码书自适应的大规模体数据矢量量化算法

A CODEBOOK ADAPTIVE VECTOR QUANTISATION ALGORITHM FOR LARGE SCALE VOLUME DATA

针对大规模体数据矢量量化VQ(Vector Quantisation)编码时间长的问题,提出一种基于图形处理器的码书自适应的大规模体数据矢量量化算法。该算法首先提取原始体数据密度分布特征,据此选取合适的初始码书生成算法,将矢量数据分批先后载入图形处理器进行并行计算,每读入一批数据,根据该批数据的码准值对第一批数据产生的码书进行优化及扩充,随后完成该批数据的编码。实验结果表明,该算法提高了图像的编码速度及还原质量,明显缩短了图像的压缩时间,同时保证了体数据重构质量。

To solve the problem of time costing in encoding large scale volume data with vector quantisation method,we propose an algorithm of GPU-based adaptive codebook vector quantisation（ACVQ） for large scale volume data.The algorithm first extracts the density distribution feature of the original volume data,and hereby selects the appropriate initial codebook generation algorithm,then it loads the vector data into GPU successively by batch to carry out parallel computation.For each loading,we optimise and expand the first codebook generated by first batch according to the code reference of the data of that batch,followed by completing the data encoding of this batch.Experimental results suggest that ACVQ improves the encoding speed and the restored quality of image,evidently shorten the compression time of image,at the same time it ensures the reconstruction quality of volume data.