DirectCompute加速图像处理方法的研究  

DirectCompute method to accelerate image processing

作　　者：张益男[1] 袁杰[2] 

机构地区：[1]南京大学金陵学院信息科学与工程学院,江苏南京210089 [2]南京大学电子科学与工程学院,江苏南京210093

出　　处：《现代电子技术》2012年第22期55-58,共4页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(60501003)

摘　　要：当代图像处理,特别是基于大量密集科学计算的数据处理,需要用高度的并行处理技术来解决计算效率问题。然而处理速度更多依赖于具体的算法和系统实现间的相互联系,这就对进行并行处理的工作人员的并行技术及相关经验提出了一定的要求。为完成并行处理并加大并行计算效益,采用了DirectCompute方法。针对工程中的角点匹配计算时延问题,凭借有效的硬件资源,对其进行了并行处理,提高了程序执行效率,并缩短了计算延时。Contemporary image processing, especially data processing based on a large number of intensive scientific com- puting, needs a high degree parallel processing technology to improve the computational efficiency. However, processing speed is more dependent on interrelation between specific algorithms and system implementation, which puts forward the re- quirements on parallel technology and relevant experience ot＂ the workers who are doing parallel processing. To complete paral- lel processing and make the parallel effectiveness highest, DirectCompute method was used. Relyed on the effective hardware resources, parallel processing was adopted to solve the calculation delay problem of corner matching in projects. It improved the program execution efficiency and shortened the calculation delay.

关 键 词：图像三维重建 并行计算 GPU DirectCompute 

分 类 号：TN919-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]