Catálogo de publicaciones - libros

This work presents a new method to automatically locate frontal facial feature points under large scene variations (illumination, pose and facial expressions). First, we use a kernel-based tracker to detect and track the facial region in an image sequence. Then the results of the face tracking, i.e. face region and face pose, are used to constrain prominent facial feature detection and tracking. In our case, eyes and mouth corners are considered as prominent facial features. In a final step, we propose an improvement to the Bayesian Tangent Shape Model for the detection and tracking of the full shape model. A constrained regularization algorithm is proposed using the head pose and the accurately aligned prominent features to constrain the deformation parameters of the shape model. Extensive experiments demonstrate the accuracy and effectiveness of our proposed method.

In this paper, a new framework is presented for the quantitative evaluation of the performance of appearance models composed of an object descriptor and a similarity measure in the context of object tracking. The evaluation is based on natural videos, and takes advantage of existing ground-truths from object tracking benchmarks. The proposed metrics evaluate the ability of an appearance model to discriminate an object from the clutter. This allows comparing models which may use diverse kinds of descriptors or similarity measures in a principled manner. The performances measures can be global, but time-oriented performance evaluation is also presented. The insights that the proposed framework can bring on appearance models properties with respect to tracking are illustrated on natural video data.

This paper first presents a review of eigenface methods for face recognition and then introduces a new algorithm in this class. The main difference with previous approaches is the definition of the database. Classically, an image is exploited as a single vector, by concatenating its rows, while here we simply use all the rows as vectors during the training and the recognition stages. The new algorithm reduces the computational complexity of the classical eigenface method and also reaches a higher percentage of recognition. It is compared with other algorithms based on wavelets, aiming at reducing the computational burden. The most efficient wavelet families and other relevant parameters are discussed.

We propose a model based approach for the problem of face localization. Traditionally, images are represented in the RGB color space, which is a 3-dimensional space that includes the illumination factor. However, the human skin color of different ethnic groups has been shown to change because of brightness. We therefore propose to transform the RGB images into the HSV color-space. We then exclude the V component, and use the HS-domain to represent skin pixels using a Gaussian probability model. The model is used to obtain a skin likelihood image which is further transformed into a binary image using the fuzzy C-mean clustering (FCM) technique. The candidate skin regions are checked for some facial properties and finally a template face matching approach is used to localize the face.. The developed algorithm is found robust and reliable under various imaging conditions and even in the presence of structural objects like hairs, spectacles, etc.

The gait recognition is presented for human identification from a sequence of noisy silhouettes segmented from video. The proposed gait recognition algorithm gives better performance than the baseline algorithm because of segmentation of the object by using active shape model (ASM) algorithm. For the experiment, we used the HumanID Gait Challenge data set, which is the largest gait benchmarking data set with 122 objects, For realistic simulation we use various values for the following parameters; i) viewpoint, ii) shoe, iii) surface, iv) carrying condition, and v) time.

Detection and classification of skin regions plays important roles in many image processing and vision applications. In this paper, we present a statistical approach for fast skin detection in MPEG-compressed videos. Firstly, conditional probabilities of skin and non-skin pixels are extracted from manual marked training images. Then, candidate skin pixels are identified using the Bayesian maximum a posteriori decision rule. An optimal threshold is then obtained by analyzing of probability error on the basis of the likelihood ratio histogram of skin and non-skin pixels. Experiments from sequences with varying illuminations have demonstrated the effectiveness of our approach.

A double layer background representation to detect novelty in image sequences is shown. The model is capable of handling non-stationary scenarios, such as vehicle intersections. In the first layer, an adaptive pixel appearance background model is computed. Its subtraction with respect to the current image results in a blob description of moving objects. In the second layer, motion direction analysis is performed by a Mixture of Gaussians on the blobs. We have used both layers for representing the usual space of activities and for detecting unusual activity. Our experiments clearly showed that the proposed scheme is able to detect activities such as vehicles running on red light or making forbidden turns.

Real-time facial expression capture is an essential part for on-line performance animation. For efficiency and robustness, special devices such as head-mounted cameras and face-attached markers have been used. However, these devices can possibly cause some discomfort that may hinder a face puppeteer from performing natural facial expressions. In this paper, we propose a comprehensive solution for real-time facial expression capture without any of such devices. Our basic idea is first to capture the 2D facial features and 3D head motion exploiting anthropometric knowledge and then to capture their time-varying 3D positions only due to facial expression. We adopt a Kalman filter to track the 3D features guided by their captured 2D positions while correcting their drift due to 3D head motion as well as removing noises.

Most existing feature-point matching algorithms rely on photometric region descriptors to distinct and match feature points in two images. In this paper, we propose an efficient feature-point matching algorithm for finding point correspondences between two uncalibrated images separated by small or mid camera baselines. The proposed algorithm does not rely on photometric descriptors for matching. Instead, only the motion smoothness constraint is used, which states that the correspondence vectors within a small neighborhood usually have similar directions and magnitudes. The correspondences of feature points in a neighborhood are collectively determined in such a way that the smoothness of the local correspondence field is maximized. The smoothness constraint is self-contained in the correspondence field and is robust to the camera motion, scene structure, illumination, etc. This makes the entire point-matching process texture-independent, descriptor-free and robust. The experimental results show that the proposed method performs much better than the intensity-based block-matching technique, even when the image contrast varies clearly across images.

We present in this article a new supervised evaluation criterion that enables the quantification of the quality of region segmentation algorithms. This criterion is compared with seven well-known criteria available in this context. To that end, we test the different methods on natural images by using a subjective evaluation involving different experts from the French community in image processing. Experimental results show the benefit of this new criterion.