摘要
现有的图像去雾算法已经能取得较好的去雾效果,但其算法的时间复杂度较高.针对暗通道先验去雾算法不能实时去雾处理的问题,提出对单幅图像的去雾算法做基于OpenCL的并行化优化.在并行优化中,首先,传统算法中大气光值的求解并不适合做并行计算,且图像中白色物体会干扰大气光值的选取,改进的算法能有效的并行执行和准确的选取大气光值;其次,通过尽可能减少内核数量并利用局部内存,降低内存的频繁拷贝和利用高速内存,从而实现对数据的快速读写.最后经过对算法的并行优化,实验对1280×720尺寸图像的处理时间为20ms,实现了实时去雾处理.
The existing image dehazing algorithm can achieve better defogging effect,but the algorithm has higher time complexity. For the problem that the dark channel prior dehazing algorithm can not deal with the fog in real time,the dehazing algorithm for a single image is proposed to be based on OpenCL with parallel optimization. In parallel optimization,first of all,the solution of atmospheric light values in traditional algorithms is not suitable for parallel computing,and white objects in the image interfere with the selection of atmospheric light values. The improved algorithm can effectively perform parallel execution and accurately select atmospheric light values. Secondly,by reducing the number of kernels as much as possible and using local memory,the frequent copying of memory and the use of high-speed memory are reduced,thereby realizing fast reading and w riting of data. Finally,after parallel optimization of the algorithm,the processing time for the 1280 × 720 size image is 20 ms,and real-time defogging processing is realized.
作者
刘明
路锦正
LIU Ming;LU Jin-zheng(School of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,China;Robot Technology Used for Special Environment Key Laboratory of Sichuan Province,Mianyang 621010,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2019年第4期845-850,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61601382
61401379)资助
四川省教育厅重点项目(15ZA0118)资助
特殊环境机器人技术四川省重点实验室开放基金项目(13zxtk0505)资助
西南科技大学博士基金项目(13zx7112)资助
关键词
去雾
优化
OPENCL
并行
实时
defogging
optimization
OpenCL
parallel
real-time
