A novel segmentation algorithm for clustered slender-particles

A novel algorithm based on watershed and concavities is proposed to segment the clustered slender-particles, such as the clustered rice kernels. First, the distance and watershed transform is used to the binary image of clustered slender-particles. Secondly, the watershed post-processing of over-segmentation is dealt with by utilizing concavity features of related shapes. Thirdly, the candidate splitting lines of touching clusters is found by matching the concavities to the un-segmentations left. Finally, the supplementary criterions are applied, such as the shortest distance, the opposite orientation, the splitting path orientation, etc., to determine whether a candidate splitting line can be accepted or not. Experimental results show that the algorithm can segment the large-scale clustered slender-particles efficiently, where such a quantitative analysis was previously infeasible.