Abstract: According to the characteristics of real-time, rapidness, robustness in object recognition and the slow changrog of the most edge directions, a new memory-based edge crawling method was proposed in this paper. An obvious flaw of the edge crawling was that the bug was easy to be trapped and kept crawling around a certain local region. Memory-based searching algorithm could remedy the shortcoming of the edge crawling effectively. This method segmented different objects having the similar colours by marking the extracting edge during the course of edge crawling. The experimental results demonstrated that, compared with traditional approaches, the adopted approach was able to get more complete and clear contour, shorten the time of image processing and therefore optimize the accuracy and robustness of object recognition.

Key words: Edge detection, Edge crawling, Image segmentation, Robot vision, Object recognition