Exploiting the GPU Power for Intensive Geometric and Imaging Data Computation
Exploiting the GPU Power for Intensive Geometric and Imaging Data Computation
Jianqing Wang

Computer graphics, imaging and visualization nowadays requires dealing with and performing various analysis on large amount of data. For example, deformation and skeletal animation of complex geometries in real-time, multiple-level discrete wavelet transform on high resolution imaging data, etc. These operations usually are computational intensive and impose a heavy burden on the CPU, which are also hard to achieve real-time performance. On the other hand, with the recent advances in consumer-level graphics hardware, the current new generation of graphics accelerator now consists of a graphics processing unit (GPU) which offers SIMD-based parallel processing power. It doesn't merely do the job of rendering texture-mapped polygons, but also provide us with high-precision rendering pipeline and high programmability nowadays. We can perform other general purpose computing on it when carefully designed. In this thesis, we have successfully exploited the power of GPU for computation and processing of both geometric data and imaging data in the 2 applications -- real-time character facial and skeletal animation; and multiple levels of discrete wavelet transform. We develop parallel algorithms to map our solutions to the GPU SIMD architecture and leverage its power to give a great performance gain in both applications. These approaches also offload the intensive computing tasks from CPU and achieve load balancing between CPU and GPU. At the same time, for the virtual character project, we also propose a simple and efficient framework for both facial and skeletal animation, as well as the seamless integration of them for rendering. The solutions for supporting multilingual lip synchronization and integration with text-to-speech system of multiple languages are also developed.

Related publication:


Home
Copyright 1996-2012 Tien-Tsin Wong. All rights reserved.