Research And Application Of Compress Fusion Volume Rendering Based On GPU

Great advancements in commodity graphics hardware have favored GPU-based volume rendering as the main adopted solution for interactive exploration of rectilinear scalar volumes on commodity platforms. Nevertheless, long data transfer times and GPU memory size limitations are often the main limiting factors, especially for massive data. To address this issue, a variety of compression techniques have been introduced. In order to improve capabilities and performance over the entire storage and rendering pipeline, the encoding/decoding process is typically highly asymmetric, and systems should ideally compress at data production time and decompress on demand at rendering time. Lossless compression methods are applied to volume data rendering, among them, vector quantization is often chosen, because it works for GPU decompressed. But, for large volume data, it wastes a lot of time to get quality codebook, which affects its application.This thesis researches about the problems existing in application of compressed volume rending in geophysical exploration field. The major contributions of the present thesis are as follows:1. A codebook initialization technique based on data stream clustering use GPU is presented. The earthquake data size is usually more than the GPU storage space, right now, the basic idea of codebook initial use GPU is copy all data to GPU, then, make the codebook initialization algorithm execute in GPU, this is only use the high parallel process ability of GPU. If the data size larger than GPU storage space, we must copy data from CPU memory to GPU memory many times, this will result in extra time-consuming. Aiming at this problem, a codebook initialization technique based on data stream clustering use GPU is presented. Our idea is to split the data into chunks and train each chunk single. Then, the codebooks extracted from chunks are aggregated at the final stage. With our algorithm, the original data only copy from CPU memory to GPU memory once, so reduce the extra time-consuming. Experimental results show that the proposed method could solve the problem of codebook initialization for large seismic data compressed volume rendering.2. In this thesis, we combine multi-attributes fusion volume rendering technique with vector quantization. Multi-attributes fusion volume rendering technique can improve the SNR of the target. The traditional multi-attributes fusion first fusion the original data, and get a new volume data. Fusion volume rendering technology could interactive exploration of multi-attributes volume data. But, it is difficult to load more volume data into GPU memory at the same time. Aiming at this problem, we combine multi-attributes fusion volume rendering technique with vector quantization. Our idea is first to compress the original data with vector quantization, then, load the compressed files into GPU memory. Meanwhile we decompress the compressed files and fusion them in GPU while rending. The results show that this can make the data amount suitable for GPU memory size.3. Design and implement a memory management module for volume rendering simulation platform based on CUDA. This module provide two means to support large-scale seismic processing which are file format and management.