【摘要】 采用可编程图形硬件对大规模体数据进行直接体绘制时常常受到图形卡容量的限制,导致数据在内存与显存之间频繁交换,从而成为绘制的瓶颈.为此,提出一种大规模体数据矢量量化压缩算法.首先对体数据分块,并依据块内数据平均梯度值是否为0对该块进行分类;然后用3层结构表示梯度值非0的块,对其中次高层和最高层采用基于主分量分析分裂法产生初始码书,用LBG算法进行码书优化和量化,而对最低层以及梯度值为0的块采用定比特量化.实验结果表明,在保证较好图像重构质量的前提下,该算法可获得50倍以上的压缩比和更快的解压速度.更多还原

【Abstract】 The size of large scale volume data sets to be visualized by direct volume rendering on programmable graphics hardware is often limited by the amount of available graphics memory,as it will lead to frequently data transfer between memory and GPU. To get rid of this limitation,an efficient large-scale volume data compression algorithm based on VQ is presented. The volume data set is first divided into smaller regular blocks and each block is classified according to whether its average gradient value is zero or not. Then,blocks with non-zero gradient values are re-organized into a three-level hierarchical representation. To the top two levels,a splitting scheme based on principal component analysis is applied to find their initial codebooks. LBG algorithm is then conducted for codebook refinement and quantization. Blocks in the lowest level and those with zero average gradient values are quantized with fixed number of bits. Experimental results show that,in addition to good fidelity,the presented algorithm can obtain more than 50 times compression rate with fast decoding speed.更多还原