GMM-Based Markov Random Field For Blood Vessels Segmentation And Extract In Color Retinal Image

Some diseases, particularly cardiovascular disease, will change the shape and structure of retinal vessels. Observation and detection of retinal vessels play an important role in the diagnosis of disease. Traditional diagnosis of retinal vessels that ophthalmologist perform under artificial visual attending, its test results showed inaccurate, poor repeatability, inefficient and poor comparability etc. Relying on the rapid development of digital image processing technology, the detection of retinal vessels images with a computer system that provides accurate results for the effective treatment of medical diagnosis is particularly important and urgent.After reading a lot of related references, I carry out some exploration on retinal vessel segmentation. It is worth mentioning that this article first uses Gaussian-Markov random mixed model in retinal vessels segmentation and achieved certain results. The overall research can be divided into three large pieces.(1) Color fundus image preprocessing. After comparison each channel of RGB color fundus image, selected network of blood vessels with the highest degree of the green channel background image as a comparison object for subsequent processes. As brightness adjustment on the background of uneven illumination is performed. Then, the retinal vessels enhancement by using two-dimensional Gaussian matched filter witch with12different angles every15degrees.(2)Segmentation of retinal vessels is performed by the use of Gauss Markov random mixed model. Based on the theory MRF-MAP, by conducting mathematical derivation and estimating model parameters derived from the past by expectation maximization (EM) algorithm, this model is derived. This model can estimate simultaneously Gaussian mixture model parameters and the Markov random field model parameters, thus improving the convergence rate of the parameter estimation.(3) In post-processing, remove fundus image border interference by the use of image mask, process segmentation results debris by the use of the area filtering method. There are some advantages of the algorithm mentioned in this article. On the one hand, by the use of mixed Gaussian Markov random field model, effective segmentation of retinal vessels is performed, and convergence efficiency is improved by simultaneously estimating Markov and Gaussian parameters. On the other hand, based on the Gaussian Markov random mixed model, image priori probability is added that provide pixel space constraint information, thus segmentation accuracy is improved. Through simulation experiment, a reliable retinal vascular segmentation results is derived, which verified the feasibility of this algorithm.