Catálogo de publicaciones - libros

Although screening mammography is widely used for the detection of breast tumors, it is difficult for a radiologist to interpret correctly a mammogram. It is possible to improve this task by using a computer aided diagnosis system (CAD) which highlights the areas most likely to contain cancer cells. In this paper, we present and compare five different feature extraction methods for breast cancer detection in digitized mammograms. All the methods are based on features extracted from a local window and on a k -nearest neighbor classifier with fast search.

This paper presents a robust multiscale method to create thinned ridge map of fingerprint for automatic recognition by employing an elaborately designed wavelet function. Properties of the new wavelet function are substantially investigated. Some desirable characteristics of the local minimum produced by wavelet transform show that they are suitable to describe skeleton of ribbon-shape objects. A multiscale thinning algorithm based on the modulus minima of wavelet transform is proposed. The proposed algorithm is able to improve the skeleton representation of the ridge of fingerprint without side-effects and limitations of previous methods. Thinned ridge map helps to facilitate minutiae extraction for matching. Experiments validated effectiveness and efficiency of the proposed method.

We present an inherently discriminative approach to face recognition. This is achieved by automatically selecting key points from lines that sketch the face and extracting textural information at these locations. As the distribution of the lines depend on each individual face, the selected points will be person-dependent, achieving discrimination in an early stage of the recognition process. A robust shape matching algorithm has been used for the correspondence problem, and Gabor responses have been extracted at final points so that both shape and textural information are combined to measure similarities between faces. Face verification results are reported over the well known XM2VTS database.

In this paper, we present a novel fingerprinting method for image authentication, the fingerprint length of which is very short (only 81 bytes) and independent on image sizes. First, we extract features based on the block DCT coefficients of images, and binarize the feature map to get key blocks. Then we apply Principal Components Analysis (PCA) to the DCT Coefficients of key blocks. Finally, we take the quantized eigenvector matrix (9×9) as fingerprints. Experimental results show that the proposed method is discriminative, robust against compression, and sensitive to malicious modifications.

In this paper, we propose an original fingerprint matching algorithm using a set of intrinsic coordinate systems which each one is attached to each minutia according to its orientation estimated from fingerprint image. Exploiting these coordinate systems, minutiae locations can be redefined by means of projection of these minutiae coordinates on the relative reference of each orientation minutia. Thus, our matching algorithm use these relative minutiae coordinate to calculate the matching score between two fingerprints. To avoid the directional field estimation errors, we propose a minutia orientation variation in order to manage the projection errors of the location minutiae. With this technique, our approach doesn’t embedding fingerprint alignment into the minutia matching stage to design the robust algorithms for fingerprint recognition. The algorithm matching was tested on a fingerprint database DB2 used in FVC2000 and the results are promising.

In this paper we propose a face recognition algorithm that combines internal and external information of face images. Most of the previous works dealing with face recognition use only internal face features to classify, not considering the information located at head, chin and ears. Here we propose an adaptation of a top-down segmentation algorithm to extract external features from face images, and then we combine this information with internal features using a modification of the non parametric discriminant analysis technique. In the experimental results we show that the contribution of external features to face classification problems is clearly relevant, specially in presence of occlusions.

This paper presents an efficient iris recognition method based on wavelet multi-scale decompositions. A two-dimensional iris image should be transformed into a set of one-dimensional signals initially and then the wavelet coefficients matrix is generated by one-dimensional quadratic spline wavelet multi-scale decompositions. From the basic principles of probability theory, the elements at the same position in different wavelet coefficients matrices can be considered as a high correlated sequence. By applying a predetermined threshold, these wavelet coefficients matrices are finally transformed into a binary vector to represent iris features. The Hamming distance classifier is adopted to perform pattern matching between two feature vectors. Using an available iris database, final experiments show promising results for iris recognition with our proposed approach.

Speech Recognition and Verification are important components of a Keyword Spotting System whose performance is determined by both of them. In this paper, a new model-distance based utterance verification algorithm is proposed to improve the performance of the keyword spotting system. Furthermore, a substitution-error recovery module is built for improving of the system, which enhances the detection rate with the false-alarm rate remaining almost the same. Experiment shows that with the novel utterance verification method and new added substitution-error recovery module, the system performance is improved greatly.

In practical speech recognition applications, channel/environment conditions may not match those of the corpus used to estimate the acoustic models. A straightforward methodology is proposed in this paper by which the speech recognizer can match the acoustic conditions of input utterances, thus allowing instantaneous adaptation schemes. First a number of clusters is determined in the training material in a fully unsupervised way, using a dissimilarity measure based on shallow acoustic models. Then accurate acoustic models are estimated for each cluster, and finally a fast match strategy, based on the shallow models, is used to choose the most likely acoustic condition for each input utterance. The performance of the clustering algorithm was tested on two speech databases in Spanish: SENGLAR (read speech) and CORLEC-EHU-1 (spontaneous human-human dialogues). In both cases, speech utterances were consistently grouped by gender, by recording conditions or by background/channel noise. Furthermore, the fast match methodology led to noticeable improvements in preliminary phonetic recognition experiments, at 20-50% of the computational cost of the ML match.

This paper deals with the application of adaptive signal models for parametric speech and audio compression. The matching pursuit algorithm is used for extracting sinusoidal components and transients in audio signals. The resulting residue is perceptually modelled as a noise like signal. When a transient is detected, psychoacoustic-adapted matching pursuits are accomplished using a wavelet-based dictionary followed of an harmonic one. Otherwise, matching pursuit is applied only to the harmonic dictionary. This multi-part model (Sines + Transients + Noise) is successfully applied for speech and audio coding purposes, assuring high perceptual quality at low bit rates (close to 16 kbps for most of the signals considered for testing).