Catálogo de publicaciones - libros

This paper presents a framework for efficient content-based motion retrieval. To bridge the gap between user’s vague perception and explicit motion scene description, we propose a Scene Description Language that can translate user’s input into a series of set operations between inverted lists . Our Scene Description Language has three-layer structures, each describing scenes at different levels of granularity. By introducing automatic transition strategy into our retrieval process, our system can search motions that do not exist in a motion database. This property makes our system have potentials to serve as motion synthesis purpose. Moreover, by using various kinds of qualitative features and adaptive segments of motion capture data stream, we obtain a robust clustering that is flexible and efficient for constructing motion graph . Some experimental examples are given to demonstrate the effectiveness and efficiency of proposed algorithms.

Direct volume rendering (DVR) is an effective way to visualize 3D vascular images for diagnosis of different vascular pathologies and planning of surgical treatments. Angiograms are typically noisy, fuzzy, and contain thin vessel structures. Therefore, some kinds of enhancements are usually needed before direct volume rendering can start. However, without visualizing the 3D structures in angiograms, users may find it difficult to select appropriate parameters and assess the effectiveness of the enhancement results. In addition, traditional enhancement techniques cannot easily separate the vessel voxels from other contextual structures with the same or very similar intensity. In this paper, we propose a framework to integrate enhancement and direct volume rendering into one visualization pipeline using multi-dimensional transfer function tailored for visualizing the curvilinear and line structures in angiograms. Furthermore, we present a feature preserving interpolation method to render very thin vessels which are usually missed using traditional approaches. To ease the difficulty in vessel selection, a MIP-guided method is suggested to assist the process.

Most existing techniques of foreground extracting work only in interactive mode. This paper introduces a novel algorithm of automatic foreground extraction for special object, and verifies its effectiveness with head shoulder images. The main contribution of our idea is to make the most use of the prior knowledge to constrain the processing of foreground extraction. For human head shoulder images, we first detect face and a few facial features, which helps to estimate an approximate mask covering the interesting region. The algorithm then extracts the hard edge of foreground from the specified area using an iterative graph cut method incorporated with an improved Gaussian Mixture Model. To generate accurate soft edges, a Bayes matting is applied. The whole process is fully automatic. Experimental results demonstrate that our algorithm is both robust and efficient.

The visualization of 3D volume data of proteins synthesized by quantum mechanics is a new topic and is of great importance in modern bio-computing. In this paper, we introduce our primary attempts on the volume visualization of the 3D macro-molecular scalar field. Firstly, we transform one protein molecular structure into a regularly sampled 3D scalar field according to the theories in quantum chemistry, in which each node records the combined effect of different actions in protease. We then exploit volume rendering techniques to find the macro-structure inside the data field based on a convenient mapping mechanism. We also propose an improved transfer function mode, facilitating the flexible visualization of the 3D protein data sets. Finally, combined with the iso-surface extraction technique, our approach allows for interactive exploration of the potential “tunnel” region which exhibits biological sense. With our approach, we show the escape route of water molecules hidden in the HIV-1 protease, which conforms to the experimental results.

A second order forward differences based subdivision depth computation technique for extra-ordinary Catmull-Clark subdivision surface (CCSS) patches is presented. The new technique improves a previous technique in that the computation of the subdivision depth is based on the patch’s curvature distribution, instead of its dimension. Hence, with the new technique, no excessive subdivision is needed for extra-ordinary CCSS patches to meet the precision requirement and, consequently, one can make trimming, finite element mesh generation, boolean operations, and tessellation of CCSSs more efficient.

This paper presents an approach for embedding regular analytic shapes within subdivision surfaces. The approach is illustrated through the construction of compound Spherical-Catmull-Clark subdivision surfaces. It starts with a subdivision mechanism that can generate a perfect sphere. This mechanism stems from the geometric definition of the sphere shape. Thus, it comes with a trivial proof that the target of the construction is what it is. Furthermore, the similarity of this mechanism to the Catmull-Clark subdivision scheme is exploited to embed spherical surfaces within Catmull-Clark Surfaces, which holds a great potential for many practical applications.

We present a novel adaptive radial basis function network to reconstruct smooth closed surfaces and complete meshes from non-uniformly sampled noisy range data. The network is established using a heuristic learning strategy. Neurons can be inserted, removed or updated iteratively, adapting to the complexity and distribution of the underlying data. This flexibility is particularly suited to highly variable spatial frequencies, and is conducive to data compression with network representations. In addition, a greedy neighbourhood Extended Kalman Filter learning method is investigated, leading to a significant reduction of computational cost in the training process with desired prediction accuracy. Experimental results demonstrate the performance advantages of compact network representation for surface reconstruction from large amount of non-uniformly sampled incomplete point-clouds.

There are a number of applications in computer graphics and computer vision that require the accurate estimation of normal vectors at arbitrary vertices on a mesh surface. One common way to obtain a vertex normal over such models is to compute it as a weighted sum of the normals of facets sharing that vertex. But numerical tests and asymptotic analysis indicate that these proposed weighted average algorithms for vertex normal computation are all linear approximations. An open question proposed in [ CAGD,17:521-543, 2000 ] is to find a linear combination scheme of the normals of the triangular faces, based on geometric considerations, that is quadratic convergence in the general mesh case. In this paper, we answer this question in general triangular mesh case. When tested on a few random mesh with valence 4, the scheme proposed by this paper is of second order accuracy, while the existing schemes only provide first order accuracy.

In virtual colonoscopy, it is crucial to generate the camera path rapidly and accurately. Most of the existing path generation methods are computationally expensive since they require a lengthy preprocessing step and the 3D positions of all path points should be generated. In this paper, we propose a visibility-based automatic path generation method by emulating the ray propagation through the conduit of the colon. The proposed method does not require any preliminary data preprocessing steps, and it also dramatically reduces the number of points needed to represent the camera path using control points. The result is a perceivable increase in computational efficiency and easier colon navigation with the same level of accuracy.

In this paper a computational model called dynamic medial axis ( $\mathcal{DMA}$ ) is proposed to describe the internal evolution of planar shapes. To define the $\mathcal{DMA}$ , a symbolic representation called matching list is proposed to depict the topological structure of the medial axis. As shown in this paper with provable properties, the $\mathcal{DMA}$ exhibits an interesting dynamic skeleton structure for planar shapes. Finally an important application of the proposed $\mathcal{DMA}$ — computing the medial axis of multiply-connected planar shapes with curved boundaries — is presented.