Interface caligráfica de escrita no ar

Na atualidade, está a emergir um novo paradigma de interação, designado por Natural User Interface (NUI) para reconhecimento de gestos produzidos com o corpo do utilizador. O dispositivo de interação Microsoft Kinect foi inicialmente concebido para controlo de videojogos, para a consola Xbox360. Este dispositivo demonstra ser uma aposta viável para explorar outras áreas, como a do apoio ao processo de ensino e de aprendizagem para crianças do ensino básico. O protótipo desenvolvido visa definir um modo de interação baseado no desenho de letras no ar, e realizar a interpretação dos símbolos desenhados, usando os reconhecedores de padrões Kernel Discriminant Analysis (KDA), Support Vector Machines (SVM) e $N. O desenvolvimento deste projeto baseou-se no estudo dos diferentes dispositivos NUI disponíveis no mercado, bibliotecas de desenvolvimento NUI para este tipo de dispositivos e algoritmos de reconhecimento de padrões. Com base nos dois elementos iniciais, foi possível obter uma visão mais concreta de qual o hardware e software disponíveis indicados à persecução do objetivo pretendido. O reconhecimento de padrões constitui um tema bastante extenso e complexo, de modo que foi necessária a seleção de um conjunto limitado deste tipo de algoritmos, realizando os respetivos testes por forma a determinar qual o que melhor se adequava ao objetivo pretendido. Aplicando as mesmas condições aos três algoritmos de reconhecimento de padrões permitiu avaliar as suas capacidades e determinar o $N como o que apresentou maior eficácia no reconhecimento. Por último, tentou-se averiguar a viabilidade do protótipo desenvolvido, tendo sido testado num universo de elementos de duas faixas etárias para determinar a capacidade de adaptação e aprendizagem destes dois grupos. Neste estudo, constatou-se um melhor desempenho inicial ao modo de interação do grupo de idade mais avançada. Contudo, o grupo mais jovem foi revelando uma evolutiva capacidade de adaptação a este modo de interação melhorando progressivamente os resultados.

Classification Performance of Multilayer Perceptrons with Different Risk Functionals

In the present paper we assess the performance of information-theoretic inspired risks functionals in multilayer perceptrons with reference to the two most popular ones, Mean Square Error and Cross-Entropy. The information-theoretic inspired risks, recently proposed, are: HS and HR2 are, respectively, the Shannon and quadratic Rényi entropies of the error; ZED is a risk reflecting the error density at zero errors; EXP is a generalized exponential risk, able to mimic a wide variety of risk functionals, including the information-thoeretic ones. The experiments were carried out with multilayer perceptrons on 35 public real-world datasets. All experiments were performed according to the same protocol. The statistical tests applied to the experimental results showed that the ubiquitous mean square error was the less interesting risk functional to be used by multilayer perceptrons. Namely, mean square error never achieved a significantly better classification performance than competing risks. Cross-entropy and EXP were the risks found by several tests to be significantly better than their competitors. Counts of significantly better and worse risks have also shown the usefulness of HS and HR2 for some datasets.

Feature Discretization with Relevance and Mutual Information Criteria

Feature discretization (FD) techniques often yield adequate and compact representations of the data, suitable for machine learning and pattern recognition problems. These representations usually decrease the training time, yielding higher classification accuracy while allowing for humans to better understand and visualize the data, as compared to the use of the original features. This paper proposes two new FD techniques. The first one is based on the well-known Linde-Buzo-Gray quantization algorithm, coupled with a relevance criterion, being able perform unsupervised, supervised, or semi-supervised discretization. The second technique works in supervised mode, being based on the maximization of the mutual information between each discrete feature and the class label. Our experimental results on standard benchmark datasets show that these techniques scale up to high-dimensional data, attaining in many cases better accuracy than existing unsupervised and supervised FD approaches, while using fewer discretization intervals.

Exploiting the bin-class histograms for feature selection on discrete data

In machine learning and pattern recognition tasks, the use of feature discretization techniques may have several advantages. The discretized features may hold enough information for the learning task at hand, while ignoring minor fluctuations that are irrelevant or harmful for that task. The discretized features have more compact representations that may yield both better accuracy and lower training time, as compared to the use of the original features. However, in many cases, mainly with medium and high-dimensional data, the large number of features usually implies that there is some redundancy among them. Thus, we may further apply feature selection (FS) techniques on the discrete data, keeping the most relevant features, while discarding the irrelevant and redundant ones. In this paper, we propose relevance and redundancy criteria for supervised feature selection techniques on discrete data. These criteria are applied to the bin-class histograms of the discrete features. The experimental results, on public benchmark data, show that the proposed criteria can achieve better accuracy than widely used relevance and redundancy criteria, such as mutual information and the Fisher ratio.

Estimation of 3D shapes using active surface models

This paper addresses the estimation of surfaces from a set of 3D points using the unified framework described in [1]. This framework proposes the use of competitive learning for curve estimation, i.e., a set of points is defined on a deformable curve and they all compete to represent the available data. This paper extends the use of the unified framework to surface estimation. It o shown that competitive learning performes better than snakes, improving the model performance in the presence of concavities and allowing to desciminate close surfaces. The proposed model is evaluated in this paper using syntheticdata and medical images (MRI and ultrasound images).

A MAP approach to evidence accumulation clustering

The Evidence Accumulation Clustering (EAC) paradigm is a clustering ensemble method which derives a consensus partition from a collection of base clusterings obtained using different algorithms. It collects from the partitions in the ensemble a set of pairwise observations about the co-occurrence of objects in a same cluster and it uses these co-occurrence statistics to derive a similarity matrix, referred to as co-association matrix. The Probabilistic Evidence Accumulation for Clustering Ensembles (PEACE) algorithm is a principled approach for the extraction of a consensus clustering from the observations encoded in the co-association matrix based on a probabilistic model for the co-association matrix parameterized by the unknown assignments of objects to clusters. In this paper we extend the PEACE algorithm by deriving a consensus solution according to a MAP approach with Dirichlet priors defined for the unknown probabilistic cluster assignments. In particular, we study the positive regularization effect of Dirichlet priors on the final consensus solution with both synthetic and real benchmark data.

Hyperspectral imagery framework for unmixing and dimensionality estimation

In hyperspectral imagery a pixel typically consists mixture of spectral signatures of reference substances, also called endmembers. Linear spectral mixture analysis, or linear unmixing, aims at estimating the number of endmembers, their spectral signatures, and their abundance fractions. This paper proposes a framework for hyperpsectral unmixing. A blind method (SISAL) is used for the estimation of the unknown endmember signature and their abundance fractions. This method solve a non-convex problem by a sequence of augmented Lagrangian optimizations, where the positivity constraints, forcing the spectral vectors to belong to the convex hull of the endmember signatures, are replaced by soft constraints. The proposed framework simultaneously estimates the number of endmembers present in the hyperspectral image by an algorithm based on the minimum description length (MDL) principle. Experimental results on both synthetic and real hyperspectral data demonstrate the effectiveness of the proposed algorithm.

Text classification using compression-based dissimilarity measures

Arguably, the most difficult task in text classification is to choose an appropriate set of features that allows machine learning algorithms to provide accurate classification. Most state-of-the-art techniques for this task involve careful feature engineering and a pre-processing stage, which may be too expensive in the emerging context of massive collections of electronic texts. In this paper, we propose efficient methods for text classification based on information-theoretic dissimilarity measures, which are used to define dissimilarity-based representations. These methods dispense with any feature design or engineering, by mapping texts into a feature space using universal dissimilarity measures; in this space, classical classifiers (e.g. nearest neighbor or support vector machines) can then be used. The reported experimental evaluation of the proposed methods, on sentiment polarity analysis and authorship attribution problems, reveals that it approximates, sometimes even outperforms previous state-of-the-art techniques, despite being much simpler, in the sense that they do not require any text pre-processing or feature engineering.

Tradutor da língua gestual portuguesa modelo de tradução bidireccional

O projeto tem como objetivo desenvolver e avaliar um modelo que facilita o acesso para pessoas surdas ou com deficiência auditiva, o acesso ao conteúdo digital - em particular o conteúdo educacional e objetos de aprendizagem – a criação de condições para uma maior inclusão social de surdos e deficientes auditivos. Pretende-se criar um modelo bidirecional, em que permite a pessoas com deficiências auditivas, possam se comunicar com outras pessoas, com a tradução da Língua Gestual Portuguesa (LGP) para a Língua Portuguesa (LP) e que outras pessoas não portadoras de qualquer deficiência auditiva possam por sua vez comunicar com os surdos ou deficientes auditivos através da tradução da LP para a LGP. Há um conjunto de técnicas que poderíamos nos apoiar para desenvolver o modelo e implementar a API de tradução da LGP em LP. Muitos estudos são feitos com base nos modelos escondidos de Markov (HMM) para efetuar o reconhecimento. Recentemente os estudos estão a caminhar para o uso de técnicas como o “Dynamic Time Warping” (DTW), que tem tido mais sucesso do que outras técnicas em termos de performance e de precisão. Neste projeto optamos por desenvolver a API e o Modelo, com base na técnica de aprendizagem Support Vector Machines (SVM) por ser uma técnica simples de implementar e com bons resultados demonstrados em reconhecimento de padrões. Os resultados obtidos utilizando esta técnica de aprendizagem foram bastante ótimos, como iremos descrever no decorrer do capítulo 4, mesmo sabendo que utilizamos dois dispositivos para capturar dados de descrição de cada gesto. Toda esta tese integra-se no âmbito do projeto científico/ investigação a decorrer no grupo de investigação GILT, sob a coordenação da professora Paula Escudeiro e suportado pela Fundação para Ciência e Tecnologia (FCT).

Real-Time Visual Ground-Truth System for Indoor Robotic Applications

The robotics community is concerned with the ability to infer and compare the results from researchers in areas such as vision perception and multi-robot cooperative behavior. To accomplish that task, this paper proposes a real-time indoor visual ground truth system capable of providing accuracy with at least more magnitude than the precision of the algorithm to be evaluated. A multi-camera architecture is proposed under the ROS (Robot Operating System) framework to estimate the 3D position of objects and the implementation and results were contextualized to the Robocup Middle Size League scenario.

Structured Light System Calibration for Perception in Underwater Tanks

The process of visually exploring underwater environments is still a complex problem. Underwater vision systems require complementary means of sensor information to help overcome water disturbances. This work proposes the development of calibration methods for a structured light based system consisting on a camera and a laser with a line beam. Two different calibration procedures that require only two images from different viewpoints were developed and tested in dry and underwater environments. Results obtained show, an accurate calibration for the camera/projector pair with errors close to 1 mm even in the presence of a small stereos baseline.

Aquisição e Processamento de Imagens de Microscopia em Balística Forense

As novas tecnologias aplicadas ao processamento de imagem e reconhecimento de padrões têm sido alvo de um grande progresso nas últimas décadas. A sua aplicação é transversal a diversas áreas da ciência, nomeadamente a área da balística forense. O estudo de evidências (invólucros e projeteis) encontradas numa cena de crime, recorrendo a técnicas de processamento e análise de imagem, é pertinente pelo facto de, aquando do disparo, as armas de fogo imprimirem marcas únicas nos invólucros e projéteis deflagrados, permitindo relacionar evidências deflagradas pela mesma arma. A comparação manual de evidências encontradas numa cena de crime com evidências presentes numa base de dados, em termos de parâmetros visuais, constitui uma abordagem demorada. No âmbito deste trabalho pretendeu-se desenvolver técnicas automáticas de processamento e análise de imagens de evidências, obtidas através do microscópio ótico de comparação, tendo por base algoritmos computacionais. Estes foram desenvolvidos com recurso a pacotes de bibliotecas e a ferramentas open-source. Para a aquisição das imagens de evidências balísticas foram definidas quatro modalidades de aquisição: modalidade Planar, Multifocus, Microscan e Multiscan. As imagens obtidas foram aplicados algoritmos de processamento especialmente desenvolvidos para o efeito. A aplicação dos algoritmos de processamento permite a segmentação de imagem, a extração de características e o alinhamento de imagem. Este último tem como finalidade correlacionar as evidências e obter um valor quantitativo (métrica), indicando o quão similar essas evidências são. Com base no trabalho desenvolvido e nos resultados obtidos, foram definidos protocolos de aquisição de imagens de microscopia, que possibilitam a aquisição de imagens das regiões passiveis de serem estudadas, assim como algoritmos que permitem automatizar o posterior processo de alinhamento de imagens de evidências, constituindo uma vantagem em relação ao processo de comparação manual.

Algorithms for information extraction and signal annotation on long-term biosignals using clustering techniques

Dissertação para obtenção do Grau de Mestre em Engenharia Biomédica

Time series morphological analysis applied to biomedical signals events detection

Dissertation submitted in the fufillment of the requirements for the Degree of Master in Biomedical Engineering

Automatic eye blink detection using consumer web cameras

This research aims to advance blinking detection in the context of work activity. Rather than patients having to attend a clinic, blinking videos can be acquired in a work environment, and further automatically analyzed. Therefore, this paper presents a methodology to perform the automatic detection of eye blink using consumer videos acquired with low-cost web cameras. This methodology includes the detection of the face and eyes of the recorded person, and then it analyzes the low-level features of the eye region to create a quantitative vector. Finally, this vector is classified into one of the two categories considered —open and closed eyes— by using machine learning algorithms. The effectiveness of the proposed methodology was demonstrated since it provides unbiased results with classification errors under 5%