Abstract:

In order to solve the over-segmentation problem of the traditional watershed algorithms，an improved watershed segmentation algorithm for the bridge image is proposed. Firstly，input images are filtered by top-hat transformation and bottom-hat transformation. Secondly, a multi-scale algorithm for computing morphological gradient images is proposed, and the threshold for marker-extraction is automatically calculated according to the statistics of local extreme points in the gradient map. Thirdly, the watershed algorithm is applied to the modified gradient map to segment the image, and the over-segmentation regions of the initial watershed segmentation is settled by region merging based on fisher distance, which is ended according to divergence principle．Experimental results show that oversegmentation phenomenon can be reduced significantly, and the image accuracy is improved while maintaining strong robustness and adaptability．

Key words: morphological filtering;watershed;over-segmentation