Catálogo de publicaciones - libros

This paper proposes a new steganography algorithm for 3D models using a spatial subdivision technique. Our algorithm first decomposes the bounding volume of the cover model into voxels based on a Binary Space Partitioning (BSP) tree. The voxels are then further categorized into eight subspaces, each of which is numbered and represented as three-digit binary characters. In the embedding process, we first traverse the BSP tree, locating a leaf voxel; then we embed every three bits of the payload message into the vertex inside the leaf voxel. This is realized by translating a vertex’s current position to the corresponding numbered subspace. This technique is a substitutive blind extraction scheme, where messages embedded can be extracted without the aid of the original cover model. This technique achieves high data capacity, equivalent to at least three times the number of the embedded vertices in the cover model. In addition, the stego model has insignificant visual distortion. Finally, this scheme is robust against similarity transformation attacks.

The existing Collaborative Virtual Environment (CVE) systems have limited scalability due to the constraints in computer processing power and network bandwidth of participating hosts. In this paper, we propose a new approach for heterogeneous internet based users to construct large-scale CVE (LCVE) system: Mobile Agent Based Framework for LCVE (MACVE). In MACVE, the system workloads are decomposed into independent and fine grained tasks. The tasks are modelled as mobile agents which are not bound to any fixed nodes as the traditional CVE architectures do. As the mobile agents can migrate or clone dynamically at any suitable participating host include traditional servers and qualified user hosts, the system workloads can be distributed more pervasively to avoid potential bottleneck. This improves the scalability of LCVE. Experiments results have demonstrated the scalability of our proposed approach.

Automatic generation of tiling patterns with the symmetry of the extended Picard group in three-dimensional hyperbolic space is considered. We generate the patterns by repeating the fundamental patterns created in the fundamental region to all other equivalent regions. We also produce such a kind of tiling patterns in the unit sphere by conformal mappings. The method provides a novel approach for devising exotic symmetric tiling patterns from a dynamical system’s point of view.

This paper proposes a keyframe extraction method based on a novel layered curve simplification algorithm for motion capture data. Bone angles are employed as motion features and keyframe candidates can be selected based on them. After that, the layered curve simplification algorithm will be used to refine those candidates and the keyframe collection can be gained. To meet different requirements for compression and level of detail of motion abstraction, adaptive extraction parameters are also applied. The experiments demonstrate that our method can not only compress and summarize the motion capture data efficiently, but also keep the consistency of keyframe collection between similar human motion sequences, which is of great benefit to further motion data retrieval or editing.

The genome is the gene complement of an organism and it comprises the information of the entire genetic material of an organism. Many researchers use the whole genome alignment method to detect a genomic meaning between genomes. In this paper, we introduce a new method for whole genome alignment with LOD(Level-of-Detail) representation. It helps us to understand a relationship between two genomes and determine candidate sets from the whole genome alignment result.

We present a data hiding algorithm for 3D models. It is based on a substitutive procedure in the spatial domain. We propose a Virtual Multi-Level Embed Procedure to embed information based on shifting the message point by its virtual geometrical property, the order of which is assigned by principal component analysis. We have defined and validated an effective metric of distortion anticipation, which can help us easily anticipate and control the distortion rate. Experimental results show that the proposed technique is efficient and secure, has high capacity and low distortion, and is robust against affine transformations. It provides a reversible method and has proven to be feasible in data hiding.

Mesh chartification is an important tool for processing meshes in various applications. In this paper, we present a novel feature sensitive mesh chartification technique that can handle huge meshes with limited main memory. Our technique adapts the mesh chartification approach using Lloyd-Max quantization to out-of-core processing. While the previous approach updates chartification globally at each iteration of Lloyd-Max quantization, we propose a local update algorithm where only a part of the chartification is processed at a time. By repeating the local updates, we can obtain a chartification of a huge mesh that cannot fit into the main memory. We verify the accuracy of the serialized local updates by comparing the results with the global update approach. We demonstrate that our technique can successfully process huge meshes for applications, such as mesh compression, shape approximation, and remeshing.

In this paper, we propose a new method for simulating reactive motions for motion capture animation. The goal is to generate realistic behaviors under unexpected external forces. A set of techniques are introduced to select a motion capture sequence which follows an impact, and then synthesize a believable transition to this found clip for character interaction. Utilizing a parallel simulation, our method is able to predict a character’s motion trajectory under dynamics, which ensures that the character moves towards the target sequence and makes the character’s behavior more life-like. In addition, the mechanism of parallel simulation with different time steps is flexible for simulation of multiple contacts in a series when multiple searches are necessary. Our controller is designed to generate physically plausible motion following an upcoming motion with adjustment from biomechanics rules, which is a key to avoid an unconscious look for a character during the transition.

This paper presents a purely hardware-accelerated shadow volume algorithm for subdivision surface based models. By introducing SP (subdivision patterns), all procedures, including subdivision evaluation, silhouette extraction, shadow volume generation, and shadow rendering are executed on GPU (Graphics Process Units) efficiently. This not only alleviates the burden of CPU, but also guarantees the consistency of data among different processing stages. This also makes it possible to integrate some special effects imposed by other shaders, e.g., displacement mapping or vertex texturing, with the shadow volume algorithm. Experiments show that the algorithm is efficient, robust, and can be easily extended to other subdivision schemes and parametric surfaces.

A model-based method is proposed in this paper for 3-dimensional human motion recovery, taking un-calibrated monocular data as input. This method is designed to recover smooth human motions with high efficiency, while its outputs are guaranteed to resemble the original motion not only from the same viewpoint the sequence was taken, but also look natural and reasonable from any other viewpoint. The proposed method is called “Motion trend prediction (MTP)”. To evaluate the accuracy of the MTP, it is first tested on some “synthesized” input. After that experiments are conducted on real video data, which demonstrate that the proposed method is able to recover smooth human motions from their 2D image features with high accuracy.