基于标记的分水岭图像分割算法研究

Research on Image Segmentation Based on Marked-Watershed

本文针对分水岭算法的过分割问题,设计了一种结合形态学运算的基于标记的分水岭算法。算法首先对梯度图像进行形态学开重建运算和形态学闭重建运算滤波平滑;然后去掉图像中像素个数小于20的局部极小值区域,提取前景标记;用Otsu法求阈值,对二值图像进行距离变换,分水岭分割,得到的分水线图像作为背景标记;最后用强制最小技术修改梯度图像,进行分水岭分割。实验结果表明,此方法能有效抑制过分割,得到与人工标注的分割目标更接近的结果。

A mark-based watershed segmentation algorithm combined mathematical morphological was proposed in this paper to deal with over-segmentation. Firstly, the gradient image was smoothed and filtered by morphological opening reconstruction operation and closing reconstruction operation. Secondly, local minimal regions less than 20 pixels were removed and the foreground marker was extracted. Then, the background marker was extracted by watershed segmentation based on Otsu threshold and distance transform. Finally, these markers were used to modified the gradient image based on the minima imposition procedure. The experimental results indicate that this method can well suppress the over-segmentation and obtain a satisfactory result, which are more accordant to human-vision segmentation.