Catálogo de publicaciones - libros

In this paper we present a novel approach for mimicking expressions in 3D from a monocular video sequence. To this end, first we construct a high resolution semantic mesh model through automatic global and local registration of a low resolution range data. Such a model is represented using predefined set of control points in a compact fashion, and animated using radial basis functions. In order to recover the 2D positions of the 3D control points in the observed sequence, we use cascade Adaboost-driven search. The search space is reduced through the use of predictive expression modeling. The optimal configuration of the Adaboost responses is determined using combinatorial linear programming which enforces the anthropometric nature of the model. Then the displacement can be reproduced on any version of the model, registered on another face. Our method doesn’t require dense stereo estimation and can then produce realistic animations, using any 3D model. Promising experimental results demonstrate the potential of our approach.

In this paper we propose a novel approach for visual object recognition. The main idea is to consider the object recognition task as an process which is guided by multi-cue attentional indexes, which at the same time correspond to object’s parts. In this method, a mechanism is carried out. It does not correspond to a different stage (or module) of the recognition process; on the contrary, it is inherent in the recognition strategy itself. Recognition is achieved by means of a sequential search of object’s parts: parts selection depends on the current state of the recognition process. The detection of each part constraints the process state in order to reduce the search space (in the overall feature space) for future parts matching. As an illustration, some results for face and pedestrian recognition are presented.

This paper presents a robust segmentation method based on the integrated squared error or estimation (). Formulated under the Finite Gaussian Mixture (FGM) framework, the new model (FGML2E) has a strong discriminative ability in capturing the major parts of intensity distribution without being affected by outlier structures or heavy noise. Comparisons are made with two popular solutions, the and algorithms, and the experimental results clearly show the improvement made by our model.

In this paper, we systematically examine two fundamental questions in information visualization – how to define effective visualization and how to measure it. Through a literature review, we point out that the existing definitions of effectiveness are incomplete and often inconsistent – a problem that has deeply affected the design and evaluation of visualization. There is also a lack of standards for measuring the effectiveness of visualization as well as a lack of standardized procedures. We have identified a set of basic research issues that must be addressed. Finally, we provide a more comprehensive definition of effective visualization and discuss a set of quantitative and qualitative measures. The work presented in this paper contributes to the foundational research of information visualization.

We present a design methodology for automatic machine vision application aiming at detecting the size ratio of tobacco leaves which will be feedback to adjust running parameters of manufacture system. Firstly, the image is represented by Markov Random Field(MRF) model which consists of a label field and an observation field. Secondly, according to Bayes theorem, the segmentation problem is translated into Maximum a Posteriori(MAP) estimation of the label field and the estimation problem is solved by Iterated Conditional Model(ICM) algorithm. Finally we give the setup of the inspection system and experimented with a real-time image acquired from it, the experiment shows better detection results than Otsu’s segmentation method especially in the larger leaf regions.

In this paper, a hierarchical shape decomposition algorithm is proposed, which integrates the advantages of skeleton-based and minima-rule-based meaningful segmentation algorithms. The method makes use of new geometrical and topological functions of skeleton to define initial cutting critical points, and then employs salient contours with negative minimal principal curvature values to determine natural final boundary curves among parts. And sufficient experiments have been carried out on many meshes, and shown that our framework can provide more reasonable perceptual results than single skeleton-based [8] or minima-rule-based [15] algorithm. In addition, our algorithm not only can divide a mesh of any genus into a collection of genus zero, but also partition level-of-detail meshes into similar parts.

Many computer graphics rendering algorithms and techniques use ray tracing for generation of natural and photo-realistic images. The efficiency of the ray tracing algorithms depends, among other techniques, upon the data structures used in the back ground. -trees are some of the most commonly used data structures for acceler ating ray tracing algorithms. Data structures using cost optimization techniques based upon Surface Area Heuristics (SAH) are generally considered to be best and of high quality. During the last decade, the trend has been moved from off-line rendering towards real time rendering with the introduction of high speed computers and dedicated Graphical Processing Units (GPUs). In this situation, SAH-optimized structures have been considered too slow to allow real-time rendering of complex scenes. Our goal is to demonstrate an accelerated approach in building SAH-based data structures to be used in real time rendering algorithms. The quality of SAH-based data structures heavily depends upon split-plane locations and the major bottleneck of SAH techniques is the time consumed to find those optimum split locations. We present a parabolic interpolation technique combined with a golden section search criteria for predicting -tree split plane locations. The resulted structure is 30% faster with 6% quality degradation as compared to a standard SAH approach for reasonably complex scenes with around 170k polygons.

We present our work on . Every radiance sample in space has a location and a direction from which it is received. These degrees of freedom make up a phase space. The rendering problem of generating a discrete image from single radiance values is reduced to reconstruct a continuous radiance function from sparse samples in its phase space. The problem of reconstruction in a sparsely sampled space is solved by utilizing scattered data interpolation (SDI) methods. We provide numerical and visual evaluations of experiments with three SDI methods.

We propose an algorithm to produce automatically a 3-D CAD model from a set of range data, based on non-uniform rational B-splines (NURBS) surface fitting technique. Our goal is to construct automatically continuous geometric models, assuming that the topology of the surface is unknown. In the propose algorithm, the triangulated surface is partitioned in quadrilateral patches, using Morse theory. The quadrilateral regions on the mesh are then regularized using geodesic curves and B-splines to obtain an improved smooth network on which to fit NURBS surfaces. NURBS surfaces are fitted and optimized using evolutionary strategies. In addition, the patches are smoothly joined guaranteeing continuity. Experimental results are presented.

This paper presents a technique that allows test engineers to visually analyze and explore within memory chip test data. We represent the test results from a generation of chips along a traditional and a . We also show correspondences in the test results across multiple generations of memory chips. We use simple geometric “glyphs” that vary their spatial placement, color, and texture properties to represent the critical attribute values of a test. When shown together, the glyphs form visual patterns that support exploration, facilitate discovery of data characteristics, relationships, and highlight trends and exceptions in the test data that are often difficult to identify with existing statistical tools.