Abstract: Retinal vessel appearance is an important indicator for many early diagnoses, including diabetes, hypertension, and arteriosclerosis. Computer processing can help doctors' work and a retinal vessel image segmentation algorithm was proposed. A local normalization algorithm was used to eliminate background differences. Expectation-maximization algorithm was used to classify the pixels into several classes to obtain segmentation results. At last, based on the anatomical and physiological characteristics of the eye, a reconstruction method to reconstruct three-dimensional retinal vessel was realized by inverse projection theory. The model can be viewed from different directions.