Abstract:Zebrafish, a model organism, is widely used today. Quantitative analysis of the physiological effects of drugcultured zebrafish embryos during development and obtaining accurate index values by visual observation is difficult. Therefore, it is necessary to use computers. The basis for the quantitative analysis of zebrafish images is that the zebrafish embryo images can be segmented in the image according to the zebrafish’s physiological structure, i.e. the head, trunk and yolk. However, because of the limited number of experimental groups for specific drugs in zebrafish drug experiments, they cannot be segmented by machine learning such as deep learning, and only image processing modeling can be used for segmentation. In this paper, the image was semantically segmented by using distance transformation combined with watershed algorithm, subtraction clustering combined with Kmeans clustering algorithm and subtraction clustering combined with floodfilling algorithm. Finally, the physiological structure segmentation effect of floodfilling algorithm combined with subtraction clustering meet the purpose of the study, and the segmentation laid a good foundation for subsequent quantitative analysis in the medical research.