Catálogo de publicaciones - libros

Phrase-based statistical models constitute one of the most competitive pattern-recognition approaches to machine translation. In this case, the source sentence is fragmented into phrases, then, each phrase is translated by using a stochastic dictionary. One shortcoming of this phrase-based model is that it does not have an adequate generalization capability. If a sequence of words has not been seen in training, it cannot be translated as a whole phrase. In this paper we try to overcome this drawback. The basic idea is that if a source phrase is not in our dictionary (has not been seen in training), we look for the most similar in our dictionary and try to adapt its translation to the source phrase. We are using the well known edit distance as a measure of similarity. We present results from an English-Spanish task (XRCE).

This paper presents a novel approach for motion segmentation from feature trajectories with missing data. It consists of two stages. In the first stage, missing data are filled in by applying a factorization technique to the matrix of trajectories. Since the number of objects in the scene is not given and the rank of this matrix can not be directly computed, a simple technique for matrix rank estimation, based on a frequency spectra representation, is proposed. In the second stage, motion segmentation is obtained by using a clustering approach based on the normalized cuts criterion. Finally, the shape and motion of each of the obtained clusters (i.e., single objects) are recovered by applying classical SFM techniques. Experiments with synthetic and real data are provided in order to demonstrate the viability of the proposed approach.

In this paper we evaluate an automatic segmentation algorithm able to identify the set of rigidly moving points within a deformable object given the 2D measurements acquired by a perspective camera. The method is based on a RANSAC algorithm with guided sampling and an estimation of the fundamental matrices from pairwise frames in the sequence. Once the segmentation of rigid and non-rigid points is available, the set of rigid points could be used to estimate the internal camera calibration parameters, the overall rigid motion and the non-rigid 3D structure.

Most applications on Human Computer Interaction (HCI) require to extract the movements of user faces, while avoiding high memory and time expenses. Moreover, HCI systems usually use low-cost cameras, while current face tracking techniques strongly depend on the image resolution. In this paper, we tackle the problem of eyelid tracking by using Appearance-Based Models, thus achieving accurate estimations of the movements of the eyelids, while avoiding cues, which require high-resolution faces, such as edge detectors or colour information. Consequently, we can track the fast and spontaneous movements of the eyelids, a very hard task due to the small resolution of the eye regions. Subsequently, we combine the results of eyelid tracking with the estimations of other facial features, such as the eyebrows and the lips. As a result, a hierarchical tracking framework is obtained: we demonstrate that combining two appearance-based trackers allows to get accurate estimates for the eyelid, eyebrows, lips and also the 3D head pose by using low-cost video cameras and in real-time. Therefore, our approach is shown suitable to be used for further facial-expression analysis.

This work describes an approach for the interpretation and explanation of human behavior in image sequences, within the context of a . The information source is the geometrical data obtained by applying tracking algorithms to an image sequence, which is used to generate conceptual data. The spatial characteristics of the scene are automatically extracted from the resuling tracking trajectories obtained during a training period. Interpretation is achieved by means of a rule-based inference engine called and a behavior modeling tool called . These tools are used to generate conceptual descriptions which semantically describe observed behaviors.

In this paper we evaluate the performance of the two most successful state-of-the-art descriptors, applied to the task of visual object detection and localization in images. In the first experiment we use these descriptors, combined with binary classifiers, to test the presence/absence of object in a target image. In the second experiment, we try to locate faces in images, by using a structural model. The results show that HMAX performs slightly better than SIFT in these tasks.

A moment-based approach is developed to constructing tree-structured descriptions of patterns given by region-based shapes with grayscale attributes. The proposed representation is approximately invariant with respect to the pattern rotation, translation, scale, and level of brightness. The tree-like structure of the pattern representations provides their independent encoding into prefix code words. Due to this fact, a pattern recognition procedure amounts to decoding a code word of the pattern by the nearest code word from a tree of the code words of selected templates. Efficient application of the pattern representation technique is illustrated by experimental results on signature and hand gesture recognition.

By applying recent results in optimization transfer, a new algorithm for kernel Fisher Discriminant Analysis is provided that makes use of a non-smooth penalty on the coefficients to provide a parsimonious solution. The algorithm is simple, easily programmed and is shown to perform as well as or better than a number of leading machine learning algorithms on a substantial benchmark. It is then applied to a set of extreme small-sample-size problems in virtual screening where it is found to be less accurate than a currently leading approach but is still comparable in a number of cases.

Bernoulli mixture models have been recently proposed as simple yet powerful probabilistic models for binary images in which each image pattern is modelled by a different Bernoulli prototype (component). A possible limitation of these models, however, is that usual geometric transformations of image patterns are not explicitly modelled and, therefore, each natural transformation of an image pattern has to be modelled using a different, prototype. In this work, we propose a simple technique to make these rigid prototypes more flexible by explicit modelling of invariances to translation, scaling and rotation. Results are reported on a task of handwritten Indian digits recognition.

Pedestrian detection from images of the visible spectrum is a high relevant area of research given its potential impact in the design of pedestrian protection systems. There are many proposals in the literature but they lack a comparative viewpoint. According to this, in this paper we first propose a common framework where we fit the different approaches, and second we use this framework to provide a comparative point of view of the details of such different approaches, pointing out also the main challenges to be solved in the future. In summary, we expect this survey to be useful for both novel and experienced researchers in the field. In the first case, as a clarifying snapshot of the state of the art; in the second, as a way to unveil trends and to take conclusions from the comparative study.