Automatic verification of personal identity using facial images is the central topic of the thesis. This problem can be stated as follows. Given two face images, it must be determined automatically whether they are images of the same person or of different persons. Due to many factors such as variability of facial appearance, sensitivity to noise, template aging, etc., the problem is difficult. We can overcome some of these difficulties by combining different information sources for the classification/recognition task. In this thesis we propose strategies on how to combine the different information sources, i.e. fusion strategies, in order to improve the verification accuracy. We have designed and thoroughly optimised a number of face verification algorithms. Their individual properties such as how their accuracy depends on algorithm parameters, image size, or sensitivity to mis-registrations have been studied. We have also studied how to combine the outputs of the different algorithms in order to reduce the verification error rates. Another decision fusion aspect considered in this thesis is the fusion of confidences obtained sequentially on several video frames of the same person's face. Finally multimodal fusion has been studied. In this case, the speech and face of the same subject are recorded and processed by different algorithms which output separate opinions. These two opinions are then conciliated at the fusion stage. It is shown that in all cases, information fusion allows a considerable performance improvement if the fusion stage is carefully designed.

Made available in DSpace on 2014-12-17T15:47:49Z (GMT). No. of bitstreams: 1 DiogoFO.pdf: 866073 bytes, checksum: bf59c2597aef9b7382b7e14bd4914265 (MD5) Previous issue date: 2008-09-01 / Conselho Nacional de Desenvolvimento Cient?fico e Tecnol?gico / Multi-classifier systems, also known as ensembles, have been widely used to solve several problems, because they, often, present better performance than the individual classifiers that form these systems. But, in order to do so, it s necessary that the base classifiers to be as accurate as diverse among themselves this is also known as diversity/accuracy dilemma. Given its importance, some works have investigate the ensembles behavior in context of this dilemma. However, the majority of them address homogenous ensemble, i.e., ensembles composed only of the same type of classifiers. Thus, motivated by this limitation, this thesis, using genetic algorithms, performs a detailed study on the dilemma diversity/accuracy for heterogeneous ensembles / Sistemas Multiclassificadores, tamb?m conhecidos como comit?s de classificadores, t?m sido amplamente utilizados para resolver os mais variados problemas, pois em geral t?m melhores desempenhos que os classificadores base que formam esses sistemas. Para que isso ocorra, por?m, ? necess?rio que os classificadores base sejam t?o acurados quanto diversos entre si isso ? conhecido como dilema da diversidade-acur?cia. Dado a sua import?ncia, alguns trabalhos sobre o estudo do omportamento dos comit?s no contexto desse dilema foram propostos. Entretanto, a maioria dos trabalhos estudou tal problema para comit?s homog?neos, ou seja, comit?s formados apenas por classificadores do mesmo tipo. Sendo assim, motivado por esta limita??o, esta disserta??o, usando algoritmos gen?ticos, efetua um estudo mais detalhado sobre o dilema da diversidade-acur?cia em comit?s heterog?neos

In a wide range of applications, decisions must be made by combining information from multiple agents with varying levels of trust and expertise. For example, citizen science involves large numbers of human volunteers with differing skills, while disaster management requires aggregating information from multiple people and devices to make timely decisions. This thesis introduces efficient and scalable Bayesian inference for decision combination, allowing us to fuse the responses of multiple agents in large, real-world problems and account for the agents’ unreliability in a principled manner. As the behaviour of individual agents can change significantly, for example if agents move in a physical space or learn to perform an analysis task, this work proposes a novel combination method that accounts for these time variations in a fully Bayesian manner using a dynamic generalised linear model. This approach can also be used to augment agents’ responses with continuous feature data, thus permitting decision-making when agents’ responses are in limited supply. Working with information inferred using the proposed Bayesian techniques, an information-theoretic approach is developed for choosing optimal pairs of tasks and agents. This approach is demonstrated by an algorithm that maintains a trustworthy pool of workers and enables efficient learning by selecting informative tasks. The novel methods developed here are compared theoretically and empirically to a range of existing decision combination methods, using both simulated and real data. The results show that the methodology proposed in this thesis improves accuracy and computational efficiency over alternative approaches, and allows for insights to be determined into the behavioural groupings of agents.

Made available in DSpace on 2014-12-17T15:47:44Z (GMT). No. of bitstreams: 1 LauraEASS.pdf: 1648653 bytes, checksum: 0aa1d6a5cd26175688d09f2c09459503 (MD5) Previous issue date: 2008-02-01 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / In systems that combine the outputs of classification methods (combination systems), such as ensembles and multi-agent systems, one of the main constraints is that the base components (classifiers or agents) should be diverse among themselves. In other words, there is clearly no accuracy gain in a system that is composed of a set of identical base components. One way of increasing diversity is through the use of feature selection or data distribution methods in combination systems. In this work, an investigation of the impact of using data distribution methods among the components of combination systems will be performed. In this investigation, different methods of data distribution will be used and an analysis of the combination systems, using several different configurations, will be performed. As a result of this analysis, it is aimed to detect which combination systems are more suitable to use feature distribution among the components / Em sistemas que combinam as sa?das de classificadores de padr?es, sistemas de combina??o, como comit?s e sistemas multiagentes para classifica??o, um dos principais problemas ? que os componentes do sistema (classificadores ou agentes) devem ser diversos entre si. Em outras palavras, n?o existe ganho de desempenho em sistemas formados por um conjunto de componentes id?nticos. Um modo de aumentar a diversidade do sistema ? distribuir os dados do padr?o entre os classificadores que comp?em o sistema. Neste trabalho ser? feita uma investiga??o sobre o impacto do uso de t?cnicas de distribui??o de dados, mais especificamente distribui??o de caracter?sticas, entre os componentes de sistemas de combina??o de classificadores. Nesta investiga??o, diferentes t?cnicas de distribui??o de caracter?sticas ser?o usadas e uma an?lise comparativa entre diferentes sistemas de combina??o, usando diferentes configura??es, ser? feita. Como resultado desta an?lise, espera-se detectar que sistemas de combina??o s?o mais adequados para usar distribui??o de caracter?sticas entre os componentes

Made available in DSpace on 2014-12-17T15:48:03Z (GMT). No. of bitstreams: 1 AntonioAFN_DISSERT.pdf: 3187796 bytes, checksum: c8d44014d0b75e991f4f3b3473a8dcd5 (MD5) Previous issue date: 2012-08-24 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / Committees of classifiers may be used to improve the accuracy of classification systems, in other words, different classifiers used to solve the same problem can be combined for creating a system of greater accuracy, called committees of classifiers. To that this to succeed is necessary that the classifiers make mistakes on different objects of the problem so that the errors of a classifier are ignored by the others correct classifiers when applying the method of combination of the committee. The characteristic of classifiers of err on different objects is called diversity. However, most measures of diversity could not describe this importance. Recently, were proposed two measures of the diversity (good and bad diversity) with the aim of helping to generate more accurate committees. This paper performs an experimental analysis of these measures applied directly on the building of the committees of classifiers. The method of construction adopted is modeled as a search problem by the set of characteristics of the databases of the problem and the best set of committee members in order to find the committee of classifiers to produce the most accurate classification. This problem is solved by metaheuristic optimization techniques, in their mono and multi-objective versions. Analyzes are performed to verify if use or add the measures of good diversity and bad diversity in the optimization objectives creates more accurate committees. Thus, the contribution of this study is to determine whether the measures of good diversity and bad diversity can be used in mono-objective and multi-objective optimization techniques as optimization objectives for building committees of classifiers more accurate than those built by the same process, but using only the accuracy classification as objective of optimization / Comit?s de classificadores podem ser empregados para melhorar a acur?cia de sistemas de classifica??o, ou seja, diferentes classificadores aplicados ? solu??o de um mesmo problema podem ser combinados gerando um sistema de maior acur?cia, denominado de comit?s de classificadores. Para que se obtenha sucesso ? necess?rio que os classificadores apresentem erros em diferentes objetos do problema para que assim os erros de um classificador sejam suprimidos pelo acerto dos demais na aplica??o do m?todo de combina??o do comit?. A caracter?stica dos classificadores de errarem em objetos diferentes ? denominada de diversidade. No entanto, as maiorias das medidas de diversidade n?o conseguiam descrever essa import?ncia. Recentemente, foram propostas duas medidas de diversidade (boa e m? diversidade) as medidas de boa e m? diversidade com o objetivo de auxiliar a gera??o de comit?s mais acurados. Este trabalho efetua uma an?lise experimental dessas medidas aplicadas diretamente na constru??o de comit?s de classificadores. O m?todo de constru??o adotado ? modelado como um problema de busca pelo melhor conjunto de caracter?sticas das bases de dados do problema e pelo melhor conjunto de membros do comit? a fim de encontrar o comit? de classificadores que apresente ? maior acur?cia de classifica??o. Esse problema ? resolvido atrav?s de t?cnicas de otimiza??o metaheur?sticas, nas vers?es mono e multiobjetivo. S?o efetuadas an?lises estat?sticas para verificar se usar ou adicionar as medidas de boa e m? diversidade como objetivos de otimiza??o resulte comit?s mais acurados. Assim, a contribui??o desse trabalho ? determinar se as medidas de boa e m? diversidade podem ser utilizadas em t?cnicas de otimiza??o mono e multiobjetivo como objetivos de otimiza??o para constru??o de comit?s de classificadores mais acurados que aqueles constru?dos pelo mesmo processo, por?m utilizando somente a acur?cia de classifica??o como objetivo de otimiza??o