基于浸没改进的标记分水岭图像分割算法

Advanced merge based watershed image segmentation

针对分水岭变换图像分割算法浸没效率低的缺陷,本文在分水岭浸没阶段通过使用无效盆地标记数组来优化浸没处理.首先在预处理阶段采用形态学梯度算子使原始图像的灰度级跃变更为急剧,然后通过巴特沃斯低通滤波较好地过滤掉噪声;接着在分水岭浸没阶段识别出标记错误的盆地,并将这些错误盆地存储在一个无效盆地标记数组中;最后通过该数组对错误盆地统一进行一次合并.实验结果表明:该方法能显著提高分水岭浸没效率,同时较好地抑制过分割.

This paper proposes an improved approach of watershed image-segmentation to ameliorate its efficiency. Invalid-marked basin array is introduced in submerge process. In preprocessing phase, mor phological gradients and Butterworth filter are used to reduce noise in images while preserving essential outline. In submerge phase, basins are extracted from marked gradient image and stored in Invalid marked basin array. Comparing to other methods, the approach significantly improves efficiency of wa tershed image-segmentation while suppressing excessive segmentation.