基于改进FAST与分水岭算法的颗粒图像分割

Image segmentation of touching particles based on improved FAST and watershed algorithm

为准确、高效分割粘连颗粒,提出一种改进FAST特征点检测算法,并与基于h-maxima变换的分水岭算法相结合实现粘连颗粒的分割。该算法通过h-maxima分水岭算法得到候选边缘分割点和候选分割线;采用改进FAST算法遍历二值图像边缘获得图像边缘特征点;利用边缘特征点对所有候选边缘分割点进行非极大值抑制得到待边缘分割点;通过待边缘分割点从候选分割线中提取待分割线;利用边缘特征点识别待分割线中的伪边缘分割点,并去除与其相连的分割线;获得正确分割线进行图像分割;对典型颗粒进行分割实验,并与现有3种图像分割算法进行比较。结果表明:该算法相对运算时间最短,分割正确率最高且均大于95%。

In order to segment touching particles accurately and efficiently,an improved FAST feature point detection algorithm was proposed,which was combined with the watershed algorithm based on h-maxima transformation to realize the segmentation of touching particles.First,the candidate segmentation points and lines were determined by the h-maxima watershed algorithm.Next,the feature points of the edge image were obtained by traversing edges of the binary image based on the improved FAST feature point detection algorithm.Then,the method of non-maximum suppression was used to gain pending segmentation points from all the candidate ones according to the distribution of feature points.The candidate segmentation lines were selected by the pending segmentation points to get the pending segmentation lines.Finally,the pseudo segmentation points in the pending segmentation lines were identified according to the edge feature points and the segmentation lines connected with the pseudo segmentation points were removed,and the correct segmentation lines were determined.Segmentation experiments on images of representative particles were carried out,compared with the three existing image segmentation algorithms.Experiment results show that computation time of the improved algorithm is the shortest and the segmentation accuracy is greater than 95%.