Catálogo de publicaciones - libros

A wide variety of systems require reliable personal recognition schemes to either confirm or determine the identity of an individual requesting their services. The purpose of such schemes is to ensure that the rendered services are accessed only by a legitimate user, and not anyone else. Examples of such applications include secure access to buildings, computer systems, laptops, cellular phones and ATMs. In the absence of robust person recognition schemes, these systems are vulnerable to the wiles of an impostor. Biometric recognition, or simply biometrics, refers to the automatic recognition of individuals based on their physiological and/or behavioral characteristics. By using biometrics it is possible to confirm or establish an individual’s identity based on who she is, rather than by what she possesses (e.g., an ID card) or what she remembers (e.g., a password). Although biometrics emerged from its extensive use in law enforcement to identify criminals, i.e., forensics, it is being increasingly used today to carry out person recognition in a large number of civilian applications (e.g., national ID card, e-passport and smart cards) [1],[2]. Most of the emerging applications can be attributed to increased security threats as well as fraud associated with various financial transactions (e.g., credit cards).

We compare three hierarchical structures, , , , that are used to steer a segmentation process in 3d voxel images. There is an important topological difference between and both others that we will study. A quantitative evaluation of the quality of the three segmentation techniques based on several hundred experiments is presented.

In this paper we present an unconventional image segmentation approach which is devised to meet the requirements of image understanding and pattern recognition tasks. Generally image understanding assumes interplay of two sub-processes: image information content discovery and image information content interpretation. Despite of its widespread use, the notion of “image information content” is still ill defined, intuitive, and ambiguous. Most often, it is used in the Shannon’s sense, which means information content assessment averaged over the whole signal ensemble. Humans, however, rarely resort to such estimates. They are very effective in decomposing images into their meaningful constituents and focusing attention to the perceptually relevant image parts. We posit that following the latest findings in human attention vision studies and the concepts of Kolmogorov’s complexity theory an unorthodox segmentation approach can be proposed that provides effective image decomposition to information preserving image fragments well suited for subsequent image interpretation. We provide some illustrative examples, demonstrating effectiveness of this approach.

Deformable contours are now widely used in image segmentation, using different models, criteria and numerical schemes. Some theoretical comparisons between some deformable model methods have already been published [1]. Yet, very few experimental comparative studies on real data have been reported. In this paper, we compare a levelset with a B-spline based deformable model approach in order to understand the mechanisms involved in these widely used methods and to compare both evolution and results on various kinds of image segmentation problems. In general, both methods yield similar results. However, specific differences appear when considering particular problems.

This paper generalizes the interactive method for region segmentation of grayscale images based on graph cuts by Boykov & Jolly (ICCV 2001) to colour and textured images. The main contribution lies in incorporating new functions handling colour and texture information into the graph representing an image, since the previous method works for grayscale images only. The suggested method is semi-automatic since the user provides additional constraints, i.e. s(he) establishes some seeds for foreground and background pixels. The method is steerable by a user since the change in the segmentation due to adding or removing seeds requires little computational effort and hence the evolution of the segmentation can easily be controlled by the user. The foreground and background regions may consist of several isolated parts. The results are presented on some images from the Berkeley database.

A novel distinguished region detector, complementary to existing approaches like Harris-corner detectors, Difference of Gaussian detectors (DoG) or Maximally Stable Extremal Regions (MSER) is proposed. The basic idea is to find distinguished regions by clusters of interest points. In order to determine the number of clusters we use the concept of maximal stableness across scale. Therefore, the detected regions are called: Maximally Stable Corner Clusters (MSCC). In addition to the detector, we propose a novel joint orientation histogram (JOH) descriptor ideally suited for regions detected by the MSCC detector. The descriptor is based on the 2D joint occurrence histograms of orientations. We perform a comparative detector and descriptor analysis based on the recently proposed framework of Mikolajczyk and Schmid, we present evaluation results on additional non-planar scenes and we evaluate the benefits of combining different detectors.

Importance of multi-spectral colour information has been remarkably increasing in imaging science. This is because the original spectrum contains much more information about the surface of target objects than perceived colour by human. This article describes our attempts to visualize the invisible information, such as the constituent distribution and internal microstructure of food and plant responses to the environmental stress, by a spectral imaging technique.

The problem of image segmentation can be formulated in the framework of Bayesian statistics. We use a Markov random field as the prior model of the spacial relationship between image pixels, and approximate an observed image by a Gaussian mixture model. In this paper, we introduce into the statistical model a hierarchical prior structure from which model parameters are regarded as drawn. This would give an efficient Gibbs sampler for exploring the joint posterior distribution of all parameters given an observed image and could make the estimation more robust.

Algorithms based on SAR images for the purpose of detecting illegal oil spill pollution in the marine environment are studied. This paper focus on the feature extraction step, aiming at identifying features that lead to significant improvements in classification performance compared to earlier reported results. Both traditional region descriptors, features tailored to oil spill detection and techniques originally associated with other applications are evaluated. Experimental results show an increase from 89% to 97% in the number of suspected oil spills detected.

Light fields are known for their potential in generating 3D reconstructions of a scene from novel viewpoints without need for a model of the scene.Reconstruction of novel views, however, often leads to ghosting artefacts, which can be relieved by correcting for the depth of objects within the scene using disparity compensation. Unfortunately, reconstructions from this disparity information suffer from a lack of information on the orientation and smoothness of the underlying surfaces. In this paper, we present a novel representation of the surfaces present in the scene using a planar patch approach.We then introduce a reconstruction algorithm designed to exploit this patch information to produce visually superior reconstructions at higher resolutions. Experimental results demonstrate the effectiveness of this reconstruction technique using high quality patch data when compared to traditional reconstruction methods.