基于CUDA的Scharr算子并行化研究
Research on Parallelization of Scharr Operators Based on CUDA
张晗 1李芳1
作者信息
摘要
传统边缘检测算子采用逐行求梯度的方法顺序进行,遇到图像尺寸大或计算速度高效时,较难胜任这类计算密集度高的需求.文章从并行化角度对Scharr算子进行设计,采用CUDA语言对二维数据并行计算上进行算法优化,提出了多线程块偏移计算的设计思路,同时采取流处理的方式缩短传输开销.实验结果表明,与传统Scharr算子相比,在7 000×7 000尺寸图像识别上呈现了高效的识别速度,加速比提高了300倍左右,有较高的应用价值.
Abstract
Traditional edge detection operators use the method of seeking gradients row by row in order,which makes it difficult to meet the high computational density requirements of large image sizes or high computational speed.This paper designs the Scharr operator from the perspective of parallelization,optimizes the algorithm on two-dimensional data parallel computing using CUDA language,proposes a design idea for multithreaded block offset calculation,and adopts stream processing to reduce transmission overhead.Experiment results show that compared with the traditional Scharr operator,it exhibits efficient recognition speed in image recognition of sizes on 7 000×7 000,with an acceleration ratio increased by about 300 times,and has high application value.
关键词
边缘检测/Scharr算子/并行计算Key words
edge detection/Scharr operator/parallel computing引用本文复制引用
基金项目
2020年自治区创新环境建设专项(2020D01C001)
出版年
2024
国家科技期刊平台
NSTL
万方数据