基于标记分水岭和区域合并的彩色图像分割

Color Image Segmentation Based on Marked-Watershed and Region-Merger

针对分水岭算法的过分割问题,设计了一种结合标记分水岭和区域合并的彩色图像分割算法.算法首先根据梯度图像局部极小值的综合信息自适应提取标记,在标记的基础上进行分水岭变换获得初始分割结果;然后结合人眼对区域相似性的实际感知定义了融合颜色距离、方差和边缘信息的综合距离度量,通过综合距离度量控制区域进行快速合并得到高级分割结果.仿真结果表明,新设计的算法能够有效抑制过分割,获得更小的局部一致性错分率,与人工标注的分割结果更为接近.

A color image segmentation algorithm combined marker-based watershed and region merger was proposed in this paper to deal with over-segmentation.Firstly,it extracted minima-marker adaptively according to the information of local minima of gradient,and watershed the marked gradient image to get the pre-segmentation result.Then,it modified region distance according to real sense of region-similarity of human-vision,and defined an integration distance measurement considering color distance,color difference and edge information,which would lead the pre-segmentation regions to merge to a final segmentation result.The experimental results indicate that the proposed algorithm can well suppress the over-segmentation and obtain a satisfactory result,which owns a lower Local Consistency Error and are more accordant to the human-vision segmentation.