Face detection in video sequence is becoming popular in surveillance applications. The trade-off between obtaining discriminative features to achieve accurate detection versus computational overhead of extracting these features, which affects the classification speed, is a persistent problem. Two ideas are introduced to increase the features’ discriminative power. These ideas are used to implement two frontal face detectors examined on a 2D low-resolution surveillance sequence. First contribution is the parallel classifier. High discriminative power features are achieved by fusing the decision from two different features trained classifiers where each type of the features targets different image structure. Accurate and fast to train classifier is achieved. Co-occurrence of Local Binary Patterns (CoLBP) features is proposed, the pixels of the image are targeted. CoLBP features find the joint probability of multiple LBP features. These features have computationally efficient feature extraction and provide high discriminative features; hence, accurate detection is achieved.

This thesis examines the application of a Support Vector Machine (SVM) classifier to automatically detect sleep and quiet wake (rest) behavior in mice from pressure signals on their cage floor. Previous work employed Neural Networks (NN) and Linear Discriminant Analysis (LDA) to successfully detect sleep and wake behaviors in mice. Although the LDA was successful in distinguishing between the sleep and wake behaviors, it has several limitations, which include the need to select a threshold and difficulty separating additional behaviors with subtle differences, such as sleep and rest. The SVM has advantages in that it offers greater degrees of freedom than the LDA for working with complex data sets. In addition, the SVM has direct methods to limit overfitting for the training sets (unlike the NN method). This thesis develops an SVM classifier to characterize the linearly non separable sleep and rest behaviors using a variety of features extracted from the power spectrum, autocorrelation function, and generalized spectrum (autocorrelation of complex spectrum). A genetic algorithm (GA) optimizes the SVM parameters and determines a combination of 5 best features. Experimental results from over 9 hours of data scored by human observation indicate 75% classification accuracy for SVM compared to 68% accuracy for LDA.

Background Subtraction is one of the fundamental pre-processing steps in video processing. It helps to distinguish between foreground and background for any given image and thus has numerous applications including security, privacy, surveillance and traffic monitoring to name a few. Unfortunately, no single algorithm exists that can handle various challenges associated with background subtraction such as illumination changes, dynamic background, camera jitter etc. In this work, we propose a Multiple Background Model based Background Subtraction (MB2S) system, which is universal in nature and is robust against real life challenges associated with background subtraction. It creates multiple background models of the scene followed by both pixel and frame based binary classification on both RGB and YCbCr color spaces. The masks generated after processing these input images are then combined in a framework to classify background and foreground pixels. Comprehensive evaluation of proposed approach on publicly available test sequences show superiority of our system over other state-of-the-art algorithms.

The performance of a classifier is affected by a number of factors including classifier type, the input features and the desired output. This thesis examines the impact of feature selection and classification problem division on classification accuracy and complexity. Proper feature selection can reduce classifier size and improve classifier performance by minimizing the impact of noisy, redundant and correlated features. Noisy features can cause false association between the features and the classifier output. Redundant and correlated features increase classifier complexity without adding additional information. Output selection or classification problem division describes the division of a large classification problem into a set of smaller problems. Problem division can improve accuracy by allocating more resources to more difficult class divisions and enabling the use of more specific feature sets for each sub-problem. The first part of this thesis presents two methods for creating feature-selected hierarchical classifiers. The feature-selected hierarchical classification method jointly optimizes the features and classification tree-design using genetic algorithms. The multi-modal binary tree (MBT) method performs the class division and feature selection sequentially and tolerates misclassifications in the higher nodes of the tree. This yields a piecewise separation for classes that cannot be fully separated with a single classifier. Experiments show that the accuracy of MBT is comparable to other multi-class extensions, but with lower test time. Furthermore, the accuracy of MBT is significantly higher on multi-modal data sets. The second part of this thesis focuses on input feature selection measures. A number of filter-based feature subset evaluation measures are evaluated with the goal of assessing their performance with respect to specific classifiers. Although there are many feature selection measures proposed in literature, it is unclear which feature selection measures are appropriate for use with different classifiers. Sixteen common filter-based measures are tested on 20 real and 20 artificial data sets, which are designed to probe for specific feature selection challenges. The strengths and weaknesses of each measure are discussed with respect to the specific feature selection challenges in the artificial data sets, correlation with classifier accuracy and their ability to identify known informative features. The results indicate that the best filter measure is classifier-specific. K-nearest neighbours classifiers work well with subset-based RELIEF, correlation feature selection or conditional mutual information maximization, whereas Fisher's interclass separability criterion and conditional mutual information maximization work better for support vector machines. Based on the results of the feature selection experiments, two new filter-based measures are proposed based on conditional mutual information maximization, which performs well but cannot identify dependent features in a set and does not include a check for correlated features. Both new measures explicitly check for dependent features and the second measure also includes a term to discount correlated features. Both measures correctly identify known informative features in the artificial data sets and correlate well with classifier accuracy. The final part of this thesis examines the use of feature selection for time-series data by using feature selection to determine important individual time windows or key frames in the series. Time-series feature selection is used with the MBT algorithm to create classification trees for time-series data. The feature selected MBT algorithm is tested on two human motion recognition tasks: full-body human motion recognition from joint angle data and hand gesture recognition from electromyography data. Results indicate that the feature selected MBT is able to achieve high classification accuracy on the time-series data while maintaining a short test time.

Face detection in video sequence is becoming popular in surveillance applications. The trade-off between obtaining discriminative features to achieve accurate detection versus computational overhead of extracting these features, which affects the classification speed, is a persistent problem. Two ideas are introduced to increase the features’ discriminative power. These ideas are used to implement two frontal face detectors examined on a 2D low-resolution surveillance sequence. First contribution is the parallel classifier. High discriminative power features are achieved by fusing the decision from two different features trained classifiers where each type of the features targets different image structure. Accurate and fast to train classifier is achieved. Co-occurrence of Local Binary Patterns (CoLBP) features is proposed, the pixels of the image are targeted. CoLBP features find the joint probability of multiple LBP features. These features have computationally efficient feature extraction and provide high discriminative features; hence, accurate detection is achieved.

As more and more remotely sensed data becomes available it is becoming increasingly harder to analyse it with the more traditional labour intensive, manual methods. The commonly used techniques, that involve expert evaluation, are widely acknowledged as providing inconsistent results, at best. We need more general techniques that can adapt to a given situation and that incorporate the strengths of the traditional methods, human operators and new technologies. The difficulty in interpreting remotely sensed data is that often only a small amount of data is available for classification. It can be noisy, incomplete or contain irrelevant information. Given that the training data may be limited we demonstrate a variety of techniques for highlighting information in the available data and how to select the most relevant information for a given classification task. We show that more consistent results between the training data and an entire image can be obtained, and how misclassification errors can be reduced. Specifically, a new technique for attribute selection in neural networks is demonstrated. Machine learning techniques, in particular, provide us with a means of automating classification using training data from a variety of data sources, including remotely sensed data and expert knowledge. A classification framework is presented in this thesis that can be used with any classifier and any available data. While this was developed in the context of vegetation mapping from remotely sensed data using machine learning classifiers, it is a general technique that can be applied to any domain. The emphasis of the applicability for this framework being domains that have inadequate training data available.

This paper evaluates the classification accuracy of three neural network classifiers on a satellite image-based pattern classification problem. The neural network classifiers used include two types of the Multi-Layer-Perceptron (MLP) and the Radial Basis Function Network. A normal (conventional) classifier is used as a benchmark to evaluate the performance of neural network classifiers. The satellite image consists of 2,460 pixels selected from a section (270 x 360) of a Landsat-5 TM scene from the city of Vienna and its northern surroundings. In addition to evaluation of classification accuracy, the neural classifiers are analysed for generalization capability and stability of results. Best overall results (in terms of accuracy and convergence time) are provided by the MLP-1 classifier with weight elimination. It has a small number of parameters and requires no problem-specific system of initial weight values. Its in-sample classification error is 7.87% and its out-of-sample classification error is 10.24% for the problem at hand. Four classes of simulations serve to illustrate the properties of the classifier in general and the stability of the result with respect to control parameters, and on the training time, the gradient descent control term, initial parameter conditions, and different training and testing sets. (authors' abstract) / Series: Discussion Papers of the Institute for Economic Geography and GIScience

Objects and parts are crucial elements for achieving automatic image understanding. The goal of the object detection task is to recognize and localize all the objects in an image. Similarly, semantic part detection attempts to recognize and localize the object parts. This thesis proposes four contributions. The first two make object detection more efficient by using active search strategies guided by image context. The last two involve parts. One of them explores the emergence of parts in neural networks trained for object detection, whereas the other improves on part detection by adding object context. First, we present an active search strategy for efficient object class detection. Modern object detectors evaluate a large set of windows using a window classifier. Instead, our search sequentially chooses what window to evaluate next based on all the information gathered before. This results in a significant reduction on the number of necessary window evaluations to detect the objects in the image. We guide our search strategy using image context and the score of the classifier. In our second contribution, we extend this active search to jointly detect pairs of object classes that appear close in the image, exploiting the valuable information that one class can provide about the location of the other. This leads to an even further reduction on the number of necessary evaluations for the smaller, more challenging classes. In the third contribution of this thesis, we study whether semantic parts emerge in Convolutional Neural Networks trained for different visual recognition tasks, especially object detection. We perform two quantitative analyses that provide a deeper understanding of their internal representation by investigating the responses of the network filters. Moreover, we explore several connections between discriminative power and semantics, which provides further insights on the role of semantic parts in the network. Finally, the last contribution is a part detection approach that exploits object context. We complement part appearance with the object appearance, its class, and the expected relative location of the parts inside it. We significantly outperform approaches that use part appearance alone in this challenging task.

Resumo: A citricultura é importante para a economia brasileira devido aos recordes de exportações e pela geração de empregos direto e indireto. O sensoriamento remoto é a fonte primária de informações sobre a cobertura do solo em decorrência dos avanços tecnológicos nos sistemas sensores. A incorporação de procedimentos geoestatísticos em estudos ambientais baseado em técnicas de Krigagem tem sido utilizada por profissionais da área de sensoriamento remoto. O objetivo deste trabalho foi comparar a precisão de métodos de classificação de imagens orbitais na determinação de áreas cultivadas com citros, na Mesoregião de Bauru e Araraquara, através de métodos de sensoriamento remoto e geoestatísticos, para a discriminação e quantificação da área plantada. A área de estudo utilizada nesta pesquisa esta localizada na região central do Estado de São Paulo, com uma área de 56.146,78 ha. Foi utilizado o SIG-SPRING para o processamento dos dados. Foram utilizadas as bandas 2, 3 e 4 da imagem digital proveniente do satélite CBERS 2B, câmera CCD (Câmera Imageadora de Alta Resolução) que fornece imagens com uma resolução espacial de 20 metros. A imagem foi de 16/04/2009, nas órbitas/ponto 157/124 e 157/125. No processo de classificação das imagens foram estudados três métodos de classificação, sendo o CLUSTER não-supervisionado, o MAXVER (Máxima Verossimilhança) e a KI (Krigagem Indicativa) supervisionados, além da classificação em tela tida como verdade terrestre. As fidedignidades das classificações foram avaliadas pelo índice Kappa. Pelos resultados obtidos nas classificações de imagem para a discriminação e quantificação de áreas cultivadas com citros pode-se concluir: os classificadores obtiveram melhor qualidade de classificação para as áreas maiores com CITROS;a Krigagem Indicativa unificou as áreas pequenas de CITROS em uma única área... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The citrus industry is important for the Brazilian economy due to record exports and generating direct and indirect jobs. Remote sensing is the primary source of information on land cover as a result of technological advances in sensor systems. The incorporation of geostatistics procedures in environmental studies based on Kriging techniques has been used by professionals in areas of remote sensing. The aim of this study was to compare the accuracy of classification methods in the determination of satellite images of areas cultivated with citrus, Mesoregião in Araraquara, Bauru, and, through methods of remote sensing and geostatistics to discrimination and quantification of the area planted. The study area used in this research is located in the central region of São Paulo, with an area of 56,146.78 ha. It was used the GIS-SPRING for data processing. We used bands 2, 3 and 4 of the digital image from the satellite CBERS 2B, CCD camera (Camera high resolution image) that provides images with a spatial resolution of 20 meters. The image dated of 04/16/2009, the path/row 157/124 and 157/125. In the process of image classification were studied three methods of classification, and unsupervised clustering, the MLC (Maximum Likelihood) and KI (kriging) supervised classification beyond the screen taken as ground truth. The validation of classifications were evaluated by Kappa index. The results obtained of image classification onto discrimination and quantification of citrus areas were: the classifiers had improved quality of classification for wide citrus areas, the Indicative Kriging unified the small citrus areas in a single area, the Indicative Kriging was the classifier that less classified riparian vegetation area such as citrus, the others changed riparian vegetation by citrus. The validation showed that Indicative Kriging was the classifier which had the lower quality rating than... (Complete abstract click electronic access below) / Orientador: Célia Regina Lopes Zimback / Coorientador: Paulo Milton Barbosa Landim / Banca: Julião Soares de Souza Lima / Banca: Sergio Lazaro de Lima / Mestre

Neste trabalho é analisada a aplicação da técnica de Sistemas Imunológicos Artificiais (SIA) a um problema do mundo real: como predizer fraudes e furtos de energia elétrica. Vários trabalhos tem mostrado que épossível detectar padrões de dados anormais a partir dos dados de consumidores de energia elétrica e descobrir problemas como fraude e furto. Sistemas Imunológicos Artificiais é um ramo recente da Inteligência Computacional e tem diversas possíveis aplicações, sendo uma delas o reconhecimento de padrões. Mais de um algoritmo pode ser empregado para criar um SIA; no escopo deste trabalho será empregado o algoritmo Clonalg. A eficácia deste algoritmo é medida e comparada com a de outros métodos de classificação. A amostra de dados usada para validar este trabalho foi fornecida por uma companhia de energia elétrica. Os dados fornecidos foram selecionados e transformados com o objetivo de eliminar redundância e normalizar valores. / In this paper, we analyze the application of an Artificial Immune System (AIS) to a real world problem: how to predict electricity fraud and theft. Various works have explained that it is possible to detect abnormal data patterns from electricity consumers and discover problems like fraud and theft. Artificial Immune Systems is a recent branch of Computational Intelligence and has several possible applications, one of which is pattern recognition. More than one algorithm can be employed to create an AIS; we selected the Clonalg algorithm for our analysis. The efficiency of this algorithm is measured and compared with that of other classifier methods. The data sample used to validate this work was provided by an electrical energy company. The provided data were selected and transformed with the aim of eliminating redundant data and to normalize values.