking NSRs and the capacity of non-coding RNAs to modulate splicing

Aujourd'hui, la manipulation d'images à des fins non éthiques est courante, notamment sur les réseaux sociaux et dans la publicité. Les utilisateurs malveillants peuvent par exemple créer des images synthétiques convaincantes pour tromper le public ou dissimuler des messages dans des images numériques, posant des risques pour la sécurité nationale. Les chercheurs en analyse forensique d'image travaillent donc avec les forces de l'ordre pour détecter ces manipulations. Les méthodes d'analyse forensique les plus avancées utilisent notamment des réseaux neuronaux convolutifs pour les détecter. Cependant, ces réseaux sont entraînés sur des données préparées par des équipes de recherche, qui diffèrent largement des données réelles rencontrées en pratique. Cet écart réduit considérablement l'efficacité opérationnelle des détecteurs de manipulations d'images. Cette thèse vise précisément à améliorer l'efficacité des détecteurs de manipulation d'images dans un contexte pratique, en atténuant l'impact de ce décalage de données. Deux stratégies complémentaires sont explorées, toutes deux issues de la littérature en apprentissage automatique : 1. Créer des modèles capables d'apprendre à généraliser sur de nouvelles bases de données ou 2. Sélectionner, voire construire, des bases d'entraînement représentatives des images à examiner. Pour détecter des manipulations sur un grand nombre d'images non étiquetées, les stratégies d'adaptation de domaine cherchant à plonger les distributions d'entraînement et d'évaluation dans un espace latent où elles coïncident peuvent se révéler utiles. Néanmoins, on ne peut nier la faible efficacité opérationnelle de ces stratégies, étant donné qu'elles supposent un équilibre irréaliste entre images vraies et manipulées parmi les images à examiner. En plus de cette hypothèse problématique, les travaux de cette thèse montrent que ces stratégies ne fonctionnent que si la base d'entraînement guidant la détection est suffisamment proche de la base d'images sur laquelle on cherche à évaluer, une condition difficile à garantir pour un praticien. Généraliser sur un petit nombre d'images non étiquetées est encore plus difficile bien que plus réaliste. Dans la seconde partie de cette thèse, nous abordons ce scénario en examinant l'influence des opérations de développement d'images traditionnelles sur le phénomène de décalage de données en détection de manipulation d'images. Cela nous permet de formuler des stratégies pour sélectionner ou créer des bases d'entraînement adaptées à un petit nombre d'images. Notre contribution finale est une méthodologie qui exploite les propriétés statistiques des images pour construire des ensembles d'entraînement pertinents vis-à-vis des images à examiner. Cette approche réduit considérablement le problème du décalage de données et permet aux praticiens de développer des modèles sur mesure pour leur situation
Today, it is easier than ever to manipulate images for unethical purposes. This practice is therefore increasingly prevalent in social networks and advertising. Malicious users can for instance generate convincing deep fakes in a few seconds to lure a naive public. Alternatively, they can also communicate secretly hidding illegal information into images. Such abilities raise significant security concerns regarding misinformation and clandestine communications. The Forensics community thus actively collaborates with Law Enforcement Agencies worldwide to detect image manipulations. The most effective methodologies for image forensics rely heavily on convolutional neural networks meticulously trained on controlled databases. These databases are actually curated by researchers to serve specific purposes, resulting in a great disparity from the real-world datasets encountered by forensic practitioners. This data shift addresses a clear challenge for practitioners, hindering the effectiveness of standardized forensics models when applied in practical situations.Through this thesis, we aim to improve the efficiency of forensics models in practical settings, designing strategies to mitigate the impact of data shift. It starts by exploring literature on out-of-distribution generalization to find existing strategies already helping practitioners to make efficient forensic detectors in practice. Two main frameworks notably hold promise: the implementation of models inherently able to learn how to generalize on images coming from a new database, or the construction of a representative training base allowing forensics models to generalize effectively on scrutinized images. Both frameworks are covered in this manuscript. When faced with many unlabeled images to examine, domain adaptation strategies matching training and testing bases in latent spaces are designed to mitigate data shifts encountered by practitioners. Unfortunately, these strategies often fail in practice despite their theoretical efficiency, because they assume that scrutinized images are balanced, an assumption unrealistic for forensic analysts, as suspects might be for instance entirely innocent. Additionally, such strategies are tested typically assuming that an appropriate training set has been chosen from the beginning, to facilitate adaptation on the new distribution. Trying to generalize on a few images is more realistic but much more difficult by essence. We precisely deal with this scenario in the second part of this thesis, gaining a deeper understanding of data shifts in digital image forensics. Exploring the influence of traditional processing operations on the statistical properties of developed images, we formulate several strategies to select or create training databases relevant for a small amount of images under scrutiny. Our final contribution is a framework leveraging statistical properties of images to build relevant training sets for any testing set in image manipulation detection. This approach improves by far the generalization of classical steganalysis detectors on practical sets encountered by forensic analyst and can be extended to other forensic contexts

Digital images have become the most important source of information. Due to presence of various image editing tools, images can be easily changed and altered. Therefore, the authentication of digital images has become an important issue. Forgery is performed by copying one part of an image somewhere else in same image. Copied part can be rotated, scaled or cropped while duplicating. So it is necessary to distinguish between authentic and forged images. These techniques are divided into two varieties- one being active i.e. intrusive. It means one needs to embed something in image example watermark, if the image is modified then the embedded data is also modified. Another one is passive i.e non-intrusive. It is a signature based technique. The work presents and compares feature selection algorithms for the detection of image forgery in tampered images. Various features are extracted from normal and spliced using spatial gray level dependence method and many more. Support vector machine has been used for classification. A very difficult problem in classification techniques is the choice of features to distinguish between classes. The feature optimization problem is addressed using a genetic algorithm (GA) as a search method. Classical sequential methods and floating search algorithm are compared against the genetic approach in terms of the best recognition rate achieved and the optimal number of features.

Tese de mestrado, Bioinformática e Biologia Computacional (Bioinformática) Universidade de Lisboa, Faculdade de Ciências, 2017
As florestas de sobreiro (Quercus suber L.) são recursos únicos e emblemáticos em Portugal, com elevado impacto económico, ecológico e social. A disponibilidade recente da sequência do genoma de sobreiro forneceu um importante contributo para revitalizar a pesquisa em temas como desenvolvimento de cortiça e melhoramento da planta, assim como promover a competitividade da indústria da cortiça. No entanto, é ainda necessário adicionar mais detalhe à anotação estrutural do genoma, nomeadamente ao nível dos transcritos, incluindo previsão de eventos de splicing alternativo. O splicing alternativo (AS) é um processo usado durante a expressão génica que origina diferentes variantes de transcritos (isoformas) e produtos proteicos a partir um único gene. No presente estudo, procedemos à análise de dezasseis bibliotecas de RNA-seq, preparadas a partir de quatro tecidos de sobreiro (folhas, felema, entrecasco e xilema), de modo a prever novas formas de AS para genes já previstos e melhorar a anotação estrutural do genoma. Um protocolo bioinformático foi definido para testar o desempenho do software HISAT2 e STAR para mapeamento de reads de RNAseq no genoma de referência, e do software Cufflinks e StringTie para (re)construção de transcritos. O alinhamento de reads no genoma efetuado com STAR resultou em taxas de mapeamento (de 84,22% a 86,86%) superiores aos resultados atingidos com HISAT2 (73,88% a 76,55%). Assim, os resultados de mapeamento com STAR foram utilizados para a (re)construção de transcritos. O uso do StringTie para este processo foi globalmente mais conservador do que com Cufflinks, gerando menos transcritos novos, mas com melhor cobertura de reads por pares de base. Para melhorar a precisão da anotação e reduzir falsos positivos, foi realizado um passo adicional de otimização com StringTie. Desta otimização resultou uma anotação que prevê a ocorrência de 7 958 novos transcritos (8% dos transcritos totais), dos quais 5 453 são novas isoformas para genes previstos na anotação de referência. Esta nova anotação foi utilizada como referência para estimar a abundância dos transcritos em cada um dos tecidos estudados e efetuar a análise de expressão diferencial. Cerca de 16% de todos os genes expressos nos quatro tecidos e que contêm intrões apresentaram splicing alternativo, e os principais eventos de splicing foram alternative acceptor site e intron retention. Grupos de transcritos com expressão diferencial entre os quatro tecidos foram identificados e a análise de enriquecimento funcional confirmou os principais processos biológicos esperados para cada tecido: os transcritos mais expressos nas folhas e no xilema estavam relacionados com a fotossíntese e com transporte, respetivamente; transcritos mais expressos na periderme (felema e entrecasco) mostraram um enriquecimento em categorias funcionais relacionadas com a síntese de suberina e outros componentes de parede celular presentes nas células de cortiça. Estes grupos específicos mostraram também um enriquecimento em transcritos envolvidos na resposta ao stresse (biótico ou abiótico). Nos tecidos que compõem a periderme, este enriquecimento foi observado principalmente no entrecasco, enquanto que no felema foi detetado um enriquecimento em transcritos envolvidos no metabolismo secundário. A presente tese permitiu a definição de um protocolo padrão que poderá ser usado para estudar o splicing alternativo no sobreiro e para uma análise mais aprofundada na nova versão do genoma, que estará disponível em breve.
Cork oak (Quercus suber L.) forests are unique and emblematic resources for Portugal, with high economical, ecological and social significance. The recent availability of the cork oak genome sequence provided an important contribution to reinvigorate research in fundamental topics such as cork development and plant improvement, and to promote the competitiveness of cork industry. Yet, further analysis is required to add detail to genome structure annotation, namely at the transcript level, also taking into account alternative splicing. Alternative splicing (AS) is a process used during gene expression to yield different transcript variants and protein products derived from a single gene. In the present study, we analyzed sixteen RNA-seq libraries prepared from four cork oak tissues (leaf, xylem, phellem and inner bark), in order to predict new AS forms for the already predicted genes and improve genome structural annotation. A bioinformatics pipeline was defined in order to test the performance of HISAT2 and STAR for read mapping against the reference genome, and Cufflinks and StringTie for transcript assembly. STAR yielded higher mapping efficiencies (84.22% to 86.86%) for the cork oak datasets, as compared to HISAT2 (73.88% to 76.55%), and the corresponding mapping data was selected for transcript assembly. The use of StringTie for this step was globally more conservative than Cufflinks, generating less novel transcripts, but with better support by read per base coverage. A further optimization step was performed using StringTie in order to improve annotation precision. The final transcript annotation was selected from this optimization step, predicting 7,958 novel transcripts (8% of total transcripts in the new annotation), 5,453 of which were novel isoforms for genes in reference annotation. This new annotation was used as reference to estimate transcript abundance in each tissue and differential expression analysis. Approximately 16% of all intron-containing genes expressed in the four tissues were alternatively spliced and the main event found in the four cork oak tissues was alternative acceptor site, followed by intron retention. Transcript clusters showing differential expression among the four tissues were identified and functional enrichment analysis confirmed the main biological processes expected for each tissue: transcripts highly expressed in leaves and xylem were mostly related to photosynthesis and transport, respectively; transcripts highly expressed in peridermis (phellem and inner bark) showed an enrichment in functional categories related to the synthesis of suberin and other component of cork cell walls. These tissue-specific clusters also showed an enrichment in transcripts involved in the response to stress (biotic or abiotic). Yet, in peridermis, this enrichment was mostly observed in inner bark samples, while phellem samples showed an enrichment in transcripts related to secondary metabolism. This thesis allowed the definition of a standard workflow that can be used to study alternative splicing in cork oak and used for further analysis on the new improved genome version that will be available soon.

碩士
國立臺北科技大學
電機工程系所
101
Due to the rapid development of image editing software, it becomes easy to tamper a digital image and create a fake image. One of the common and simple ways of the digital image tampering is image splicing, which is to copy a region of an image to another image, and add the contents not belonging to the original image. Therefore, the authenticity of digital image has become an important issue. Since different sources of images are productions of different sensors and post-processing, they will contain different noise levels. Based on this fact, we propose a multi-scale PCA noise level estimation algorithm to detect image splicing, and improve the detection rate of the existing works. In addition, our method can locate the tampering region by clustering the estimated noise with the integration of EGB segmentation. In the experiments, we use the Columbia uncompressed image splicing detection evaluation dataset to test our method. The experimental results show that our method can detect the image splicing robustly and accurately.

碩士
國立東華大學
資訊工程學系
99
Digital images can be easily forged with various image processing tools; nowadays the issue of digital image forgery becomes more and more important. As JPEG has been popularly used in image compression standard, forgery detection of JPEG images plays an important role now. Forging on compressed images often involves recompression and tends to erase those forging traces existed in un-compressed images. We could, however, attempt to discover new traces caused by recompression and use these traces to detect the recompressed image’s forgery. The artifacts introduced by lossy JPEG compression can be regarded as an inherent feature for recompressed images. In this thesis, a novel forgery image detection for splicing and copy-move forgery image is proposed. We first use a forgery image detection approach by periodicity analysis with the double compression effect in spatial and DCT domain. Then, the feature extracted by SURF descriptors is applied to resisting the variation of rotation and/or scaling. Experimental results demonstrate that the proposed technique is performed well on the detection of forgery localization. Especially for the copy-move forgery images, the proposed technique is able to locate the forged regions and recognize the non-original regions.

abstract: Splicing of digital images is a powerful form of tampering which transports regions of an image to create a composite image. When used as an artistic tool, this practice is harmless but when these composite images can be used to create political associations or are submitted as evidence in the judicial system they become more impactful. In these cases, distinction between an authentic image and a tampered image can become important. Many proposed approaches to image splicing detection follow the model of extracting features from an authentic and tampered dataset and then classifying them using machine learning with the goal of optimizing classification accuracy. This thesis approaches splicing detection from a slightly different perspective by choosing a modern splicing detection framework and examining a variety of preprocessing techniques along with their effect on classification accuracy. Preprocessing techniques explored include Joint Picture Experts Group (JPEG) file type block line blurring, image level blurring, and image level sharpening. Attention is also paid to preprocessing images adaptively based on the amount of higher frequency content they contain. This thesis also recognizes an identified problem with using a popular tampering evaluation dataset where a mismatch in the number of JPEG processing iterations between the authentic and tampered set creates an unfair statistical bias, leading to higher detection rates. Many modern approaches do not acknowledge this issue but this thesis applies a quality factor equalization technique to reduce this bias. Additionally, this thesis artificially inserts a mismatch in JPEG processing iterations by varying amounts to determine its effect on detection rates.
Dissertation/Thesis
Masters Thesis Computer Science 2015

The widespread availability of image editing tools and improvements in image processing techniques make image manipulation feasible for the general population. Oftentimes, easy-to-use yet sophisticated image editing tools produce results that contain modifications imperceptible to the human observer. Distribution of forged images can have drastic ramifications, especially when coupled with the speed and vastness of the Internet. Therefore, verifying image integrity poses an immense and important challenge to the digital forensic community. Satellite images specifically can be modified in a number of ways, such as inserting objects into an image to hide existing scenes and structures. In this thesis, we describe the use of a Conditional Generative Adversarial Network (cGAN) to identify the presence of such spliced forgeries within satellite images. Additionally, we identify their locations and shapes. Trained on pristine and falsified images, our method achieves high success on these detection and localization objectives.

碩士
國立成功大學
電腦與通信工程研究所
107
With the Internet development and the availability of image editing tools, digital images can be easily manipulated and edited. Therefore, the credibility of digital images has faced severe challenges. In digital image forensic, the copy-move and splicing forgeries are popular forgery attacks. For copy-move forgery, a part of the image is copied and pasted elsewhere in the same image in order to cover possible important messages. However, the image splicing is to duplicate a region of another image to the original image so as to add the contents not belonging to the original image. In this thesis, a convolutional neural network (CNN) model is proposed to detect such tampering. First, the image is divided into fixed-size non-overlapping patches and the Radon transform is applied to each patch to compute the features. After the network is trained, the proposed model can classify the tampered and the authentic patches. By classifying each patch in the images, the duplicated regions can be detected. The experimental results demonstrate that the accuracy of proposed method is better than other methods.

Alternative mRNA processing can result in the generation of multiple, qualitatively different RNA transcripts from the same gene and is a powerful engine of complexity in higher organisms. Recent deep sequencing studies have indicated that essentially all human genes containing more than a single exon generate multiple RNA transcripts. Functional roles of alternative processing have been established in virtually all areas of biological regulation, particularly in development and cancer. Changes in alternative mRNA processing can now be detected from over a billion dollars' worth of conventional gene expression microarray data archived over the past 20 years using a program we created called SplicerAV. Application of SplicerAV to publicly available microarray data has granted new insights into previously existing studies of oncogene over-expression and clinical cancer prognosis.

Adaptation of SplicerAV to the new Affymetrix Human Exon arrays has resulted in the creation of SplicerEX, the first program that can automatically categorize microarray detected changes in alternative processing into biologically pertinent categories. We use SplicerEX's automatic event categorization to identify changes in global mRNA processing during B cell transformation and show that the conventional U133 platform is able to detect 3' located changes in mRNA processing five times more frequently than the Human Exon array.

Dissertation