Rozprawy doktorskie na temat „PERCEPTUAL LOOS”

Image-to-semantic labels classification is a very challenging task in image processing. Convolutional neural networks (CNN) have managed to achieve the state-of-the-art quality of the segmented image in semantic segmentation tasks. Still, the classification capability of such algorithms is not satisfactory to segment images that contain complex object boundaries and minimal regions. Recently, the Generative Adversarial Networks (GAN) were introduced, which can solve the overfitting of the generator network using the adversarial loss. In this paper, a GAN-based segmentation model is proposed, in which the Conditional Generative Adversarial Networks (CGAN) model is used as base architecture. Perceptual loss is introduced in this composite model to solve the identification and classification of visually small elements in images. A pre-trained deep convolution neural network is adopted to generate improved segmentation masks to calculate Perceptual loss. The usage of Perceptual loss has demonstrated the high quality of the output labels. The evaluation of the proposed model on the cityscapes dataset has shown the effectiveness of GAN-based architecture in semantic segmentation of multiclass images. The proposed model achieved 83.3% accuracy on the test dataset, which is superior to most semantic segmentation state-of-the-art methods.

The tradition of broadcast video is today complemented with user generated content, as portable devices support video coding. Similarly, computing is becoming ubiquitous, where Internet of Things (IoT) incorporate heterogeneous networks to communicate with personal and/or infrastructure devices. Irrespective, the emphasises is on bandwidth and processor efficiencies, meaning increasing the signalling options in video encoding. Consequently, assessment for pixel differences applies uniform cost to be processor efficient, in contrast the Human Visual System (HVS) has non-uniform sensitivity based upon lighting, edges and textures. Existing perceptual assessments, are natively incompatible and processor demanding, making perceptual video coding (PVC) unsuitable for these environments. This research allows existing perceptual assessment at the native level using low complexity techniques, before producing new pixel-base image quality assessments (IQAs). To manage these IQAs a framework was developed and implemented in the high efficiency video coding (HEVC) encoder. This resulted in bit-redistribution, where greater bits and smaller partitioning were allocated to perceptually significant regions. Using a HEVC optimised processor the timing increase was < +4% and < +6% for video streaming and recording applications respectively, 1/3 of an existing low complexity PVC solution. Future work should be directed towards perceptual quantisation which offers the potential for perceptual coding gain.

The thesis that perceptual experience is transparent has received considerable air-time in contemporary philosophy of mind and perception. Debate over its truth-value has reached an impasse. I diagnose this mired debate, and pursue a reformulation of the “transparency thesis” such that it can be more readily evaluated form the perspective of perceptual psychology and related subdisciplines. I argue that the empirical methods characteristic of these disciplines are important for evaluating the transparency thesis. Both historical and contemporary empirical results but substantial pressure on the transparency thesis.

Thesis: Ph. D., Harvard-MIT Program in Health Sciences and Technology, 2016.
Cataloged from PDF version of thesis.
Includes bibliographical references.
The global hearing healthcare field is faced with two principal challenges. First, the demand for basic audiometric testing services far exceeds the capacity of trained clinicians even in high income countries, and this supply/demand mismatch is expected to worsen secondary to population aging. Next, once patients are identified as having a hearing loss, the treatments that are provided (hearing aids) do not sufficiently address their primary complaint, namely that they have trouble hearing in noisy environments. To begin to address the first problem, we executed a proof-of-concept study to ask whether mobile consumer electronics could be used to replace manually performed clinic-based testing with self-directed hearing measurements from home. We found that self-administered home hearing measurements were largely equivalent to standard clinical measures. To begin to address the second problem (hearing in noise challenges of patients), we performed three additional experiments. Inspired by promising findings of enhanced visual attention following action videogame training, we developed a closed-loop audiomotor training application and asked if playing a game that focused on tone in noise discriminations would provide generalized benefit for speech recognition in noise abilities. In young normally hearing adults, closed-loop training for one month provided a 12 percentage point improvement in speech understanding in noise scores. Next, we recruited older adults who wore hearing aids to play a similar closed-loop training game and observed a 10 percentage point enhancement of speech recognition in noise abilities secondary to gameplay, suggesting that this training could be coupled with standard treatments to improve patient outcomes. Finally, we studied the neurophysiological correlates of audiomotor signal in noise training in a rodent model, where we observed enhanced resistance to noise suppression in auditory cortical neurons following three months of training, perhaps contributing to the perceptual benefits that we observed in human subjects.
by Jonathon Whitton.
Ph. D.

This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections.
Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 2018
Cataloged from student-submitted PDF version of thesis.
Includes bibliographical references (pages 51-53).
Product perceptual maps are visualizations of the perceptions of products by customers. They provide many advantages to businesses, such as identifying gaps in the market, understanding competition, and finding how new products fit into a market. Conventional product perceptual mapping methods exhibit limitations, particularly in capturing the highly nonlinear structure in product perceptual categories. Therefore, given only a set of images and triplet data representing product co-occurence by consumers, we propose and use a Gaussian mixture variational autoencoder (GMVAE) with triplet loss to create product embeddings. These product embeddings are then flattened into a 2D perceptual map able to be interpreted by human judgment. We test the GMVAE approach on three datasets: (1) a dataset of simple generated data; (2) the MNIST dataset, a dataset of handwritten digits; and (3) the Amazon Fashion dataset, a dataset of product images, product categories, and similar products. The GMVAE method is quantitatively evaluated on its ability to capture product "latent" categories, and qualitatively evaluated on the quality of its 2D perceptual maps compared with those produced by using a conventional perceptual mapping method. We find that across the experiments, the GMVAE method could reasonable capture "latent" perceptual product categories and is more effective than the conventional perceptual mapping baseline in correctly identifying and predicting latent product categories.
by Mike Wang.
M. Eng.
M.Eng. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science

Thesis (Ph. D.) -- University of Maryland, College Park, 2006.
Thesis research directed by: Psychology. Title from t.p. of PDF. Includes bibliographical references. Published by UMI Dissertation Services, Ann Arbor, Mich. Also available in paper.

Aim : The purpose of this study was to examine the impact of simulated hearing loss on the acoustic contrasts between declarative questions and declarative statements and on the perception of speech intonation. A further purpose of the study was to investigate whether any such effects are universal or language specific. Method: Speakers included four native speakers of English and four native speakers of Mandarin and Taiwanese, with two female and two male adults in each group. Listeners included ten native English and ten native speakers of Mandarin and Taiwanese, with five female and five male adults in each group. All participants were aged between 19 and 55 years old. The speaker groups were asked to read a list of 28 phrases, with each phrase expressed as a declarative statement or a declarative question separately. These phrases were then filtered through six types of simulated hearing loss configurations, including three levels of temporal jittering for simulating a loss in neural synchrony, a high level of temporal jittering in combination with a high-pass or a low-pass filter that simulate falling and rising audiometric hearing loss configurations, and a vocoder processing procedure to simulate cochlear implant processing. A selection of acoustic measures was derived from the sentences and from some embedded vowels, including /i/, /a/, and /u/. The listener groups were asked to listen to the tokens in their native language and indicate if they heard a statement or a question. Results: The maximum fundamental frequency (F0) of the last syllable (MaxF0-last) and the maximum F0 of the remaining sentence segment (MaxF0-rest) were found to be consistently higher in declarative questions than in declarative statements. The percent jitter measure was found to worsen with simulated hearing loss as the level of temporal jittering increased. The vocoder-processed signals showed the highest percent jitter measure and the spread of spectral energy around the dominant pitch. Results from the perceptual data showed that participants in all three groups performed significantly worse with vocoder-processed tokens compared to the original tokens. Tokens with temporal jitter alone did not result in significantly worse perceptual results. Perceptual results from the Taiwanese group were significantly worse than the English group under the two filtered conditions. Mandarin listeners performed significantly worse with the neutral tone on the last syllable, and Taiwanese listeners performed significantly worse with the rising tone on the last syllable. Perception of male intonation was worse than female intonation with temporal jitter and high-pass filtering, and perception of female intonation was worse than male intonation with most temporal jittering conditions, including the temporal jitter and low-pass filtering condition. Conclusion: A rise in pitch for the whole sentence, as well as that in the final syllable, was identified as the main acoustic marker of declarative questions in all of the three languages tested. Perception of intonation was significantly reduced by vocoder processing, but not by temporal jitter alone. Under certain simulated hearing loss conditions, perception of intonation was found to be significantly affected by language, lexical tone, and speaker gender.

With reductions in the age criterion for cochlear implantation, the need for age-appropriate measures of speech perception skills has increased. One recently developed tool that shows great promise for the clinical assessment of auditory speech perception capacity in young children with cochlear implants is the On-Line Imitative Test of Speech-Pattern Contrast Perception (OlimSpac). The OlimSpac requires a child to imitate nonword utterances by providing a verbal response. The child's perceptual abilities are inferred from the child's productions through having a listener, who is masked to the stimulus select the utterance produced by the child in an eight-alternative force-choice task. Although the OlimSpac has the potential for use in children with cochlear implants, the specific role of measured speech production abilities on performance has yet to be systematically examined. Thus, the main objective of the present study was to examine the influence of speech production abilities on OlimSpac performance in an auditory-visual (A-V) and auditory-only (A-O) condition in young children with cochlear implants relative to an articulation-matched sample of normal hearing peers. A secondary objective was to determine whether the presentation modality affected a child's OlimSpac performance. A matched pair design was used to compare the OlimSpac performance of ten children with cochlear implants (ages 29 to 76 months) to normal hearing peers (ages 27 to 73 months). Each child with cochlear implants was matched to a child with normal hearing from a sampled population of 22 normal hearing participants based on word-level articulation skills, as measured by the GFTA-2, accounting for hearing age and gender. The OlimSpac software generated a score for a single contrast and a single presentation modality (A-V or A-O). The score was based on eight binary trials. Due to the small trial size, individual contrast scores were interpreted as pass/fail, given that only a score of seven or eight is significantly better than chance. Therefore, OlimSpac performance was determined by a composite score reflecting the average across all six contrasts, based on 48 binary trials. Average composite scores for both the A-V and A-O conditions were lower, albeit not significant, for young children with cochlear implants compared to their articulation-matched controls. Examination of individual phonemic contrast scores revealed that the articulation-matched sample of normal hearing participants most often failed the post-alveolar consonant place contrast, whereas children with cochlear implants most often failed the consonant voicing contrast. There were also no significant within group difference in speech perception performance between the A-V and A-O conditions of the OlimSpac. The results of this study demonstrated that children with cochlear implants achieved similar speech perception performance to their articulation-matched normal hearing peers. Although children's speech production abilities partially influenced their OlimSpac performance, knowledge of word-level articulation skills allows clinicians to make appropriate judgments when interpreting composite scores, thus validating the OlimSpac as an indirect measure of a child's speech perception capacity and a direct measure of speech perception skills. Objective scores obtained from a child's OlimSpac performance may be used to assess outcomes of cochlear implant use, guide cochlear implant mapping, and plan habilitative intervention. A greater understanding of the potential effect of speech production performance on estimates of speech perception ability may also assist in highlighting other developmental, linguistic and/or cognitive delays masked by a child's hearing loss.

De nombreuses études ont relevé un biais perceptif (BP) gauche, c'est-à-dire une tendance à utiliser préférentiellement les informations issues de l'hémi-visage gauche (du point de vue de l'observateur), lors du traitement des visages. Ce BP pourrait être lié à la dominance de l'hémisphère droit pour le traitement des visages. Par ailleurs, certaines études ont mis ce BP en lien avec un biais oculomoteur (BO) gauche : l'observateur effectue des fixations en nombre et/ou en durée supérieurs sur l'hémi-visage gauche (Butler et al., 2005 ; Megreya & Havard, 2011). Au cours du vieillissement, le BP gauche semble être toujours présent, mais d'apparition plus tardive, nécessitant un temps d'exposition aux visages plus long (Butler & Harvey, 2008 ; Coolican et al., 2008). Dans ce travail de thèse, nous avons étudié l'évolution de ces deux biais au cours du vieillissement au cours d'une tâche de jugement de genre, à l'aide de visages normaux et chimériques (composés de moitiés de visages d'homme et de femme) en manipulant le nombre de saccades permettant d'explorer le visage (aucune, 1, 2 et 3 saccades) et la position de présentation des visages (Haut, Bas, Centre, Gauche et Droite). Bien que dans l'ensemble, les personnes jeunes témoignent d'un BP gauche, cela n'est pas le cas des participants âgés. En outre, ce BP dépend de la position de présentation des visages (BP de proximité pour les positions latérales), mais également du nombre de saccades exécutées (BP gauche apparaissant à partir de la réalisation d'une saccade). En outre, une certaine variabilité interindividuelle est notée, certains participants présentant un BP droit et d'autre une absence de BP. Le BO dépend également de la position de présentation des visages et reste stable au cours du vieillissement. Il semble résulter d'une combinaison de l'effet du centre de gravité (Bindemann et al., 2009) et de l'effet de rang (Kapoula, 1985), la position d'arrivée se situant autour du centre du visage, orientée légèrement vers le point de fixation initial. Enfin, ce BO n'est pas affecté par le BP des participants, qu'ils soient jeunes ou âgés, l'exploration des visages étant la même qu'ils répondent en se basant sur la partie gauche ou droite des visages. Dans l'ensemble, ces résultats remettent en question l'explication jusqu'ici avancée pour ces deux biais, à savoir la dominance de l'HD pour le traitement des visages.

A utilização do mascaramento contralateral durante a pesquisa do potencial evocado auditivo de tronco encefálico é uma temática bastante discutida, visto a inexistência de um consenso na literatura acerca de sua necessidade. O objetivo deste estudo foi verificar a necessidade do uso do mascaramento contralateral no potencial evocado auditivo de tronco encefálico com estímulo tone burst apresentado com fone de inserção 3A e propor a aplicabilidade do ruído mascarador white noise na perda auditiva sensorioneural unilateral de graus severo ou profundo. Foram avaliados 15 indivíduos, de ambos os sexos, com o diagnóstico audiológico prévio de perda auditiva sensorioneural unilateral de graus severo ou profundo, com idades entre 2 a 40 anos. Para tanto, foi pesquisada a resposta neural definida pela presença da onda V na orelha com perda auditiva, nas frequências de 500, 1000, 2000 e 4000 Hz. Nos casos em que houve o registro, a pesquisa foi repetida com mascaramento white noise na orelha contralateral. Os registros foram analisados por dois juízes experientes em eletrofisiologia, a fim de verificar a concordância entre as análises. O Coeficiente Kappa (1,00) apresentou força de concordância quase perfeita e o de Correlação Intraclasse (1,00) foi excelente, entre as duas avaliadoras, em todas as análises. Como resultado, foi observada a audição cruzada, ao menos uma vez, em todas as frequências analisadas. Os níveis de mascaramento contralateral mínimos necessários para eliminar a participação da orelha não testada variaram de 10 a 20 dBnNA acima do limiar eletrofisiológico da orelha sem perda auditiva. Não foi possível realizar a análise para a frequência de 500 Hz devido à presença de artefato e ruído elétricos. Conclui-se que o uso do mascaramento contralateral no potencial evocado auditivo de tronco encefálico com estímulo tone burst apresentado com fone de inserção 3A é necessário nas frequências de 1000, 2000 e 4000 Hz em indivíduos com perda auditiva sensorioneural unilateral de graus severo ou profundo, nas intensidades de 15, 20 e 10 dB acima dos limiares eletrofisiológicos, respectivamente.
The use of contralateral masking on the investigation of auditory brainstem response is a topic widely discussed, however there is no consensus in literature about its necessity. The aim of this study was to verify the need of using masking in auditory brainstem response with tone burst stimulus presented with 3A insert earphones and to propose the applicability of white noise masking in severe or profound unilateral sensorineural hearing loss. Fifteen individuals of both genders with previous audiological diagnosis of severe or profound unilateral sensorineural hearing loss, aged between 2 to 40 years, were evaluated. For this, the neural response, which is defined by the presence of the V wave in the ear with hearing loss, was investigated in frequencies of 500, 1000, 2000 and 4000 Hz. In cases where there was a record, the search was repeated using white noise masking in the contralateral ear. The records were analyzed by two judges experienced in electrophysiology, in order to verify the concordance between the analyses. The Kappa coefficient (1.00) had almost perfect agreement strength and the Intraclass Correlation (1.00) was excellent between the two evaluators in all analyses. As a result, cross-audition was observed at least once in all analyzed frequencies. The minimum contralateral masking levels required to eliminate the untested ear involvement varied from 10 to 20 dBnNA above the electrophysiological threshold of the ear without hearing loss. It was not possible to perform the analysis for the frequency of 500 Hz due to the presence of electrical artifact and noise. It is concluded that the use of contralateral masking in the auditory brainstem response with tone burst stimulus presented with 3A insert earphone is necessary in frequencies of 1000, 2000 and 4000 Hz in individuals with severe or profound unilateral sensorineural hearing loss in intensities of 15, 20 and 10 dB above the electrophysiological thresholds, respectively.

La communication parlée peut être vue comme un processus interactif impliquant un couplage fonctionnel entre les systèmes moteur et sensoriel. L’objectif de ce travail de thèse est de tester ces possibles liens perceptivo-moteurs, aussi bien lors de la perception que de la production de la parole, au travers de multiples paradigmes comportementaux et auprès de différentes populations de participants. Le protocole expérimental mis en place est composé de trois paradigmes expérimentaux classiques : un premier paradigme de close-shadowing,visant à explorer le format partiellement moteur de stimuli auditifs et audiovisuels de parole ; un second paradigme permettant de mettre en évidence des corrélations entre la production et la perception de voyelles ; et enfin, un dernier paradigme d’imitation consciente et inconsciente de fréquence fondamentale.Suite à la validation de notre protocole expérimental auprès d’une population contrôle de jeunes adultes normo-entendants, nous avons étudié une seconde population constituée d’adultes normo-entendants séniors, et ceci afin d’évaluer la conséquence d'un déclin des fonctions cognitives et langagières. Les résultats obtenus ont permis de suggérer une activation fonctionnelle des liens perceptivo-moteurs lors de la perception et de la production de parole chez l’ensemble des participants. La troisième population testée était constituée de patients sourds post-linguaux puis implantés, afin de déterminer l'impact d'une déprivation sensorielle ainsi que les éventuels réapprentissages liés à leur implantation sur ces liens perceptivo-moteurs. De manièresurprenante, les résultats ont mis en évidence des relations sensori-motrices actives chez ces participants, et ce même très peu de temps après l’implantation. Pris ensemble, les résultats observés avec ces trois paradigmes expérimentaux et auprès de ces trois groupes de participants attestent de la nature perceptivo-motrice de la parole. De manière importante, malgré des performances dégradées, ces interactions entre systèmes sensoriels et moteur lors de la perception et de la production de parole resteraient fonctionnelles auprès des deux populations, celle des adultes normo-entendants séniors etcelle des patients sourds post-linguaux porteurs d’un implant cochléaire
Speech communication can be considered as an interactive process involving afunctional coupling between sensory and motor systems. The aim of this thesis was to test possible perceptuo-motor linkages during both speech perception and production, using distinct behavioral paradigms and populations. The experimental protocol was made of three classic experiments: a first paradigm of close-shadowing, aiming at exploring the partially motor format of audio and audiovisual stimuli; a second paradigm allowing to correlate production and perception of vowels; and a third paradigm of conscious and unconscious imitation of pitch. The experimental protocol was validated with a first group of young hearing adults. The second population studied was composed of elderly normal-hearing participants, in order to evaluate the consequences of both cognitive and linguistic declines. Results allowed us to suggest a functional activation of perceptuo-motor linkage during speech production and perception.The third population we tested comprised post-lingually deaf patients wearing acochlear implant. Our objective was to determine the impact of the sensorial deprivation and the re-learning processes, associated with their implantation, on perceptuo-motor linkages. Unexpectedly, results showed an active sensori-motor relationship in those participants, even shortly after the cochlear implantation. Altogether, our results confirmed the perceptuo-motor nature of speech. Importantly, in spite of degraded performances, these interactions between the sensory and the motor systems during speech production and perception remained functional in both the elderly normal-hearing population and the post-lingually deaf patients, wearing a cochlear implant

[ES] El control de ruido ambiental es una preocupación de primera magnitud para las sociedades avanzadas, debido a los problemas derivados que ocasionan en la salud de los ciudadanos. Una de las soluciones más extendidas para el control del ruido en su fase de transmisión en la utilización de pantallas acústicas. La aparición de nuevos materiales formados por redes de dispersores acústicos aislados, denominados cristales de sonido, está revolucionando el campo del apantallamiento acústico, posibilitando el avance tecnológico de esta área. Así, en los últimos años, las pantallas acústicas basadas en cristales de sonido se han posicionado como una alternativa viable a las pantallas acústicas tradicionales, puesto que ofrecen múltiples ventajas frente a las soluciones actuales. En el presente trabajo se muestra primeramente una recopilación de los avances realizados en el campo del apantallamiento acústico mediante esta tipología de pantallas. No obstante, aún existen líneas de investigación abiertas en esta área, que es necesario abordar para conseguir el objetivo de aplicar esta tecnología como atenuadores de sonido en las infraestructuras de transporte. Durante el periodo de formación de la doctoranda, se ha trabajado en algunas de las líneas de investigación activas en este campo del apantallamiento acústico. Una de estas investigaciones condujo al descubrimiento de interferencias entre los efectos de la resonancia y la dispersión múltiple de los cristales de sonido cuando estos efectos se producen en rangos de frecuencia cercanos. También hemos diseñado un nuevo dispositivo de reducción de ruido basado en cristales de sonido, utilizando herramientas de optimización multiobjetivo, que permitan apantallar y reflejar de forma difusa el ruido. El empleo de esta nueva herramienta de diseño identificó la necesidad de realizar un estudio comparativo de los métodos de simulación más utilizados para estimar el rendimiento de los dispositivos basados en cristales de sonido. Por último, hemos realizado un estudio psicoacústico para determinar la percepción de la reducción de molestia que proporcionan las pantallas acústicas basadas en cristales sonido y las barreras tradicionales, determinando si los parámetros objetivos que evalúan su rendimiento coinciden con la respuesta subjetiva de los usuarios.
[CA] El control de soroll ambiental és una preocupació de primera magnitud per a les societats avançades, a causa dels problemes derivats que ocasionen en la salut dels ciutadans. Una de les solucions més esteses per al control del soroll en la seua fase de transmissió en la utilització de pantalles acústiques. L'aparició de nous materials formats per xarxes de dispersors acústics aïllats, denominats cristals de so, està revolucionant el camp de l'apantallament acústic, possibilitant l'avanç tecnològic d'esta àrea. Així, en els últims anys, les pantalles acústiques basades en cristals de so s'han posicionat com una alternativa viable a les pantalles acústiques tradicionals, ja que oferixen múltiples avantatges enfront de les solucions actuals. En el present treball es mostra primerament una recopilació dels avanços realitzats en el camp de l'apantallament acústic per mitjà d'esta tipologia de pantalles. No obstant això, encara hi ha línies d'investigació obertes en esta àrea, que és necessari abordar per a aconseguir l'objectiu d'aplicar esta tecnologia com a atenuadors de so en les infraestructures de transport. Durant el període de formació de la doctoranda, s'ha treballat en algunes de les línies d'investigació actives en este camp de l'apantallament acústic. Una d'estes investigacions va conduir al descobriment d'interferències entre els efectes de la ressonància i la dispersió múltiple dels cristals de so quan estos efectes es produïxen en rangs de freqüència pròxims. També hem dissenyat un nou dispositiu de reducció de soroll basat en cristals de so, utilitzant ferramentes d'optimització multiobjectiu, que permeten apantallar i reflectir de forma difusa el soroll. L'ús d'esta nova ferramenta de disseny va identificar la necessitat de realitzar un estudi comparatiu dels mètodes de simulació més utilitzats per a estimar el rendiment dels dispositius basats en cristals de so. Finalment, hem realitzat un estudi psicoacústic per a determinar la percepció de la reducció de molèstia que proporcionen les pantalles acústiques basades en cristals so i les barreres tradicionals, determinant si els paràmetres objectius que avaluen el seu rendiment coincidixen amb la resposta subjectiva dels usuaris.
[EN] Control of environmental noise is a major concern for advanced societies because of the resulting problems for citizens' health. One of the most widespread solutions for controlling noise in its transmission phase is the use of acoustic screens. The emergence of new materials made up of arrays of isolated acoustic scatterers, called sonic crystals, is revolutionizing the field of acoustic screening. In recent years, acoustic screens based on sonic crystals have positioned themselves as a viable alternative to traditional acoustic screens, as they offer multiple advantages over current traditional solutions. This Doctoral dissertation compiles the advances in the field of acoustic screening using this type of sonic crystals. However, there is still active research in this area which needs to be addressed and studied in order to apply this technology as noise reduction devices in transport infrastructures. Therefore, during the PhD student's training period, we have researched the acoustic phenomena produced by isolated scatterers in order to better understand the physical phenomena behind the lasts designs of this type of screen. One of these researches led to the discovery of interferences between the effects of resonance and multiple scattering of sonic crystals when occurring in nearby frequency ranges. Also we have designed a new noise reduction device based on sonic crystals, using multi-objective optimization tools, which would block and diffuse the noise. This new designing tool identified the need for a comparative study of the most commonly used simulation methods to estimate the performance of devices based on sonic crystals. Finally, we have carried out a psychoacoustic study that determined the perception of the annoyance reduction provided by acoustic screens based on sonic crystals and traditional barriers, determining whether the objective parameters that evaluate their performance match to the subjective response of the users.
Agradezco al Ministerio de Ciencia e Innovación por la ayuda concedida dentro del programa Doctores Industriales. Asimismo, a mi tutor en empresa Dr. Juan José Martín Pino, por posibilitar la realización de esta investigación dentro de la empresa BECSA. Al Departamento de Física Aplicada de la Universitat Politècnica de València, a la Comisión Académica del Programa de Doctorado de Matemáticas y al Centro de Tecnologías Físicas: Acústica, Materiales y Astrofísica.
Peiró Torres, MDP. (2021). Research of the acoustic phenomenon produced by isolated scatterers and its applicability as a noise reducing device in transport infrastructures. Search for an optimised and sustainable design [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/164903
TESIS

U radu se proučava poreklo funkcija, veza sistema povreatnih sprega sa uspostavljanjem funkcija, primarne funkcije kao funkcije fizičkog protoka između adaptivnih sistema i njihovog okruženja, perceptivno-analitičke funkcije kao funkcije informacionih protoka između adaptivnih sistema i okruženja, strukture obrazaca prirodnog i ljudskom rukom oblikovanih delova okruženja i sistemi ideja u oblikovanju okruženja.
Origins of functions, connections of feedback systems with emerging offunctions, primary functions as functions of physical flow between adaptivesystems and their environment, perceptual-analytical functions as functionsof information flow between adaptive systems and their environment,structures of patterns (levels of form) of natural and human designedelements of environment, historical developement and complexification ofrelationships of human soci-eties with their environment.

This thesis deals with image colorization and image super-resolution using neural networks. It briefly explains neural networks principles and summarizes current approaches in this domain. It also describes the design, implementation and training of various neural network architectures. The best implemented architecture can colorize images, in particular, works well with outdoor areas. The architecture for image super-resolution with residual blocks that was trained with a perceptual loss function performs a double increase in image resolution (4x more pixels in total). Part of this thesis is also an implementation of a web application that uses trained models for image modification.

This thesis presents a case of study of the development and performance analysis of a surface grading application with real-time compliance. We address the issue of spatial and temporal uniformity in the acquisition system. In a surface grading application it is crucial to ensure the uniform response of the system through time and space. All the results presented for surface grading were obtained using real data from the ceramic tile industry. The VxC TSG database is public and can be accessed at www.disca.upv.es/vision/vxctsg. We present a method based on soft colour-texture descriptors computed in perceptually uniform colour spaces. The method is parameterized and the involved factors are studied using two statistical procedures; experimental design and logistic regression. Although it is not a new theoretical contribution, we have found and demonstrate that a simple set of global colour and texture statistics, together with well-known classifiers, are powerful enough to meet stringent factory requirements for real-time and performance. Also the method is compared with two other approaches from the scientific literature; Colour Histograms and Centile-LBP. Finally, we explore the method's capacity for on-line inspection in a study of real-time compliance and parallelization.
López García, F. (2005). Real-time surface grading of ceramic tiles [Tesis doctoral no publicada]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/1909
Palancia

碩士
國立中興大學
電機工程學系所
107
In the paper, the adversarial learning based on the deep learning network architecture is proposed and the image prediction of human ageing is realized through this architecture. In the pre-existing data collection, we will prepare a photo data set for young people and the elderly (target age). The young people''s photo data set represents the input image (Domain X), and the elderly photo data set represents the output (target age) map. Like (Domain Y), our network can output images of young people of different ages by changing the image data sets of different seniors (target ages). In the framework of confrontational learning, there will be two kinds of learning networks, namely generators for generating fake images and discriminators for authenticating image realities. During the training process, our network must Compliance with the game theory, that is, the generator will get the feedback data of the discriminator to improve the quality of the generated image, and the discriminator must constantly identify the difference between the fake image and the true image. When the difference occurs, the discriminator will distinguish the true and false images with stricter standards. In the middle of the training, the generator becomes excellent, the discriminator will become excellent, and the generator and discriminator will reach the end. Equilibrium point, and the goal of the Nash Equilibrium point is that the generator can generate the same image as the target age, and the discriminator discriminates between the generated image and the target age image, and in terms of probability, the generator The probability of generating a picture similar to the target age will be close to 1, and the discriminator will have the same chance of identifying the picture and the target age picture, that is, the probability of identifying the generated picture and the target age is 0.5. In the generator part, we use the deep ResNet block. According to the experimental results, from 4 layers to 7 layers, the quality of the pictures produced by different layers is different. The results of the ageing show that the results of the 7 layers are the most excellent. Then there is the part of the discriminator. Since the generated picture is too similar to the (target age) picture, a certain degree of artifact is generated. In order to eliminate the artifact, we use a loss function called bias loss to reduce the discriminator. The degree of rigor, and the results have also improved significantly. In the aging process, we introduced the VGG16 pre-training network to take extra layer features for our Domain X and Domain Y image sets to make a loss function, and the resulting change is visible. In the photo collection, we took CACD2000 for face recognition and retrieval across ages. The dataset contains images of more than 160,000 2,000 celebrities between the ages of 16 and 62. To the best of our knowledge, it is by far the largest publicly available cross-age face dataset. Experimental results show that this method can achieve the most advanced performance on our dataset and other widely used cross-age face recognition dataset MORPH datasets. And we classify people of different ages, which are 10~20, 20~30, 30~40, 40~50, 50~60, 60 or above. For these 6 age groups, we use pictures from 10~20 years old. Our input is then taken into the image of the target age based on the age range we want to convert.

Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2008.
Source: Dissertation Abstracts International, Volume: 69-05, Section: B, page: 2885. Adviser: David M. Gooler. Includes bibliographical references (leaves 103-108) Available on microfilm from Pro Quest Information and Learning.

University of Technology Sydney. Faculty of Engineering and Information Technology.
A main goal of statistics and machine learning is to represent and manipulate high-dimensional probability distributions of real-world data, such as natural images. Generative adversarial networks (GAN), which are based on the adversarial learning paradigm, are one of the main types of methods for deriving generative models from complicated real-world data. GAN and its variants use a generator to synthesise semantic data from standard signal distributions and train a discriminator to distinguish real samples in the training dataset from fake samples synthesised by the generator. As a confronter, the generator aims to deceive the discriminator by producing ever more realistic samples. Through a two-player adversarial game played by the generator and discriminator, the generated distribution can approximate the real-world distribution and generate samples from it. This thesis aims to both improve the quality of generative modelling and manipulate generated samples by specifying multiple scene properties. A novel framework for training GAN is proposed to stabilise the training process and produce more realistic samples. Unlike existing GANs, which alternately train a generator and a discriminator using a pre-defined adversarial objective function, different adversarial training objectives are utilised as mutation operations and train a population of generators to adapt to the environment (i.e. the discriminator). The samples generated by different iterations of generators are evaluated and only well-performing generators are preserved and used for further training. In this way, the proposed framework overcomes the limitations of an individual adversarial training objective and always preserves the best offspring, contributing to the progress and success of GANs. Based on the GANs framework, this thesis devised a novel model, called a perceptual adversarial network (PAN). The proposed PAN consists of two feed-forward convolutional neural networks: a transformation network and a discriminative network. Besides generative adversarial loss, which is widely used in GANs, this thesis proposes to employ perceptual adversarial loss, which undergoes adversarial training between the transformation network and hidden layers of the discriminative network. The hidden layers and output of the discriminative network are upgraded to constantly and automatically discover discrepancies between a transformed image and the corresponding ground truth, and the image transformation network is trained to minimise the discrepancy identified by the discriminative network. Furthermore, to extend the generative models to perform more challenging re-rendering tasks, this thesis explores disentangled representations encoded in real-world samples and proposes a principled tag disentangled generative adversarial network for re-rendering new samples of the object of interest from a single image by specifying multiple scene properties. Specifically, from an input sample, a disentangling network extracts disentangled and interpretable representations, which are then used to generate new samples using the generative network. In order to improve the quality of the disentangled representations, a tag mapping net determines the consistency between the image and its tags. Finally, experiments with different challenging datasets and image synthesis tasks demonstrate the good performance of the proposed frameworks regarding the problem of interest.

碩士
國立臺灣大學
資訊工程學研究所
93
Block-based video coding cooperating with block transform and block motion compensation is the most widely adopted way to reduce the data redundancy in various video coding standards. Although the goal of de-correlations is achieved effectively by this way, the most annoying artifact known as the blocking effect also comes into existence. To both remove this artifact and improve the coding performance simultaneously, the latest video coding standard, H.264/AVC, enforces the deblocking filters inside its coding loop. In the design of deblocking filters of H.264/AVC, one pair of parameters, OffsetA and OffsetB, are provided, which allow the adaptive control of the deblocking strength in slice level. Thus, finding out better parameters for conducting the deblocking process of H.264/AVC is capable of improving visual quality of reconstructed video. Identifying which edges belong to blocking effect relies on perceptual judgment of human beings. In fact, this subjective assessment may not exactly match existing objective measurements and high PSNR does not always stand for less blocking artifacts. In this thesis, we introduce two new criteria for measuring the blocking distortion by analyzing the perceptual difference between the source and the reconstruction. The experimental results validate the proposed approaches, especially in subjective issues. On the other hand, another implicit advantage of deblocking is ignored by most encoders. It is observed that different coded images may have the same output after applying the mandatory deblocking process. Based on this observation, we integrate this concept into H.264/AVC. For eight different deblocking modes, we first derive the equations to change the input image but do not affect the final output reconstruction. By choosing those of less bitrate consumption, the proposed pre-processing approach successfully improves video coding performance. Combing advantages of both pre-process and post-process, an enhanced H.264/AVC coding system is implemented which maximizes the effect of deblocking filters. The experimental results demonstrate its improvements for H.264/AVC codec both in objective and subjective evaluations

Two amnesic patients with MTL damage that included the hippocampus and perirhinal cortex were tested along controls on a series of “oddity” discrimination tasks, in which they had to select an odd item from a visual array. Participants’ eye moments were monitored while they performed these tasks. Three types of stimuli were used: greebles, scenes, and faces. Results revealed that patients were impaired on tasks that required them to discriminate between items that shared features in common and tasks that required processing items from different viewpoints. An analysis of their eye movements revealed that their impaired performance was linked with decreased viewing times of target items compared to controls, when discriminating between greebles and scenes; their poor performance on the faces task could not be explained by the same token.