Segmentation is one of the first steps in human visual system which helps us see the world around us. Humans pre-attentively segment scenes into regions of unique textures in around 10-20 ms. In this thesis, we address the problem of segmentation by grouping dense pixel-wise descriptors. Our work is based on the fact that human vision has a feed forward and a feed backward loop, where low level feature are used to refine high level features in forward feed, and higher level feature information is used to refine the low level features in backward feed. Most vision algorithms are based on a feed-forward loop, where low-level features are used to construct and refine high level features, but they don’t have the feed back loop. We have introduced ”Shape-Tailored Local Descriptors”, where we use the high level feature information (region approximation) to update low level features i.e. the descriptor, and the low level feature information of the descriptor is used to update the segmentation regions. Our ”Shape-Tailored Local Descriptor” are dense local descriptors which are tailored to an arbitrarily shaped region, aggregating data only within the region of interest. Since the segmentation, i.e., the regions, are not known a-priori, we propose a joint problem for Shape-Tailored Local Descriptors and Segmentation (regions). Furthermore, since natural scenes consist of multiple objects, which may have different visual textures at different scales, we propose to use a multi-scale approach to segmentation. We have used a set of discrete scales, and a continuum of scales in our experiments, both resulted in state-of-the-art performance. Lastly we have looked into the nature of the features selected, we tried handcrafted color and gradient channels and we have also introduced an algorithm to incorporate learning optimal descriptors in segmentation approaches. In the final part of this thesis we have introduced techniques for unsupervised learning of descriptors for segmentation. This eliminates the problem of deep learning methods where we need huge amounts of training data to train the networks. The optimum descriptors are learned, without any training data, on the go during segmentation.

Research has typically addressed a specific emotional component of having a Learning Disability (LD), and thus has failed to capture the complete picture of what it is like to experience a LD. The current study asked university students to describe without any prompts or cues how it feels to have a LD, both retrospectively in high school and currently in university. We were interested in seeing how individuals with LDs describe their LD experience in their own words through free association. Information was collected from eight different cohorts throughout the past 11 years who were enrolled in a course on LDs for students diagnosed with LDs. All descriptors were coded into 17 different theme categories and further sorted by valence into positive, neutral, and negative categories. Participants reported more negative descriptors than positive ones, which interacted with the context in which they were reported. More negative descriptors were reported in high school compared to university and more positive descriptors were reported in university than high school. We failed to find any differences in emotional valence across the different cohorts. Latent class analyses revealed that reports of high school experiences consisted of two different LD profiles: extremely negative and negative. University experiences consisted of three different LD profiles: predominately positive, mixed emotional valence, and predominately negative. These results suggest that the experience of a LD can improve in university but that approximately 23% continue to find having LD a highly negative experience even though they are receiving support. / Social Sciences and Humanities Research Council: Joseph-Armand Bombardier CGS Master’s Scholarship

Submitted by Socorro Pontes (socorrop@ufersa.edu.br) on 2017-07-17T14:33:25Z No. of bitstreams: 1 LuizaCVM_DISSERT.pdf: 405995 bytes, checksum: a17548eebceffb6034ffb50efcf06189 (MD5) / Approved for entry into archive by Vanessa Christiane (referencia@ufersa.edu.br) on 2017-07-18T15:07:24Z (GMT) No. of bitstreams: 1 LuizaCVM_DISSERT.pdf: 405995 bytes, checksum: a17548eebceffb6034ffb50efcf06189 (MD5) / Approved for entry into archive by Vanessa Christiane (referencia@ufersa.edu.br) on 2017-07-18T15:07:42Z (GMT) No. of bitstreams: 1 LuizaCVM_DISSERT.pdf: 405995 bytes, checksum: a17548eebceffb6034ffb50efcf06189 (MD5) / Made available in DSpace on 2017-07-18T15:07:50Z (GMT). No. of bitstreams: 1 LuizaCVM_DISSERT.pdf: 405995 bytes, checksum: a17548eebceffb6034ffb50efcf06189 (MD5) Previous issue date: 2017-02-24 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Sweet potatoes are one of the most important food crops in the world. In the Northeast of Brazil, it is the main vegetable consumed and is a source of energy food. Because it is a tropical and subtropical climate plant, it is easy to grow and adaptable, presenting low production costs due to its rusticity. Brazil presents a diversity of types and forms of sweet potato, and part of its territory may be considered as a secondary center of variability of the species: hence the need to characterize and evaluate this culture. Therefore, the objective of this work was to evaluate the genetic divergence of sweet potato accesses belonging to the Germplasm Didactic Collection of the Federal Rural Semi-Arid University (UFERSA) using morphological descriptors. For the characterization, 21 accessions were used. The spatial arrangement in the field was in continuous rows (1.0 m between ridges and 0.3 m between pits, two branches per pits). We used 17 morphological descriptors: leaf color (mature and immature), general leaf profile, number of leaf lobes, leaf lobe types, central leaf lobe shape, pigmentation and petiole length, commercial root, commercial length and diameter of commercial roots, mature leaf size, film color. The hierarchical groupings of the accessions were obtained by UPGMA methods. The joint analysis of the qualitative and quantitative data to determine the genetic distance was accomplished with basis on the Gower algorithm. The relative contribution of the quantitative descriptors to the genetic divergence between the accessions was determined. In the generated dendrograms, there was distribution of the accesses in four distinct groups. The variables that contributed most to the genetic divergence among the accessions were commercial root (62.10%) and commercial root length (27.66%). Through morphological descriptors, it was possible to identify genetic divergence among the accesses of sweet potatoes belonging to the Didactic Collection of Germplasm of UFERSA / A batata-doce é uma das culturas alimentares mais importantes no mundo. No Nordeste do Brasil, é uma das principais hortaliças consumidas e constitui fonte de alimento energético. Por ser uma planta de clima tropical e subtropical, é de fácil cultivo e ampla adaptabilidade, apresentando baixo custo de produção devido à sua rusticidade. O Brasil apresenta diversidade de tipos e formas de batata-doce, e parte do seu território pode ser considerada como centro secundário de variabilidade da espécie, tornando-se importante conhecer a variabilidade que ocorre no germoplasma dessa cultura. Portanto, o objetivo deste trabalho foi avaliar a divergência genética dos acessos de batata-doce pertencentes à Coleção Didática de Germoplasma da Universidade Federal Rural do Semi-Árido (UFERSA) utilizando descritores morfológicos. Para a caracterização, foram utilizados 21 acessos. O arranjo espacial em campo foi em fileiras contínuas (1,0 m entre leiras e 0,3 m entre covas, duas ramas por covas). Foram utilizados 17 descritores morfológicos: cor da folha (madura e imatura), perfil geral da folha, número de lóbulos da folha, tipos de lóbulo da folha, forma do lóbulo central da folha, pigmentação e comprimento do pecíolo, produção de raízes comerciais e não comerciais, comprimento e diâmetro das raízes comerciais, tamanho da folha madura e cor da película. Os agrupamentos hierárquicos dos acessos foram obtidos pelos métodos de UPGMA. A análise conjunta dos dados qualitativos e quantitativos para determinação da distância genética foi determinada com base no algoritmo de Gower. Foi determinada a contribuição relativa dos descritores quantitativos para a divergência genética entre os acessos. Nos dendrogramas gerados, houve a distribuição dos acessos em quatro grupos distintos. As variáveis que mais contribuíram para a divergência genética entre os acessos foram a produção de raízes comerciais (62,10%) e comprimento das raízes comerciais (27,66%). Por meio dos descritores morfológicos, foi possível identificar divergência genética entre os acessos de batata-doce pertencentes à Coleção didática de Germoplasma da UFERSA / 2017-07-17

A textura da íris humana é uma das peculiaridades biométricas mais confiáveis, pois os padrões que compõem sua estrutura são considerados únicos e estáveis por longos anos. No entanto, amostras de íris capturadas em ambiente não cooperativo como reconhecimento de íris a distância, por exemplo, estão sujeitas a conter variações na textura, devido a mudanças comportamentais da membrana da íris. Outro problema é a complexidade do algoritmo, que o torna inviável para aplicações práticas ou em tempo real. O objetivo deste trabalho foi avaliar alguns descritores de textura locais para o reconhecimento biométrico de íris, considerando os efeitos de dilatação e contração da pupila. Para a comprovação da hipótese desta tese de doutoramento, foi utilizada uma base de dados contendo amostras de íris com a pupila contraída e dilatada, simulando assim, a aquisição natural em ambiente não cooperativo. Além disso, foram propostos dois novos descritores, denominados como Median Local Mapped Pattern (Median-LMP) e Modified Median Local Mapped Pattern (MM-LMP), que foram comparados com o método de Daugman, o Local Mapped Pattern (LMP), o Completed Modeling of Local Binary Pattern (CLBP), o Median Binary Pattern (MBP) e o Weber Law Descriptor (WLD). Os resultados da avaliação de desempenho mostraram que o algoritmo de Daugman é o melhor para o reconhecimento de íris quando é realizada a comparação entre amostras de íris com pupilas contraídas. No entanto, se a pupila está dilatada, os descritores propostos apresentaram o melhor desempenho, principalmente se uma amostra de íris com uma pupila contraída é comparada com outra íris com a pupila dilatada. Além disso, os descritores propostos e o LMP obtiveram os menores tempos de processamento, sendo mais adequados do que os demais para aplicações em tempo preditivo com implementação em hardware. / The texture of the human iris is one of the most reliable biometric traits, so the patterns that make up its structure are the only criteria and stable for long time. However, iris samples captured in a noncooperative environment as recognition of nature, for example, subject to contain variations in texture, due to behavioral changes of the iris membrane. Another problem is an algorithm complexity, which makes it impractical for practical or in real-time applications. The objective of this work is to evaluate some local texture descriptors for the biometric iris recognition, considering the effects of dilation and contraction of the pupil. In order to prove the hypothesis of this doctoral question, a database was used containing iris samples with a contracted and dilated pupil, thus simulating a natural acquisition in a noncooperative environment. In addition, two new descriptors, named Median-Local Standard Mapped (Median-LMP) and Modified Modified Local Standard Mapped (MM-LMP) were proposed, which were compared with the Daugman method, the Mapped Local Pattern (LMP), the Complete Local Binary Pattern Modeling (CLBP), the Median Binary Standard (MBP) and Weber Law Descriptor (WLD). The results of the performance evaluation show that the Daugman algorithm is the best for iris recognition when a study of iris samples with the students is performed. However, if a pupil is dilated, the proposed descriptors show the best performance, especially a sample of iris with a contracted pupil is compared to another iris with a dilated pupil. In addition, the proposed descriptors and the LMP obtained the shortest processing times, being more adequate than the others for predictive time applications with hardware implementation.

Objetivo: avaliar a percepção da dor na fibromialgia por meio de técnica metodológica quantitativa e qualitativa. Métodos: 30 clientes foram avaliadas por meio de uma entrevista analisada através de análise de conteúdo temática e do Instrumento de Descritores de Dor, sendo que para os dados resultantes foram calculados a média aritmética e o desvio padrão para determinar quais descritores caracterizam melhor a dor na fibromialgia. Resultados e discussão: a análise de conteúdo resultou na construção de categorias de análise referentes às percepções de: diagnóstico, motivações, doença, sentimentos, pensamentos e repercussões na qualidade de vida. Já o Instrumento de Descritores de Dor revelou que os descritores de maior atribuição na caracterização da dor foram incômoda, que espalha, latejante, desconfortável e persistente e os de menor atribuição foram desgraçada, demoníaca, maldita, aterrorizante e assustadora. Os dois instrumentos mostraram tendência das clientes em perceber e relatar a dor, principalmente relacionada às características sensorialdiscriminativas. Além disso apresentaram dados relativos à importância do papel da família e do profissional de saúde no manejo da dor. Conclusão: percebeu-se a necessidade de estimular a percepção e a expressão das clientes com relação à dor, abarcando sua multidimensionalidade e que, o manejo da dor deve ser realizado levando-se em conta a tríade equipe de saúde - cliente - família, face à complexidade do fenômeno. / Objective: Evaluate the perception of pain in the fibromyalgia through the quantitative and qualitative methodological technique. Method: A total of 30 clients were assessed through an interview analyzed by the thematic content and through the instrument Descriptors of Pain. Arithmetic mean and standard error were used to determine which descriptors better characterize the pain in the fibromyalgia. Results and Discussion: The result of the content analysis was the construction of categories of analysis regarding the perceptions of: diagnosis, motivation, disease, feelings, thoughts and repercussions on the quality of life. The Descriptors of Pain instrument revealed the descriptors of higher attribution in the characterization of pain were inconvenient, spreading, pulsating, uncomfortable and persistent and the descriptors with the lower attribution were miserable, demoniac, cursed, terrifying and frightening. The two instruments showed the clients\' tendency in perceiving and reporting the pain regarding to the sensorialdiscriminating characteristics. In addition, data related to the importance of the family\'s and the health professional\'s roles in managing the pain were presented. Conclusion: The need to stimulate the perception and expression of clients regarding the pain in its multidimensionality was perceived. It is concluded that the management of pain must be performed considering the complexity of the phenomenon in terms of the triad health team - client - family.

A textura da íris humana é uma das peculiaridades biométricas mais confiáveis, pois os padrões que compõem sua estrutura são considerados únicos e estáveis por longos anos. No entanto, amostras de íris capturadas em ambiente não cooperativo como reconhecimento de íris a distância, por exemplo, estão sujeitas a conter variações na textura, devido a mudanças comportamentais da membrana da íris. Outro problema é a complexidade do algoritmo, que o torna inviável para aplicações práticas ou em tempo real. O objetivo deste trabalho foi avaliar alguns descritores de textura locais para o reconhecimento biométrico de íris, considerando os efeitos de dilatação e contração da pupila. Para a comprovação da hipótese desta tese de doutoramento, foi utilizada uma base de dados contendo amostras de íris com a pupila contraída e dilatada, simulando assim, a aquisição natural em ambiente não cooperativo. Além disso, foram propostos dois novos descritores, denominados como Median Local Mapped Pattern (Median-LMP) e Modified Median Local Mapped Pattern (MM-LMP), que foram comparados com o método de Daugman, o Local Mapped Pattern (LMP), o Completed Modeling of Local Binary Pattern (CLBP), o Median Binary Pattern (MBP) e o Weber Law Descriptor (WLD). Os resultados da avaliação de desempenho mostraram que o algoritmo de Daugman é o melhor para o reconhecimento de íris quando é realizada a comparação entre amostras de íris com pupilas contraídas. No entanto, se a pupila está dilatada, os descritores propostos apresentaram o melhor desempenho, principalmente se uma amostra de íris com uma pupila contraída é comparada com outra íris com a pupila dilatada. Além disso, os descritores propostos e o LMP obtiveram os menores tempos de processamento, sendo mais adequados do que os demais para aplicações em tempo preditivo com implementação em hardware. / The texture of the human iris is one of the most reliable biometric traits, so the patterns that make up its structure are the only criteria and stable for long time. However, iris samples captured in a noncooperative environment as recognition of nature, for example, subject to contain variations in texture, due to behavioral changes of the iris membrane. Another problem is an algorithm complexity, which makes it impractical for practical or in real-time applications. The objective of this work is to evaluate some local texture descriptors for the biometric iris recognition, considering the effects of dilation and contraction of the pupil. In order to prove the hypothesis of this doctoral question, a database was used containing iris samples with a contracted and dilated pupil, thus simulating a natural acquisition in a noncooperative environment. In addition, two new descriptors, named Median-Local Standard Mapped (Median-LMP) and Modified Modified Local Standard Mapped (MM-LMP) were proposed, which were compared with the Daugman method, the Mapped Local Pattern (LMP), the Complete Local Binary Pattern Modeling (CLBP), the Median Binary Standard (MBP) and Weber Law Descriptor (WLD). The results of the performance evaluation show that the Daugman algorithm is the best for iris recognition when a study of iris samples with the students is performed. However, if a pupil is dilated, the proposed descriptors show the best performance, especially a sample of iris with a contracted pupil is compared to another iris with a dilated pupil. In addition, the proposed descriptors and the LMP obtained the shortest processing times, being more adequate than the others for predictive time applications with hardware implementation.

Process development of new complex reactions in the pharmaceutical and fine chemicals industries is challenging, and expensive. The field is beginning to see a bridging between fundamental first-principles investigations, and utilisation of data-driven statistical methods, such as machine learning. Nonetheless, process development and optimisation in these industries is mostly driven by trial-and-error, and experience. Approaches that move beyond these are limited to the well-developed optimisation of continuous variables, and often do not yield physical insights. This thesis describes several new methods developed to address research questions related to this challenge. First, we investigated whether utilising physical knowledge could aid statistics-guided self-optimisation of a C-H activation reaction, in which the optimisation variables were continuous. We then considered algorithmic treatment of the more challenging discrete variables, focussing on solvents. We parametrised a library of 459 solvents with physically meaningful molecular descriptors. Our case study was a homogeneous Rh-catalysed asymmetric hydrogenation to produce a chiral γ-lactam, with conversion and diastereoselectivity as objectives. We adapted a state-of-the-art multi-objective machine learning algorithm, based on Gaussian processes, to utilise the descriptors as inputs, and to create a surrogate model for each objective. The aim of the algorithm was to determine a set of Pareto solutions with a minimum experimental budget, whilst simultaneously addressing model uncertainty. We found that descriptors are a valuable tool for Design of Experiments, and can produce predictive and interpretable surrogate models. Subsequently, a physical investigation of this reaction led to the discovery of an efficient catalyst-ligand system, which we studied by operando NMR, and identified a parametrised kinetic model. Turning the focus then to ligands for asymmetric hydrogenation, we calculated versatile empirical descriptors based on the similarity of atomic environments, for 102 chiral ligands, to predict diastereoselectivity. Whilst the model fit was good, it failed to accurately predict the performance of an unseen ligand family, due to analogue bias. Physical knowledge has then guided the selection of symmetrised physico-chemical descriptors. This produced more accurate predictive models for diastereoselectivity, including for an unseen ligand family. The contribution of this thesis is a development of novel and effective workflows and methodologies for process development. These open the door for process chemists to save time and resources, freeing them up from routine work, to focus instead on creatively designing new chemistry for future real-world applications.

En el camp de l'Anàlisi de Documents voldríem ser capaços de processar automàticament qualsevol tipus de document digital i d'extreure la informació rellevant. és a dir, voldríem conËixer la configuració del document, identificar cadascuna de les seves parts i reconËixer els seus continguts; per a poder fer cerques entre les components del document, però també, per fer cerques entre documents diferents. Aquest és un problema difícil que ha motivat diferents línies de recerca a diferents nivells. S'ha desenvolupat tot una sèrie de tècniques destinades a pre-processar la imatge per augmentar la seva qualitat, reduint el soroll dels sistemes d'adquisició i minimitzant els efectes de la degradació dels documents. També trobem molts treballs en la segmentació destinats a separar les àrees d'interès de la resta del document. Finalment, des de finals dels anys 60 fins a l'actualitat s'han proposat molts tipus descriptors que pretenen representar i identificar aquestes àrees d'interès.En aquesta tesis ens hem centrat en el darrer d'aquests problemes, la descripció de formes però també en la fusió de classificadors per a aplicar-los a una de les apliacions de l'Anàlisi de Documents, el reconeixement de símbols gràfics. En el reconeixement de formes, moltes aplicacions han de fer front al problema de descriure un conjunt gran i complex de formes per a reconèixer-les, o per a recuperar-les de gran bases de dades. En alguns casos, a més del gran nombre de formes, podem trobar altres dificultats com són la semblança entre formes o la variabilitat de classes de símbols. En aquest casos, un punt clau en el procés de reconeixement de formes és la definició de descriptors de gran capacitat de discriminació. Malauradament, un sol tipus de descriptors no sol ser suficient per aconseguir resultats satisfactoris i per tant, hem de combinar la informació provinent de diferents fonts per a millorar el comportament global del sistema de reconeixement. Aquesta combinació de la informació la hem realitzat a travÈs de la fusió de classificadors.En relació a la descripció de formes, tradicionalment els símbols gràfics s'han representat mitjançant descriptors estructurals, construïts a partir d'una representació vectorial. Els mètodes de vectorització són sensibles al soroll i a les distorsions dels símbols esboçats. Podem intentar evitar aquest problema definint gramàtiques o construint models deformables dels símbols. Una altra possibilitat, la que hem seguit en aquest treball, és fer servir descriptors que no necessiten d'una representació vectorial. En el context de la descripció de formes hem proposat un descriptor basat en la transformada de crestetes -en anglès "ridgelets"- que, gràcies a que hem unificat la terminologia i hem introduït un vocabulari per explicar i classificar els descriptors, podem definir com: multiresolució, polar, 2D, que conserva la informació i invariant a les similituds. D'altre banda, la propietat de multiresolució de la transformada de crestetes fa que obtinguem una representació en diferents nivells de resolució que ens permet dividir-la en grups de coeficients de crestetes que es poden considerar com a descriptors. D'aquesta manera, hem entrenat un classificador per a cada descriptor, i hem proposat unes regles de combinació lineals, IN i DN, que minimitzen l'error de classificació per aquells classificadors que compleixin un conjunt de restriccions, relatives a la distribució i dependËncia dels classificadors.Aquests enfocs teòrics han estat avaluats a partir d'un conjunt d'experiments que ens han donat els següents resultats: Els descriptors de crestetes descriuen millor els símbols que altres descriptors més genèrics. Els mètodes IN i DN redueixen l'error de classificació en relació a d'altres mètodes de referència. Per últim, el mètode IN aplicat als descriptors de crestetes, en combinació amb classificadors de tipus "boosting" aconsegueix uns encerts de reconeixement propers als 100% en les proves definides per a la base de dades de símbols gràfics del GREC'03. / In the field of Document Analysis we would like to be able to automatically process any kind of digital document. We mean extracting the document layout and identifying each of its parts, recognising its contents and organising them in order to make searches of its components, through the document itself, but also through different documents. This is a challenger problem that has motivated different lines of research in the field of Document Analysis at different levels: Pre-processing techniques have been developed to upgrade the quality of the document image, reducing noise from the input devices and minimizing the effects of the degradation of documents. A deep study in segmentation has been carried out in order to separate the regions of interest from the document background. Finally, many descriptors have been proposed for representing and identifying these regions of interest since the end of 60s until now.In this thesis, we have focused on, this last problem, the shape description description and also on classifier fusion, to apply them to one of the application fields in the Document Analysis: the graphics recognition. In shape recognition, many applications have to face the problem of describing a large number of complex shapes for recognition or retrieval in large databases. Besides the large number of shapes, we can find other challenges for shape description, such as the similarity among some of the shapes or the variability of the shape classes. In these cases, one of the key issues is the design of highly discriminant shape descriptors. Unfortunately, one kind of descriptor is not usually enough to achieve satisfactory results and hence, we have to combine the information from different sources to improve the global performance of the recognition system. We have carried out this combination of information using classifier fusion. Concerning shape description, traditionally graphics have been represented using structural descriptors, which are based on a vectorial representation of the shape. Vectorization is quite sensitive to noise and to distortions of sketched symbols. We can try to overcome this problem using grammar descriptors or deformable models of shapes. Another possibility, which is the followed in this dissertation, is to propose descriptors that do not need a vectorial representation of the symbol. Thereby, in the context of shape description, we have proposed a descriptor based on the ridgelets transform which, thanks to we have unified the terminology used in shape description and the introduced vocabulary, we can define as: 2D, polar and multi-resolution descriptor information preserving and invariant to similarities. On the other hand, although ridgelets descriptor can be considered as a single descriptor, it offers a shape representation divided into groups of coefficients, which permit us to consider them as single descriptors. Thus, for each descriptor, we have trained a classifier and we have proposed two linear combination rules, IN and DN, that minimize the classification error of classifiers verifying a set of constraints concerning the dependence and the distribtuion of classifers.These theoretical approaches have been evaluated through an experimental evaluation in ridgelets descriptors, classifier fusion and applying the classifier fusion methods to ridge lets descriptors, obtaining the following results: Ridgelets descriptors have proven to represent graphics symbols better than general purpose descriptors. IN and DN methods reduce the misclassification rates regarding other reference fusion methods. Finally, the IN method applied to ridgelets descriptor, in combination of boosting algorithms, has reached recognition rates near to 100% in the test defined for the GREC'03 database.

Content-based image retrieval (CBIR) is still in its early stages, although several attempts have been made to solve or minimize challenges associated with it. CBIR techniques use such visual contents as color, texture, and shape to represent and index images. Of these, shapes contain richer information than color or texture. However, retrieval based on shape contents remains more difficult than that based on color or texture due to the diversity of shapes and the natural occurrence of shape transformations such as deformation, scaling and orientation. This thesis presents an approach for fusing several shape-based image retrieval techniques for the purpose of achieving reliable and accurate retrieval performance. An extensive investigation of notable existing shape descriptors is reported. Two new shape descriptors have been proposed as means to overcome limitations of current shape descriptors. The first descriptor is based on a novel shape signature that includes corner information in order to enhance the performance of shape retrieval techniques that use Fourier descriptors. The second descriptor is based on the curvature of the shape contour. This invariant descriptor takes an unconventional view of the curvature-scale-space map of a contour by treating it as a 2-D binary image. The descriptor is then derived from the 2-D Fourier transform of the 2-D binary image. This technique allows the descriptor to capture the detailed dynamics of the curvature of the shape and enhances the efficiency of the shape-matching process. Several experiments have been conducted in order to compare the proposed descriptors with several notable descriptors. The new descriptors not only speed up the online matching process, but also lead to improved retrieval accuracy. The complexity and variety of the content of real images make it impossible for a particular choice of descriptor to be effective for all types of images. Therefore, a data- fusion formulation based on a team consensus approach is proposed as a means of achieving high accuracy performance. In this approach a select set of retrieval techniques form a team. Members of the team exchange information so as to complement each other’s assessment of a database image candidate as a match to query images. Several experiments have been conducted based on the MPEG-7 contour-shape databases; the results demonstrate that the performance of the proposed fusion scheme is superior to that achieved by any technique individually.

Content-based image retrieval (CBIR) is still in its early stages, although several attempts have been made to solve or minimize challenges associated with it. CBIR techniques use such visual contents as color, texture, and shape to represent and index images. Of these, shapes contain richer information than color or texture. However, retrieval based on shape contents remains more difficult than that based on color or texture due to the diversity of shapes and the natural occurrence of shape transformations such as deformation, scaling and orientation. This thesis presents an approach for fusing several shape-based image retrieval techniques for the purpose of achieving reliable and accurate retrieval performance. An extensive investigation of notable existing shape descriptors is reported. Two new shape descriptors have been proposed as means to overcome limitations of current shape descriptors. The first descriptor is based on a novel shape signature that includes corner information in order to enhance the performance of shape retrieval techniques that use Fourier descriptors. The second descriptor is based on the curvature of the shape contour. This invariant descriptor takes an unconventional view of the curvature-scale-space map of a contour by treating it as a 2-D binary image. The descriptor is then derived from the 2-D Fourier transform of the 2-D binary image. This technique allows the descriptor to capture the detailed dynamics of the curvature of the shape and enhances the efficiency of the shape-matching process. Several experiments have been conducted in order to compare the proposed descriptors with several notable descriptors. The new descriptors not only speed up the online matching process, but also lead to improved retrieval accuracy. The complexity and variety of the content of real images make it impossible for a particular choice of descriptor to be effective for all types of images. Therefore, a data- fusion formulation based on a team consensus approach is proposed as a means of achieving high accuracy performance. In this approach a select set of retrieval techniques form a team. Members of the team exchange information so as to complement each other’s assessment of a database image candidate as a match to query images. Several experiments have been conducted based on the MPEG-7 contour-shape databases; the results demonstrate that the performance of the proposed fusion scheme is superior to that achieved by any technique individually.