Studies On Segmentation Methods Of Milk Somatic Cell Color Images

The accurate segmentation of milk somatic cells in microscope images may contribute to development of a successful system that automatically analyze,detect and count cells in microscope images. It is important to improve the milk quality detection and diagnose bovine mastitis. Therefore, this thesis mainly concentrates on researching the segmentation methods of milk somatic cell color images, including common color image segmentation, milk somatic cell segmentation, separation of touching and overlapping cells and somatic cells count.After analyzing the theories and applications of cell segmentation, the image pre-processing, color space conversion, mathematical morphology theory, image fusion technology and other basic theories and methods are described.To achieve the purpose of rapid and accurate milk somatic cell image segmentation, the common methods for color image segmentation are studied. Because the image segmentation methods of 3-D color spaces consume large amount of computation and have slow speed. They do not suit to real-time applications. However, the accuracy of segmentation methods of 1-D and 2-D space is not high. For those reasons, an approach for color image segmentation, called WHF2D, is presented. The approach is based on segmentation of subsets of bands using mathematical morphology followed by the fusion of the resulting segmentation"channels". For RGB color images, the band subsets are chosen as RG, RG and GB pairs. The three segmentations in 2D color spaces are obtained using the watershed algorithm. The three 2D segmentations are then combined to obtain a final result using image fusion technology. Through quantitative evaluation, this method is average 10 times faster than that of 3-D space. There is no significant difference between the two methods in the correct rate.In the process of the realization of the WHF2D method, the color space quantitative evaluation and selection method based on the contrast are proposed, the aggregate function combined with local and global information in a watershed algorithm is proposed, the quantitative evaluation method of segmentation validity based on a maximum variance criteria is proposed, and the image fusion technology based on the split and merger is studied. According to milk somatic cell image characteristics, the WHF2D segmentation method applied to milk somatic cell image segmentation could reasonably extract cytoplasm and nucleus from a complex background. Through analyzing and comparing the accuracy of the segmentations for the three methods K-means,FCM and WHF2D, it shows that WHF2D is more effective. Thereby, milk somatic cell segmentation strategy is created.Separation of touching and overlapping cells is a crucial problem. On the basis of cells segmentation and extraction of cells, a binary image composed of just cells and background, is obtained. The separation of touching and overlapping cells is performed using a distance image from the binary image, mathematical reconstruction and watershed algorithm. The somatic cell count is realized using different connected regions and combining with the criteria of milk somatic cell count.