Catálogo de publicaciones - libros

Since microcalcification clusters are primary indicators of malignant types of breast cancer, its detection is important to prevent and treat the disease. This paper proposes a method for detection of microcalcification clusters in mammograms using sequential Difference of Gaussian filters (DoG). In a first stage, fifteen DoG filters are applied sequentially to extract the potential regions, and later, these regions are classified using the following features: absolute contrast, standard deviation of the gray level of the microcalcification and a moment of contour sequence (asymmetry coefficient). Once the microcalcifications are detected, two approaches for clustering are compared. In the first one, several microcalcification clusters are detected in each mammogram. In the other, all microcalcifications are considered in a single cluster. We demonstrate that the diagnosis based on the detection of several microcalcification clusters in a mammogram is more efficient than considering a single cluster including all the microcalcifications in the image.

Neural Stem Cells (NSCs) have a remarkable capacity to proliferate and differentiate to other cell types. This ability to differentiate to desirable phenotypes has motivated clinical interests, hence the interest here to segment Neural Stem Cell (NSC) clusters to locate the NSC clusters over time in a sequence of frames, and in turn to perform NSC cluster motion analysis. However the manual segmentation of such data is a tedious task. Thus, due to the increasing amount of cell data being collected, automated cell segmentation methods are highly desired. In this paper a novel level set based segmentation method is proposed to accomplish this segmentation. The method is initialization insensitive, making it an appropriate solution for automated segmentation systems. The proposed segmentation method has been successfully applied to NSC cluster segmentation.

The work investigates the use of multi dimensional histograms for segmentation of images of chronic wounds. We employ a Support Vector Machine (SVM) classifier for automatic extraction of wound region from an image. We show that the SVM classifier can generalize well on the difficult wound segmentation problem using only 3-D dimensional color histograms. We also show that color histograms of higher dimensions provide a better cue for robust separation of classes in the feature space. A key condition for the successful segmentation is an efficient sampling of multi-dimensional histograms. We propose a multi-dimensional histogram sampling technique for generation of input feature vectors for the SVM classifier. We compare the performance of the multi-dimensional histogram sampling with several existing techniques for quantization of 3-D color space. Our experimental results indicate that different sampling techniques used for the generation of input feature vectors may increase the performance of wound segmentation by about 25%.

Recognition of faces under varied poses has been a challenging area of research due to the complex dispersion of poses in feature space when compared to that of frontal faces. This paper presents a novel and robust pose-invariant face recognition method in order to improvise over existing face recognition techniques. First, we apply the TSL color model for detecting facial region and estimate the direction of face using facial features. The estimated pose vector is decomposed into X-Y-Z axes. Second, the input face is mapped by a deformable template using these vectors and the 3D face model. Finally, the mapped face is transformed to the frontal face which appropriates for face recognition by the estimated pose vector. Through the experiments, we come to validate the application of face detection model and the method for estimating facial poses. Moreover, the tests show that recognition rate is greatly boosted through the normalization of the poses.

This paper presents a morphological- and color-based method for face localization in color images. Basically, it uses a skin-color segmentation technique in a novel color space, YCg’Cr’, and the application of the shape information and the location of the facial features to the definition of a face model. First, a color-based segmentation technique detects the skin regions inside the image, and then, a combination of morphological operators and algorithms is used for completing the segmentation masks. At last, the feature extraction lets define a model based on the best-fit ellipse determined by the position of the eyes and mouth inside the image. Finally, the detection mask based on this ellipse is used for locating the face in the image under test.

A novel technique for facial feature detection in images of frontal faces is presented. We use a set of Gabor wavelet coefficients in different orientations and frequencies to analyze and describe facial features. However, due to the lack of sufficient local structures for describing facial features, Gabor wavelets can not perfectly capture the wide range of possible variations in the appearance of facial features, and thus can give many false positive (and sometimes false negative) responses. We show that the performance of such a feature detector can be significantly improved by using the local entropy of features. Complex regions in a face image, such as the eye, exhibit unpredictable local intensity and hence high entropy. Our method is robust against image rotation, varying brightness, varying contrast and a certain amount of scaling.

In this paper we propose a new range-based face recognition for significant improvement in the recognition rate using an optimized stereo acquisition system. The optimized 3D acquisition system consists of an eyes detection algorithm, facial pose direction distinction, and principal component analysis (PCA). The proposed method is carried out in the YCbCr color space in order to detect the face candidate area. To detect the correct face, it acquires the correct distance of the face candidate area and depth information of eyes and mouth. After scaling, the system transfers the pose change according to the distance. The face is finally recognized by the optimized PCA for each area with the facial pose elements detected. Simulation results with face recognition rate of 95.83% (100cm) in the front and 98.3% with the pose change were obtained successfully. Therefore, proposed method can be used to obtain high recognition rate with an appropriate scaling and pose change according to the distance.

In many realistic face recognition applications, such as surveillance photo identification, the subjects of interest usually have only a limited number of image samples a-priori. This makes the recognition a difficult task, especially when only one image sample is available for each subject. In such a case, the performance of many well known face recognition algorithms will deteriorate rapidly and some of the algorithms even fail to apply. In this paper, we introduced a novel scheme to solve the one training sample problem by combining a specific solution learned from the samples of interested subjects and a generic solution learned from the samples of many other subjects. A multi-learner framework is firstly applied to generate and combine a set of generic base learners followed by a second combination with the specific learner. Extensive experiments based on the FERET database suggests that in the scenario considered here, the proposed solution significantly boosts the recognition performance.

The present paper introduces a novel set of facial biometrics defined in the frequency domain representing “facial asymmetry”. A comparison with previously introduced spatial asymmetry measures suggests that the frequency domain representation provides an efficient approach for performing human identification in the presence of severe expressions and also for expression classification. Feature analysis indicates that asymmetry of the different regions of the face (e.g., eyes, mouth, nose) help in these two apparently conflicting classification problems. Another advantage of our frequency domain measures is that they are tolerant to some form of illumination variations. Error rates of less than 5% are observed for human identification in all cases. We then propose another asymmetry biometric based only on the Fourier domain phase and show a potential connection of asymmetry with phase.

Due to growing demands in such application areas as law enforcement, video surveillance, banking, and security system access authentication, automatic face recognition has attracted great attention in recent years. The advantages of facial identification over alternative methods, such as fingerprint identification, are based primarily on the fact that face is fairly easy to use and well accepted by people. However it is not robust enough to be used in most practical security applications because too sensitive to variations in pose and illumination. During the last few years, many algorithms have been proposed to overcome these problems using 2-D images, but very few has been made in order to address the problem of partial occlusions. In this paper, a fractal based technique is presented; the face image is partitioned in different regions of interest, each one is indexed by means of an IFS system. A new distance function is then introduced, in order to discard unuseful information. The proposed method turns out to be faster and more robust than other approaches in the state of the art.