关键词: 分水岭算法, 路面裂缝检测, 区域合并, 梯度熵

Abstract: The traditional watershed algorithm in image processing has the advantage of remaining good weak edge information, but it is easy to cause the over-segmentation problem. Aimed at the over-segmentation problem of traditional algorithm, this paper improves the watershed merging algorithm, and proposes the concept of gradient entropy combined with characteristics of asphalt pavement cracks. At the same time, it puts forward an adaptive region merging algorithm as judged by region merging rules, the entropy as the criterion to judge whether region merging should be stopped as well. Then the method is applied in pavement cracks detection. The experimental results show that pavement cracks extracted by the method have little of noise, while details are remained well compared with the traditional threshold segmentation method, and then this method solves the over-segmentation problem effectively.

Key words: watershed algorithm, pavement crack detection, region merging, gradient entropy