基于FCM和标记分水岭的粘连岩石颗粒图像分割

Image segmentation of overlapping rock particles based on FCM and labeling watershed

模糊C均值算法可利用图像的多种特征值进行准确的图像分割,但不能分割粘连物体;传统的分水岭分割算法能够获得准确的物体边缘轮廓,但容易造成过分割.为了解决这个问题,提出基于FCM和标记分水岭的粘连图像分割.该方法首先对原始彩色图像中值滤波后进行基于LUV颜色空间的FCM聚类;对聚类后的图像用形态学方法去杂质、空洞填充后进行距离变换;然后根据距离变换图像找出局部最大值,得到种子图像;最后对距离变换图像进行基于标记的分水岭分割,得到最终的分割图像.该方法对粘连岩石颗粒图像进行分割,取得了较好的实验效果.

FCM algorithm can achieve precise image segmentation by using various features of an image, but it can not segment overlapping object. The traditional watershed algorithm can obtain precise edges, but result in oversegmented. To solve this problem, a method for overlapping particle segmentation is proposed based on FCM and labeling watershed. First, FCM algorithm based on LUV model is applied to segment an original color image. Next, the impurities of the segmented image are removed and the empties are filled. Then, a distance transform and a local maximum search are performed to find the peak pixels as the seeds of the segmented regions. Finally, a labeling watershed algorithm is applied to the distance transform image to get the final segmented image. Experimental results showed that the method we proposed is efficient for segmentation of the overlapping rock particles.