The Research And Implementation On Large Data Sets Volume Rendering Based On GPU

Direct volume rendering of large volume data sets on programmable graphics hardware is often limited by the amount of available graphics memory and the bandwidth from main memory to graphics memory. CVR(compressed volume rendering) has been shown to be an effective solution to avoid frequently transferring data between memory and GPU. CVR is a general approach for combining volume compression and volume rendering such that the decompression is coupled to rendering. In this paper, we focus on the large-scale data (seismic data, etc.) compression volume rendering on GPU. The main content and innovative points of this paper can be outlined as the following aspects. First of all, design and implement an efficient large-scale data compression algorithm. And then, decompress the compressed data in real-time on GPU and use FBO (Frame Buffer Object) and three-dimensional texture mapping to speed up rendering. Finally, design and implement a flexible and compatible seismic data processing system and integrate the compression volume rendering algorithm into it.Firstly, one of the challenges of volume rendering on the GPU is the comparatively limited amount of available memory due to the large disparity between memory capacity and volume sizes. To address this problem, an efficient large-scale volume data compression algorithm based on VQ is presented. First of all, volume data is divided into smaller regular blocks and each block is classified according as whether its average gradient value is zero or not. Then, blocks with non-zero gradient values are decomposed into a three hierarchical representation, for each of the top two levels, a splitting scheme based on a principal component analysis is adopted to find an initial codebook. Consequently, LBG algorithm is used for codebook refinement and quantization of the down-sampled blocks in those two levels. Blocks in the lowest level of the hierarchical representation and those with zero average gradient values are quantized with fixed number of bits. Experimental results show that, in the premise of good fidelity, the presented algorithm can obtain more than 50 times compression rate with a higher decoding speed.Secondly, today's graphics hardware equipped with GPU as the the core of programmable vertex shader and programmable pixel shaders can provide real-time volume rendering support. In order to solve the more and more serious conflicts between the user's demands of real-time, interactive and the low speed of the volume rendering, let GPU decompress the efficient compressed vector quantization data, and then use the FBO (Frame Buffer Object) technology and three-dimensional texture mapping method to speed up rendering. Experimental results show that in the premise of good fidelity and without any other acceleration strategy, the rendering speed has increased slightly.Finally, the three-dimensional and high-dimensional seismic data real-time visualization technology is a hot research area. Design and implement a seismic data processing system. In order to solve the diversity of formats of seismic data as well as the pluggable of variety of pre-processing of seismic data algorithm, the PROXY, FASCADE and other design patterns are applied to this system to make it more flexible and compatible, and at the same time provide an unified interface to reduce the degree of coupling to other modules. The compression volume rendering algorithm can be easily integrated to the seismic data processing system, the experimental results show that the method can speed up the rendering of multi-individual data in multi-window.