Study on retinal vascular image segmentation method based on hybrid model

Abstract

Basing on morphological characteristics of retinal vascular structure and its changing, in order to realize the early diagnosis and quantitative analysis of the severity of diabetes, cardiovascular disease, and fundus disease ect, we propose a retinal vascular image multi-scale segmentation method based on hybrid model in this paper. First, combing the statistical principle and image enhancement method, the retinal vascular image was segmented and extracted. Second, a hybrid model consisting of a Gaussian model and two exponential models for vascular fitting was developed. Then, the K-means clustering method is used to estimate initial parameters, and the estimated parameters are iteratively processed to solve model parameters; Finally, the retinal vascular image is segmented according to the maximum a posteriori criterion to extract vessels. The experimental results on DRIVE database show that our proposed segmentation method can extract retinal vascular network effectively, and the segmentation accuracy is 94.62%. The proposed segmentation method can thus help the ophthalmologists in efficient retinal image analysis and fruitful treatment to the patient community.