基于压缩域的脑成像大数据体可视化方法

发布时间：2019-07-08 11:43

【摘要】：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。
[Abstract]:Brain science is the frontier neighborhood of international scientific and technological research, and visualization of high-precision brain imaging data is the basic requirement of brain neuroscience in structural imaging. In order to solve the problems of large amount of data and low rendering efficiency in the visualization of high precision brain imaging data, a compressed domain visualization method based on classification hierarchical vector quantification and perfect space hash is proposed. Firstly, the volume data is divided into blocks, the average value of each block is recorded and classified according to whether the average gradient value of the block inner volume data is 0; secondly, the block whose average gradient value is not 0 is compressed by hierarchical vector quantification, and then the block perfect space hash technique is used to store and compress the two index values. Finally, the above compressed volume data is decoded to get the restorer data, and the residual data obtained by the difference between the original volume data and the restorer data is compressed by using the block perfect space hash. When drawing, the compressed data can only be loaded into GPU as texture, and the real-time decompression rendering can be completed in GPU. The experimental results show that the algorithm reduces the storage space of the data and improves the rendering efficiency of volume visualization on the premise of ensuring the quality of image reconstruction, so that the larger data can be processed on a single computer.
【作者单位】：国防科学技术大学计算机学院;国防科学技术大学海洋科学与工程研究院;
【基金】：国家重点基础研究计划(973计划)项目:灵长类神经回路精细结构成像的新方法和新工具(2015CB755604)资助
【分类号】：R338;TP391.41

【相似文献】