基于GPU的彩色显微光学图像快速三维可视化  

GPU-BASED 3D VISUALISATION OF COLOURFUL MICROSCOPIC OPTICAL IMAGE

作　　者：韩超[1,2] 夏威[1] 高欣[1] 

机构地区：[1]中国科学院苏州生物医学工程技术研究所医学影像室,江苏苏州215163 [2]南京理工大学电子工程与光电技术学院,江苏南京210094

出　　处：《计算机应用与软件》2016年第12期145-147,共3页Computer Applications and Software

基　　金：国家自然科学基金青年科学基金项目(61205183);国家重大科研装备研究项目

摘　　要：彩色显微光学图像三维可视化计算量较大,针对基于CPU单线程串行计算的可视化方法无法满足实时显示要求,提出一种基于图形处理器(GPU)的显微彩色图像快速三维可视化算法。该方法采用最大密度投影函数(MIP)实现可视化,通过插值计算通过物体光线上等间距点的RGB值,取其最大亮度点的RGB作为该光线对应像素的颜色值。以上过程通过构筑的内核函数在GPU上以多线程方法完成,最终使用Open GL直接绘制投影图像。利用激光共聚焦显微镜获得的小鼠肾细胞彩色图像和多层细胞样本进行算法验证。实验结果表明,与基于CPU的单线程串行计算方法相比,基于GPU的可视化方法在显示效果一致的前提下,计算速度提高了90倍。该方法极大提升了显微图像处理过程中的实时显示性能。The computation of 3 D visualisation for colour microscopic optical images is relatively large. The visualising scheme based on CPU single-thread serial computation cannot satisfy the requirement of real-time display. This paper proposes a GPU-based fast 3D visualisation algorithm of colour microscopic optical images. The method utilises the function of maximum intensity projection （MIP） to realise visualisation. By using interpolation calculation to calculate the RGB values of the equal-interval points on the light through the object, it chooses the RGB value of the point with maximum brightness as the colour value of the corresponding pixel of light. The above-mentioned process is completed by multi-thread execution through the built kernel function on GPU. Finally, the projection image is sketched directly by OpenGL. The colour image of mice＇ s kidney cells and multi-layer cell samples pictured by confocal laser scanning microscopy are employed to validate this algorithm. Experimental result turns out that： compared with the visualisation means based on CPU single-thread serial computation, on the condition of the same performance on displaying quality, the GPU-based visualisation method accelerates the computation speed by 90 times. This method significantly improves the performance of real-time display during microscopic image processing.

关 键 词：彩色显微光学图像 三维可视化 最大密度投影 CUDA GPU并行计算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]