Benjamín Hernández

Visual learning of texture descriptors for facial expression recognition in thermal imagery

Facial expression recognition is an active research area that finds a potential application in human emotion analysis. This work presents an illumination independent approach for facial expression recognition based on long wave infrared imagery. In general, facial expression recognition systems are designed considering the visible spectrum. This makes the recognition process not robust enough to be deployed in poorly illuminated environments. Common approaches to facial expression recognition of static images are designed considering three main parts: (1) region of interest selection, (2) feature extraction, and (3) image classification. Most published articles propose methodologies that solve each of these tasks in a decoupled way. We propose a Visual Learning approach based on evolutionary computation that solves the first two tasks simultaneously using a single evolving process. The first task consists in the selection of a set of suitable regions where the feature extraction is performed. The second task consists in tuning the parameters that defines the extraction of the Gray Level Co-occurrence Matrix used to compute region descriptors, as well as the selection of the best subsets of descriptors. The output of these two tasks is used for classification by a SVM committee. A dataset of thermal images with three different expression classes is used to validate the performance. Experimental results show effective classification when compared to a human observer, as well as a PCA-SVM approach. This paper concludes that: (1) thermal Imagery provides relevant information for FER, and (2) that the developed methodology can be taken as an efficient learning mechanism for different types of pattern recognition problems.

Automatic Feature Localization in Thermal Images for Facial Expression Recognition

We propose an unsupervised Local and Global feature extraction paradigm to approach the problem of facial expression recognition in thermal images. Starting from local, low-level features computed at interest point locations, our approach combines the localization of facial features with the holistic approach. The detailed steps are as follows: First, face localization using bi-modal thresholding is accomplished in order to localize facial features by way of a novel interest point detection and clustering approach. Second, we compute representative Eigenfeatures for feature extraction. Third, facial expression classification is made with a Support Vector Machine Committiee. Finally, the experiments over the IRIS data-set show that automation was achieved with good feature localization and classification performance.

A new accurate and flexible model based multi-corner detector for measurement and recognition

This paper introduces a new parametric model capable of the modeling and identification of a certain class of characteristic intensity variations described by polygonal structures of a depicted 3D scene. We propose a new parametric corner modeling based on a Unit Step Edge Function (USEF) that defines a straight line edge. The USEF function is a distribution function, which models the optical and physical characteristics found in digital imaging systems. The simplicity of the model definition provides the flexibility and generality useful in modeling complex corners. Based on our model, we have been able to create a multi-corner detector using simple operations as addition and multiplication. Once we have built a detector, it is possible to retrieve the information useful in other machine vision tasks. An example of retrieving the position of a projected L-corner is developed. A new criterion for high-accurate L-corner localization is introduced, and a comparison with five previous corner criteria is made.

Accurate L-corner measurement using USEF functions and evolutionary algorithms

Corner feature extraction is studied in this paper as a global optimization problem. We propose a new parametric corner modeling based on a Unit Step Edge Function (USEF) that defines a straight line edge. This USEF function is a distribution function, which models the optical and physical characteristics present in digital photogrammetric systems. We search model parameters characterizing completely single gray-value structures by means of least squares fit of the model to the observed image intensities. As the identification results relies on the initial parameter values and as usual with non-linear cost functions in general we cannot guarantee to find the global minimum. Hence, we introduce an evolutionary algorithm using an affine transformation in order to estimate the model parameters. This transformation encapsulates within a single algebraic form the two main operations, mutation and crossover, of an evolutionary algorithm. Experimental results show the superiority of our L-corner model applying several levels of noise with respect to simplex and simulated annealing.

Facial Expression Recognition in Nonvisual Imagery

This chapter presents two novel approaches that allow computer vision applications to perform human facial expression recognition (FER). From a prob lem standpoint, we focus on FER beyond the human visual spectrum, in long-wave infrared imagery, thus allowing us to offer illumination-independent solutions to this important human-computer interaction problem. From a methodological stand point, we introduce two different feature extraction techniques: a principal com ponent analysis-based approach with automatic feature selection and one based on texture information selected by an evolutionary algorithm. In the former, facial fea tures are selected based on interest point clusters, and classification is carried out us ing eigenfeature information; in the latter, an evolutionary-based learning algorithm searches for optimal regions of interest and texture features based on classification accuracy. Both of these approaches use a support vector machine-committee for classification. Results show effective performance for both techniques, from which we can conclude that thermal imagery contains worthwhile information for the FER problem beyond the human visual spectrum.

Flexible model-based multi-corner detector for accurate measurements and recognition

Recognition and photogrammetric tasks require models capable of detecting and measuring complex features normally found in all kind of natural and artificial scenes. Corners are special features in images, and are of great use in computing camera calibration, tracking and reconstruction. Basically, a corner is defined as the junction point of two or more straight-line edges. Previous methods devoted to corner detection are based on parametric models. However, a drawback from previous approaches is the cumbersome of the proposed models. This paper presents a new multi-corner detector based on a unit step edge function (USEF) that defines a straight line edge. Several USEFs are combined to produce complex corners using simple operators as addition and multiplication. As well as previous methods we search model parameters characterizing completely single gray-value structures by means of least squares fit of the model to the observed image intensities. Examples and experimental results illustrate the quality and efficacy of the detectors.

CUDA-based parallelization of a bio-inspired model for fast object classification

The need for highly accurate classification systems capable of working on real-time applications has increased in recent years. Nowadays, several computer vision tasks apply a classification step as part of bigger systems, hence requiring classification models that work at a fast pace. This rendered interesting the concept of real-time object classification to several research communities. In this paper, we propose to accelerate a bio-inspired model for object classification, which has given very good results when compared with other state-of-the-art proposals using the compute unified device architecture (CUDA) and exploiting computational capabilities of graphic processing units. The classification model that is used is called the artificial visual cortex, a novel bio-inspired approach for image classification. In this work, we show that through an implementation of this model in the CUDA framework it is possible to achieve real-time functionality. As a result, the proposed system is able to process images in average of up to 90 times faster than the original system.