Improvement and parallel implementation of canny edge detection algorithm based on GPU

摘要

Being an effective edge detector with single-pixel response, Canny operator has been widely used in accurately abstracting the edge information in image processing. However, taking its 4-step process into account, its real-time implementation based on CPU has become a significant problem, especially for the part of the edge tracing, which consumes a large amount of computing time. To solve the problem, GPU will be used in this paper for sake of its powerful ability of parallel processing while a new Canny operator is proposed with the introduction of parallel breakpoints detection and edge tracing without recursive operations. Experimental results show that, the improved Canny operator has obtained a high real-time performance of edge detection in image processing.