Datasets, features, learning, and models in visual recognition

Visual recognition is a fundamental research topic in computer vision. This dissertation explores datasets, features, learning, and models used for visual recognition. In order to train visual models and evaluate different recognition algorithms, this dissertation develops an approach to collect object image datasets on web pages using an analysis of text around the image and of image appearance. This method exploits established online knowledge resources (Wikipedia pages for text; Flickr and Caltech data sets for images). The resources provide rich text and object appearance information. This dissertation describes results on two datasets. The first is Berg’s collection of 10 animal categories; on this dataset, we significantly outperform previous approaches. On an additional set of 5 categories, experimental results show the effectiveness of the method. Images are represented as features for visual recognition. This dissertation introduces a text-based image feature and demonstrates that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, downloaded from the Internet. Image tags are noisy. The method obtains the text features of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed under novel circumstances (say, a new viewing direction) must rely on its visual examples. This text feature may not change, because the auxiliary dataset likely contains a similar picture. While the tags associated with images are noisy, they are more stable when appearance changes. The performance of this feature is tested using PASCAL VOC 2006 and 2007 datasets. This feature performs well; it consistently improves the performance of visual object classifiers, and is particularly effective when the training dataset is small. With more and more collected training data, computational cost becomes a bottleneck, especially when training sophisticated classifiers such as kernelized SVM. This dissertation proposes a fast training algorithm called Stochastic Intersection Kernel Machine (SIKMA). This proposed training method will be useful for many vision problems, as it can produce a kernel classifier that is more accurate than a linear classifier, and can be trained on tens of thousands of examples in two minutes. It processes training examples one by one in a sequence, so memory cost is no longer the bottleneck to process large scale datasets. This dissertation applies this approach to train classifiers of Flickr groups with many group training examples. The resulting Flickr group prediction scores can be used to measure image similarity between two images. Experimental results on the Corel dataset and a PASCAL VOC dataset show the learned Flickr features perform better on image matching, retrieval, and classification than conventional visual features. Visual models are usually trained to best separate positive and negative training examples. However, when recognizing a large number of object categories, there may not be enough training examples for most objects, due to the intrinsic long-tailed distribution of objects in the real world. This dissertation proposes an approach to use comparative object similarity. The key insight is that, given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. This dissertation develops a regularized kernel machine algorithm to use this category dependent similarity regularization. Experiments on hundreds of categories show that our method can make significant improvement for categories with few or even no positive examples.

Object detection and recognition in complex scenes

Dissertação de Mestrado, Engenharia Informática, Faculdade de Ciências e Tecnologia, Universidade do Algarve, 2014

Quantum of Vision

Visual inputs to artificial and biological visual systems are often quantized: cameras accumulate photons from the visual world, and the brain receives action potentials from visual sensory neurons. Collecting more information quanta leads to a longer acquisition time and better performance. In many visual tasks, collecting a small number of quanta is sufficient to solve the task well. The ability to determine the right number of quanta is pivotal in situations where visual information is costly to obtain, such as photon-starved or time-critical environments. In these situations, conventional vision systems that always collect a fixed and large amount of information are infeasible. I develop a framework that judiciously determines the number of information quanta to observe based on the cost of observation and the requirement for accuracy. The framework implements the optimal speed versus accuracy tradeoff when two assumptions are met, namely that the task is fully specified probabilistically and constant over time. I also extend the framework to address scenarios that violate the assumptions. I deploy the framework to three recognition tasks: visual search (where both assumptions are satisfied), scotopic visual recognition (where the model is not specified), and visual discrimination with unknown stimulus onset (where the model is dynamic over time). Scotopic classification experiments suggest that the framework leads to dramatic improvement in photon-efficiency compared to conventional computer vision algorithms. Human psychophysics experiments confirmed that the framework provides a parsimonious and versatile explanation for human behavior under time pressure in both static and dynamic environments.

Generative Neural Networks for image iper-resolution and improvement of Optical Character Recognition's performances

The usage of Optical Character Recognition’s (OCR, systems is a widely spread technology into the world of Computer Vision and Machine Learning. It is a topic that interest many field, for example the automotive, where becomes a specialized task known as License Plate Recognition, useful for many application from the automation of toll road to intelligent payments. However, OCR systems need to be very accurate and generalizable in order to be able to extract the text of license plates under high variable conditions, from the type of camera used for acquisition to light changes. Such variables compromise the quality of digitalized real scenes causing the presence of noise and degradation of various type, which can be minimized with the application of modern approaches for image iper resolution and noise reduction. Oneclass of them is known as Generative Neural Networks, which are very strong ally for the solution of this popular problem.

Improvements on ICA mixture models for image pre-processing and segmentation

Today several different unsupervised classification algorithms are commonly used to cluster similar patterns in a data set based only on its statistical properties. Specially in image data applications, self-organizing methods for unsupervised classification have been successfully applied for clustering pixels or group of pixels in order to perform segmentation tasks. The first important contribution of this paper refers to the development of a self-organizing method for data classification, named Enhanced Independent Component Analysis Mixture Model (EICAMM), which was built by proposing some modifications in the Independent Component Analysis Mixture Model (ICAMM). Such improvements were proposed by considering some of the model limitations as well as by analyzing how it should be improved in order to become more efficient. Moreover, a pre-processing methodology was also proposed, which is based on combining the Sparse Code Shrinkage (SCS) for image denoising and the Sobel edge detector. In the experiments of this work, the EICAMM and other self-organizing models were applied for segmenting images in their original and pre-processed versions. A comparative analysis showed satisfactory and competitive image segmentation results obtained by the proposals presented herein. (C) 2008 Published by Elsevier B.V.

Amygdala hyperactivation in untreated depressed individuals

The amygdala participates in the detection and control of affective states, and has been proposed to be a site of dysfunction in affective disorders. To assess amygdala processing in individuals with unipolar depression, we applied a functional MRI (fMRI) paradigm previously shown to be sensitive to amygdala function. Fourteen individuals with untreated DSM-IV major depression and 15 healthy subjects were studied using fMRI with a standardized emotion face recognition task. Voxel-level data sets were subjected to a multiple-regression analysis, and functionally defined regions of interest (ROI), including bilateral amygdala, were analyzed with MANOVA. Pearson correlation coefficients between amygdala activation and HAM-D score also were performed. While both depressed and healthy groups showed increased amygdala activity when viewing emotive faces compared to geometric shapes, patients with unipolar depression showed relatively more activity than healthy subjects, particularly on the left. Positive Pearson correlations between amygdala activation and HAM-D score were found for both left and right ROIs in the patient group. This study provides in vivo imaging evidence to support the hypothesis of abnormal amygdala functioning in depressed individuals. (C) 2009 Elsevier Ireland Ltd. All rights reserved.

Fatty liver characterization and classification by ultrasound

Steatosis, also known as fatty liver, corresponds to an abnormal retention of lipids within the hepatic cells and reflects an impairment of the normal processes of synthesis and elimination of fat. Several causes may lead to this condition, namely obesity, diabetes, or alcoholism. In this paper an automatic classification algorithm is proposed for the diagnosis of the liver steatosis from ultrasound images. The features are selected in order to catch the same characteristics used by the physicians in the diagnosis of the disease based on visual inspection of the ultrasound images. The algorithm, designed in a Bayesian framework, computes two images: i) a despeckled one, containing the anatomic and echogenic information of the liver, and ii) an image containing only the speckle used to compute the textural features. These images are computed from the estimated RF signal generated by the ultrasound probe where the dynamic range compression performed by the equipment is taken into account. A Bayes classifier, trained with data manually classified by expert clinicians and used as ground truth, reaches an overall accuracy of 95% and a 100% of sensitivity. The main novelties of the method are the estimations of the RF and speckle images which make it possible to accurately compute textural features of the liver parenchyma relevant for the diagnosis.

Dominant set approach to ECG biometrics

Electrocardiographic (ECG) signals are emerging as a recent trend in the field of biometrics. In this paper, we propose a novel ECG biometric system that combines clustering and classification methodologies. Our approach is based on dominant-set clustering, and provides a framework for outlier removal and template selection. It enhances the typical workflows, by making them better suited to new ECG acquisition paradigms that use fingers or hand palms, which lead to signals with lower signal to noise ratio, and more prone to noise artifacts. Preliminary results show the potential of the approach, helping to further validate the highly usable setups and ECG signals as a complementary biometric modality.

Road lane detection for autonomous robot guidance

Computational Vision stands as the most comprehensive way of knowing the surrounding environment. Accordingly to that, this study aims to present a method to obtain from a common webcam, environment information to guide a mobile differential robot through a path similar to a roadway.

A Java expression evaluator for nonlinear programming

Finding the optimal value for a problem is usual in many areas of knowledge where in many cases it is needed to solve Nonlinear Optimization Problems. For some of those problems it is not possible to determine the expression for its objective function and/or its constraints, they are the result of experimental procedures, might be non-smooth, among other reasons. To solve such problems it was implemented an API contained methods to solve both constrained and unconstrained problems. This API was developed to be used either locally on the computer where the application is being executed or remotely on a server. To obtain the maximum flexibility both from the programmers’ and users’ points of view, problems can be defined as a Java class (because this API was developed in Java) or as a simple text input that is sent to the API. For this last one to be possible it was also implemented on the API an expression evaluator. One of the drawbacks of this expression evaluator is that it is slower than the Java native code. In this paper it is presented a solution that combines both options: the problem can be expressed at run-time as a string of chars that are converted to Java code, compiled and loaded dynamically. To wide the target audience of the API, this new expression evaluator is also compatible with the AMPL format.

Road lane detection for autonomous robot guidance

Computational Vision stands as the most comprehensive way of knowing the surrounding environment. Accordingly to that, this study aims to present a method to obtain from a common webcam, environment information to guide a mobile differential robot through a path similar to a roadway.

Combate à fraude em jogos de casino, assistida por computador

No decorrer dos últimos anos tem-se verificado um acréscimo do número de sistemas de videovigilância presentes nos mais diversos ambientes, sendo que estes se encontram cada vez mais sofisticados. Os casinos são um exemplo bastante popular da utilização destes sistemas sofisticados, sendo que vários casinos, hoje em dia, utilizam câmeras para controlo automático das suas operações de jogo. No entanto, atualmente existem vários tipos de jogos em que o controlo automático ainda não se encontra disponível, sendo um destes, o jogo Banca Francesa. A presente dissertação tem como objetivo propor um conjunto de algoritmos idealizados para um sistema de controlo e gestão do jogo de casino Banca Francesa através do auxílio de componentes pertencentes à área da computação visual, tendo em conta os contributos mais relevantes e existentes na área, elaborados por investigadores e entidades relacionadas. No decorrer desta dissertação são apresentados quatro módulos distintos, os quais têm como objetivo auxiliar os casinos a prevenir o acontecimento de fraudes durante o decorrer das suas operações, assim como auxiliar na recolha automática de resultados de jogo. Os quatro módulos apresentados são os seguintes: Dice Sample Generator – Módulo proposto para criação de casos de teste em grande escala; Dice Sample Analyzer – Módulo proposto para a deteção de resultados de jogo; Dice Calibration – Módulo proposto para calibração automática do sistema; Motion Detection – Módulo proposto para a deteção de fraude no jogo. Por fim, para cada um dos módulos, é apresentado um conjunto de testes e análises de modo a verificar se é possível provar o conceito para cada uma das propostas apresentadas.

A Real Time Vision System for Autonomous Systems: Characterization during a Middle Size Match

in RoboCup 2007: Robot Soccer World Cup XI

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.