Human skin is made of a stack of different layers, each of which reflects a portion of impinging light, after absorbing a certain amount of it by the pigments which lie in the layer. The main pigments responsible for skin color origins are melanin and hemoglobin. Skin segmentation plays an important role in a wide range of image processing and computer vision applications. In short, there are three major approaches for skin segmentation: rule-based, machine learning and hybrid. They differ in terms of accuracy and computational efficiency. Generally, machine learning and hybrid approaches outperform the rule-based methods but require a large and representative training dataset and, sometimes, costly classification time as well, which can be a deal breaker for real-time applications. In this work, we propose an improvement, in three distinct versions, of a novel method for rule-based skin segmentation that works in the YCbCr color space. Our motivation is based on the hypotheses that: (1) the original rule can be complemented and, (2) human skin pixels do not appear isolated, i.e. neighborhood operations are taken into consideration. The method is a combination of some correlation rules based on these hypotheses. Such rules evaluate the combinations of chrominance Cb, Cr values to identify the skin pixels depending on the shape and size of dynamically generated skin color clusters. The method is very efficient in terms of computational effort as well as robust in very complex images. / A pele humana é constituída de uma série de camadas distintas, cada uma das quais reflete uma porção de luz incidente, depois de absorver uma certa quantidade dela pelos pigmentos que se encontram na camada. Os principais pigmentos responsáveis pela origem da cor da pele são a melanina e a hemoglobina. A segmentação de pele desempenha um papel importante em uma ampla gama de aplicações em processamento de imagens e visão computacional. Em suma, existem três abordagens principais para segmentação de pele: baseadas em regras, aprendizado de máquina e híbridos. Elas diferem em termos de precisão e eficiência computacional. Geralmente, as abordagens com aprendizado de máquina e as híbridas superam os métodos baseados em regras, mas exigem um conjunto de dados de treinamento grande e representativo e, por vezes, também um tempo de classificação custoso, que pode ser um fator decisivo para aplicações em tempo real. Neste trabalho, propomos uma melhoria, em três versões distintas, de um novo método de segmentação de pele baseado em regras que funciona no espaço de cores YCbCr. Nossa motivação baseia-se nas hipóteses de que: (1) a regra original pode ser complementada e, (2) pixels de pele humana não aparecem isolados, ou seja, as operações de vizinhança são levadas em consideração. O método é uma combinação de algumas regras de correlação baseadas nessas hipóteses. Essas regras avaliam as combinações de valores de crominância Cb, Cr para identificar os pixels de pele, dependendo da forma e tamanho dos agrupamentos de cores de pele gerados dinamicamente. O método é muito eficiente em termos de esforço computacional, bem como robusto em imagens muito complexas.

Human skin is made of a stack of different layers, each of which reflects a portion of impinging light, after absorbing a certain amount of it by the pigments which lie in the layer. The main pigments responsible for skin color origins are melanin and hemoglobin. Skin segmentation plays an important role in a wide range of image processing and computer vision applications. In short, there are three major approaches for skin segmentation: rule-based, machine learning and hybrid. They differ in terms of accuracy and computational efficiency. Generally, machine learning and hybrid approaches outperform the rule-based methods but require a large and representative training dataset and, sometimes, costly classification time as well, which can be a deal breaker for real-time applications. In this work, we propose an improvement, in three distinct versions, of a novel method for rule-based skin segmentation that works in the YCbCr color space. Our motivation is based on the hypotheses that: (1) the original rule can be complemented and, (2) human skin pixels do not appear isolated, i.e. neighborhood operations are taken into consideration. The method is a combination of some correlation rules based on these hypotheses. Such rules evaluate the combinations of chrominance Cb, Cr values to identify the skin pixels depending on the shape and size of dynamically generated skin color clusters. The method is very efficient in terms of computational effort as well as robust in very complex images. / A pele humana é constituída de uma série de camadas distintas, cada uma das quais reflete uma porção de luz incidente, depois de absorver uma certa quantidade dela pelos pigmentos que se encontram na camada. Os principais pigmentos responsáveis pela origem da cor da pele são a melanina e a hemoglobina. A segmentação de pele desempenha um papel importante em uma ampla gama de aplicações em processamento de imagens e visão computacional. Em suma, existem três abordagens principais para segmentação de pele: baseadas em regras, aprendizado de máquina e híbridos. Elas diferem em termos de precisão e eficiência computacional. Geralmente, as abordagens com aprendizado de máquina e as híbridas superam os métodos baseados em regras, mas exigem um conjunto de dados de treinamento grande e representativo e, por vezes, também um tempo de classificação custoso, que pode ser um fator decisivo para aplicações em tempo real. Neste trabalho, propomos uma melhoria, em três versões distintas, de um novo método de segmentação de pele baseado em regras que funciona no espaço de cores YCbCr. Nossa motivação baseia-se nas hipóteses de que: (1) a regra original pode ser complementada e, (2) pixels de pele humana não aparecem isolados, ou seja, as operações de vizinhança são levadas em consideração. O método é uma combinação de algumas regras de correlação baseadas nessas hipóteses. Essas regras avaliam as combinações de valores de crominância Cb, Cr para identificar os pixels de pele, dependendo da forma e tamanho dos agrupamentos de cores de pele gerados dinamicamente. O método é muito eficiente em termos de esforço computacional, bem como robusto em imagens muito complexas.

Submitted by Erika Demachki (erikademachki@gmail.com) on 2015-10-23T18:23:32Z No. of bitstreams: 2 Dissertação - Rafael Divino Ferreira Feitosa - 2015.pdf: 7893703 bytes, checksum: 12af470c3ca2fb4a3d0bd3885bfde46d (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Erika Demachki (erikademachki@gmail.com) on 2015-10-23T18:25:45Z (GMT) No. of bitstreams: 2 Dissertação - Rafael Divino Ferreira Feitosa - 2015.pdf: 7893703 bytes, checksum: 12af470c3ca2fb4a3d0bd3885bfde46d (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-10-23T18:25:45Z (GMT). No. of bitstreams: 2 Dissertação - Rafael Divino Ferreira Feitosa - 2015.pdf: 7893703 bytes, checksum: 12af470c3ca2fb4a3d0bd3885bfde46d (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2015-04-14 / Skin detection techniques are widely applied to locate and to track parts of the human body with the objective of posterior recognition, having received great attention in recent years in the development of research in reason to the innumerable possible applications with the detection and tracking of faces, identification of naked people, identification of hand movements, among others. The present work proposed the construction of mathematical models for the detection of human skin tones such as, white, yellow, brown and black in digital color images in the RGB and HSV color spaces. Using a set of human skin tone samples, mathematical models were constructed describing how the variables of each color pixel in the RGB and HSV systems interrelate. To understand the answer of the proposed system, the mechanistic model was chosen, dividing it into components, observing the behavior of each part and the interactions that occurred between them. The proposed RGB filter reached a 98.3657% reduction index of the spectrum, classifying only 1.6343% (253,159 tones) as possible skin tones and the HSV model reduced the likely spectrum to 2.5352% (94,030 tones), discarding 97.4648% of the colors as candidates for human skin tones. When the proposed filters, were applied to the reduction of the probable range of human skin tones, well-defined bands in the geometric representation of the color spaces were selected. The experimental validation of the effectiveness of the RGB model showed that the proposed filter has conservative characteristics in the detection of skin, mistakenly classifying as skin only 6.7163% of the sample space. The proposed RGB filter has low sensitivity of 61.0831% and high specificity of 95.2769% in the detection of human skin in digital images. The HSV model had rates of (54,6333%) low sensitivity and (92,6390%) high specificity, considered low when compared to the performance of the other algorithms. / Técnicas de detecção de pele são amplamente aplicadas para localizar e rastrear partes do corpo humano com o objetivo de posterior reconhecimento, tendo recebido nos últimos anos grande atenção no desenvolvimento de pesquisas em razão das inúmeras aplicações possíveis como detecção e rastreamento de faces, identificação de pessoas nuas, identificação de movimentos das mãos, entre outras. O presente trabalho propôs construir 2 modelos matemáticos para detecção de tons de pele humana branca, amarela, parda e preta em imagens digitais coloridas nos espaços de cores RGB e HSV. Utilizandose de um conjunto de amostras de tons de pele humana foram construídos modelos matemáticos que descrevem como as variáveis de cada pixel de cor nos sistemas RGB e HSV se relacionam. Para compreender a resposta do sistema proposto, foi escolhido o modelo mecanístico, dividindo-o em componentes e observando o comportamento de cada parte e das interações que ocorreram entre elas. O filtro RGB proposto alcançou o índice de redução de 98,3657% do espectro, classificando apenas 1,6343% (253.159 tons) como possíveis tons de pele e o modelo HSV reduziu para 2,5352% (94.030 tons) o espectro provável, descartando 97,4648% das cores como candidatas a tons de pele humana. Os filtros propostos, quando aplicados à redução do espectro provável de tons de pele humana, selecionaram faixas bem definidas na representação geométrica dos espaços de cores. A validação experimental da eficácia do modelo RGB mostrou que o filtro proposto apresenta características conservadoras na detecção de pele classificando como pele, erroneamente, apenas 4,5075% do espaço amostral. O filtro RGB proposto possui baixa sensibilidade de 56,9698% e elevada especificidade de 95,4925% na detecção de pele humana em imagens digitais. O modelo HSV apresentou taxas de baixa sensibilidade (54,6333%) e alta especificidade (92,6390%), quando comparadas ao desempenho dos demais algoritmos propostos na literatura.

>Magister Scientiae - MSc / The SASL project is in the process of developing a machine translation system that can translate fully-fledged phrases between SASL and English in real-time. To-date, several systems have been developed by the project focusing on facial expression, hand shape, hand motion, hand orientation and hand location recognition and estimation. Achmed developed a highly accurate upper body pose recognition and estimation system. The system is capable of recognizing and estimating the location of the arms from a twodimensional video captured from a monocular view at an accuracy of 88%. The system operates at well below real-time speeds. This research aims to investigate the use of optimizations and parallel processing techniques using the CUDA framework on Achmed’s algorithm to achieve real-time upper body pose recognition and estimation. A detailed analysis of Achmed’s algorithm identified potential improvements to the algorithm. Are- implementation of Achmed’s algorithm on the CUDA framework, coupled with these improvements culminated in an enhanced upper body pose recognition and estimation system that operates in real-time with an increased accuracy.

Magister Scientiae - MSc / Research has shown that five fundamental parameters are required to recognize any sign language gesture: hand shape, hand motion, hand location, hand orientation and facial expressions. The South African Sign Language (SASL) research group at the University of the Western Cape (UWC) has created several systems to recognize sign language gestures using single parameters. These systems are, however, limited to a vocabulary size of 20 – 23 signs, beyond which the recognition accuracy is expected to decrease. The first aim of this research is to investigate the use of two parameters – hand motion and hand shape – to recognise a larger vocabulary of SASL gestures at a high accuracy. Also, the majority of related work in the field of sign language gesture recognition using these two parameters makes use of Hidden Markov Models (HMMs) to classify gestures. Hidden Markov Support Vector Machines (HM-SVMs) are a relatively new technique that make use of Support Vector Machines (SVMs) to simulate the functions of HMMs. Research indicates that HM-SVMs may perform better than HMMs in some applications. To our knowledge, they have not been applied to the field of sign language gesture recognition. This research compares the use of these two techniques in the context of SASL gesture recognition. The results indicate that, using two parameters results in a 15% increase in accuracy over the use of a single parameter. Also, it is shown that HM-SVMs are a more accurate technique than HMMs, generally performing better or at least as good as HMMs.

This thesis is focused on face detection in static picture. Theoretical part contains color spaces (RGB, HSI, YCbCr), methods for skin detection (explicit, parametric or non-parametric methods), image metric, edge detection, mathematical morphology, methods for classification faces (appearance-based methods, feature invariant approaches, knowledge-based methods, template matching methods). Practical part of this thesis contains concept and practical realization two algorithms for segmentation skin in static image (simple method based on Cr chroma components and statistical method). Practical part contains concept and practical realization two algorithms for classification face (appearance-based method and template matching method) too.

This paper discusses problems of computer vision, which deals with face detection and recognition in image and video sequence at real time. All methods are designed for color images and are based on skin detection on the basis of information of human skin color. For skin detection is used very effective method Gaussian distribution. All of the areas, which have human skin color, are classified. This classification specifies, which area is or isn’t face. For face detection is used correlation method, complete with eigenfaces method. All areas classified as a face are subsequently recognized by the eigenfaces method. Result of recognition phase is information about human identity.

The master thesis deals with the topic of detecting faces in digital images. There are generally described and classified the most frequently used methods and discussed their advantages and disadvantages. More detailed is described skin color detection, eye and mouth detection and are teoretically described machine learning algorithms and detection based on Haar-classifiers. The work aims to implementation of these methods in the OpenCV library, it refers to practical application of them a finally compares different provided trained files.

This thesis is focused on gesture recognition in video. The main purpose of this thesis was to create an algorithm and an application that can recognize selected gestures using a~video obtained through a~standard webcamera. The intention was to control an application program, such as video player. The approach used to achieve this goal was to exploit methods of feature extraction, tracking, and machine learning.

This master thesis deals with the use of gestures in user interfaces. The goal of this thesis is to create library for hand tracking and gesture recognition in real time. For hand tracking was choosen algorithm Flock of Features. Classification of gestures is done by using algorithm DTW. This thesis also contains stage design, design and implementation of a system that uses this library. Within the tests was tested control of various application using this library.