Segmentation Of Milk Somatic Cell Image Based On Level Set

The type and quantity of the milk somatic cell is an important index of the milk quality appraisal and a cow's health. There is the important practical significance to analyze the milk somatic cell image with the digital image processing technology. Therefore, the thesis mainly concentrates on research of the milk somatic cell image's graying, image filter, and image segmentation.After analyzing the theories and applications of cell segmentation in domestic and foreign, considering the milk somatic cell color is monotonous, synthesized speed and efficiency factors, the gray images of milk somatic cell are used as the segmentation object. Firstly, the weighted mean approach is used to transform color image into gray image, and gauss filter is applied to remove the image noise. Secondly, improved variational level set algorithm is applied on gray image of the segmentation of milk somatic cell. The constant function in the C-V algorithm is replaced by the Local Binary Fitting function in improved variational level set algorithm. The gauss function introduced emphasizes the local characterist and automatically expands the function domain of definition to the total image domain. The distance regularizing term is introduced to avoid reinitialization in traditional level set methods.Finally, the milk somatic cell image is segmented by the traditonal level set method, the C-V variational level set method and the improved variational level set method respectively. The experimental result showed that improved variational level set method has the distinct improvement in both the segmentation efficiency and the segmentation effect. Based on the improved algorithm segmentation, the cell separation's algorithm is proposed. The method restructured the strong contrast gray image according to the contour's position. Gray image binarization processing, open and close operation is carried on to obtain the good segmentation effect. The separation of overlapping cells is performed using mathematical reconstruction and watershed algorithm.