基于图形处理器加速的医学图像分割算法研究  

Image Segmentation Method of Gibbs Random Field Accelerated by GPU

作　　者：程广斌[1] 马承华[1] 郝立巍[2] 

机构地区：[1]武警广东总队医院,广州510507 [2]南方医科大学生物医学工程学院,广州510515

出　　处：《医疗卫生装备》2008年第2期6-9,共4页Chinese Medical Equipment Journal

基　　金：国家重点基础研究发展计划(编号:2003CB716104)

摘　　要：目的:提出对基于显卡图形处理器(GPU)加速的吉布斯(Gibbs)随机场医学图像分割算法的改进。方法:利用片元作色器并行执行的逐点计算,取代了CPU串行执行的逐点计算,大大减小CPU的负载,处理效率高于单独采用CPU的效率。结果:改进后的分割算法的加速性较CPU的串行计算方式减小了CPU的运算量,效率得以提高,计算速度较原算法有明显改善(计算速度提高400%以上)。结论:采用显示卡加速的基于Gibbs随机场的模糊C均值分割算法运算接近实时,大大提高了Gibbs随机场分割算法在临床的实用性。Objective To propose an improved C-means segment method based on Gibbs random field accelerated by GPU. Methods The parallel computation of pixel shades was used to take the place of the classical point-by-point method of CPU. By this way, the efficiency was higher than merely using the CPU computation. Results The efficiency of computation was improved over 400%. The load of CPU was reduced and the effect of accelerator was obvious. Conclusion The improved C-means segment method based on Gibbs random field accelerated by GPU enhances the clinical application of image segmentation, the computer rate of which is improved distinctly and closely to real time.

关 键 词：图像分割 GIBBS随机场 图形处理器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]