Abstract: Due to the problems of excessive segmented regions in the result of existing algorithms, it constructs the importance map based on the crossed field of edges, and constrains the segmented regions by the stability of feature of the edges to improve the quality of segmentation effectively. It constructs the crossed field of edges by using Gauss integral to enhance the continuity and the stability of the line of edges, then transforms the distance by the feature of the edges to obtain the importance map and segments the image by mean-shift, merges the results of segmented regions according to the strength of importance of the adjacent regional border. Experimental results show that compared with the old ways of segmentation, this algorithm can be used to have an effective merging on detail regions as keeping the important regions of source image, and improve the quality of segmentation obviously.

Key words: image segmentation, mean-shift, importance map, regional merging

摘要： 针对已有算法结果分割区域过多问题，提出采用边缘正交场构造重要性图，通过边缘特征稳定性约束分割区域，从而有效地提高分割质量。构造边缘正交场，通过高斯积分提高边缘线的连续性和稳定性。采用边缘特征进行距离变换，生成图像的重要性图。采用均值漂移进行图像预分割，根据相邻区域边界上的重要性强度对分割区域结果进行合并。实验结果表明，和原有分割方法相比较，算法在保持原始图像重要区域的同时，对细节区域进行有效合并，明显提高分割质量。

关键词: 图像分割, 均值漂移, 重要性图, 区域合并