The Segmentation Of Overlapping Milk Somatic Cell Based On Improved Watershed Algorithm

It is important for automatic recognition of milk somatic cells to improve the quality of milk, detection efficiency and accuracy of cow mastitis. But the phenomenon of multiple cells adhesion often appears in the image. It severely affects the analysis of cell parameters. Currently, most algorithms studied on overlapping cells segmentation is for particular object. For other object, the effect is poor. Therefore, to design a suitable segmentation method for overlapping milk somatic cells in this paper.In the paper, milk somatic cell image is preprocessed firstly, including: image acquisition, gray transformation, filtering, image segmentation, and so on. Then, the shape factor chose to achieve the discrimination of overlapping cells. The main idea is: analyzing the shape factor of a group of milk somatic cells, choosing a threshold. That is the shape factor of cell than the threshold removed from the images. The remaining cells are overlapping.An improved watershed algorithm is selected to separate overlapping milk somatic cells image. The main idea is: transforming binary image to distance image, obtaining center points of the local maximum regions(seed points), and merging the redundancy seed points, deleting the region of only a single seed point, that is to delete overlapping area better, and also improve the disadvantages of the traditional watershed algorithm. The more accurate segmentation points in connection lines are selected using of the concept of positive skip variable. the final segmentation lines are found out with a recursive algorithm.Experimental results show the improved watershed method can accurately separate overlap milk somatic cells, and inhibit the over-segmentation is obvious.