摘要
针对Retinex图像增强算法中的计算密集性问题,提出了基于图形处理器GPU平台的单尺度Retinex(SSR)算法的并行加速方案.首先,简要介绍了SSR算法的基本原理;其次,根据SSR算法的并行性,利用计算统一设备架构(CUDA)软硬件体系架构,实现了SSR算法向GPU上的移植;结果表明,经过并行优化的SSR算法可到达较高的执行效率,并随着图像分辨率的增大加速比显著提高,最大加速比达到近90倍,具有实际应用价值.
In order to solve the problem of disadvantages of time-consuming calculations in the course of Retinex image enhancement,parallel acceleration strategies for Single-Scale Retinex(SSR)algorithm,based on GPU,are put forward.First,general idea of SSR is introduced,as well as the effcient architecture of GPU.Then,according to the parallelity of the algorithm,employing the CUDA(Compute Unified Device Architecture)user-friendly programming framework,SSR based on GPU was implemented to validate optimization.Experiment results indicate that the parallel optimized SSR can achieve high execution efficiency.The speedup improves significantly while the image size increases.The maxirmum speedup is up to 90X compared with CPU Implementation,which has a practical meaning.
作者
陈云善
盛磊
李一芒
高世杰
Chen Yunshan;Sheng Lei;Li Yimang;Gao Shijie(Changchun Institute of Optics,Fine Mechanics and Physics,Chinese Academy of Sciences,Changchun 130033,China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2015年第S01期189-193,共5页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(11403064)项目资助