CUDA架构下的快速Wallis影像增强算法  

Fast Wallis image enhancement algorithm with CUDA

作　　者：肖汉[1,2] 吴庆双[1,3] 冯娜 

机构地区：[1]武汉大学遥感信息工程学院,武汉430079 [2]郑州师范学院信息技术系,郑州450044 [3]安徽师范大学国土资源与旅游学院,安徽芜湖241000 [4]郑州市档案馆,郑州450007

出　　处：《沈阳工业大学学报》2011年第3期293-298,共6页Journal of Shenyang University of Technology

基　　金：国家自然科学基金资助项目(40771177);国家高技术研究与发展计划(863)资助项目(2006AA12Z136);河南省高等学校青年骨干教师计划资助项目(2009GGJS-167)

摘　　要：针对图像增强通常需要较大的计算量、用传统方法难于进行实时处理的问题,提出了一种基于图形处理器加速的Wallis变换影像增强方法.借助于图形处理器较强的运算能力,利用CUDA并行计算架构在PC机上实现了快速Wallis图像滤波算法,包括图形处理器(GPU)上任务分解、大规模计算核心的分解方法,结合使用共享存储器、全局存储器对算法进行加速,使用线程块内的共享存储器较好地解决了同一计算子空间的各线程同步问题.对比了CPU和GPU计算Wallis影像变换的时间,结果表明,随着图像分辨率的增大,Wallis并行算法可以把计算速度提高40倍.该方法具有较好的实时性,可大大提高图像增强过程的处理速度,显著地减少了计算时间.For the problem that the image enhancement needs the huge computation quantity and the real-time processing is difficult to be implemented using the traditional processing methods, a Wallis transform image enhancement method based on graphic processing unit （GPU） acceleration was proposed. With the help of the strong computing ability of GPU and the parallel computing architecture of compute unified device architecture （CUDA）, the fast Wallis image filter algorithm was implemented on a personal computer. The method of large scale thread division was put forward along with the task division on GPU. The algorithm was accelerated with both shared memory and coalesced global memory. Various threads for the same computing subspace were properly synchronized by using the shared memory in thread block. The Wallis image transformation time using CPU and GPU was compared. The results show that the WaUis parallel algorithm can increase the computing speed by 40 times as the image resolution increases. The method is excellent in real time processing. It can accelerate the image enhancement process and reduce the computing time significantly.

关 键 词：图形处理器 统一计算设备架构 单指令多线程 Wallis变换 影像增强 CUDA核 并行 滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] P237[自动化与计算机技术—计算机科学与技术]