基于压缩域的脑成像大数据体可视化方法

Volume Rendering Method of Mass Brain Imaging Data Based on Compression Domain

脑科学是当今国际科技研究的前沿邻域,而对高精度脑成像数据进行可视化是脑神经科学在结构成像方面的基础性需求。针对高精度脑成像数据可视化过程中存在的数据量大以及绘制效率低的问题,提出了基于分类分层矢量量化和完美空间哈希相结合的压缩域可视化方法。首先对体数据进行分块,记录每块的平均值并依据块内体数据的平均梯度值是否为0进行分类;其次运用分层矢量量化对平均梯度值不为0的块进行压缩;然后用分块完美空间哈希技术存储压缩得到两个索引值;最后对上面的压缩体数据进行解码得到恢复体数据,采用分块完美空间哈希对原始体数据与恢复体数据作差得到的残差数据进行压缩。绘制时,只需将压缩得到的数据作为纹理加载到GPU内,即可在GPU内完成实时解压缩绘制。实验结果表明,在保证较好图像重构质量的前提下,该算法减少了数据的存储空间,提高了体可视化的绘制效率,从而可以在单机上处理较大的数据。

Nowadays, brain science is the forefront field of international scientific and technological research, while the visualization of the high-precision brain imaging data is the fundamental requirements of the structural imaging of brain neuroscience. Aiming at the problems of great quantity of data and low efficiency during rendering the brain imaging da- ta,a compression domain visualization algorithm based on the combination of flag based classical hierarchical vector quantization and perfect spatial hashing was put forward. Firstly, the volume data is blocked, the average of each block is recorded and then the blocks are classfied according to their average gradient value. Secondly, the hierarchical vector quantization is used to compress the blocks of whose average gradient is not O. Thirdly, the perfect spatial hashing tech- nology based on blocking is used to store two index values obtained by compressing. Finally, the above compressed data is decompressed to obtain the recovered volume data, and then the perfect spatial hashing based on blocking is applied to compress the differential volume data obtained by making the original volume data minus the recovered volume data. When rendering, the compressed data is reloaded as textures to GPU, then decompression and visualization can be done in real time. The experiment results show that the algorithm reduces the data storage space and improves the compres- sion ratio and can make the single machine handle larger data under the premise of ensuring the better quality of image reconstruction.