Research On Image Processing Technology Of Fundus Retinal Vessel Segmentation

The morphology of blood vessels in retinal fundus images is an important indicator of diseases like glaucoma,hypertension and diabetic retinopathy.The accuracy of retinal blood vessels segmentation affects the quality of retinal image analysis which is used in diagnosis methods in modern ophthalmology.Due to the low clinical efficiency of color fundus image recognition,it is easily affected by the environment and the doctor’s own experience.This paper uses images semantic segmentation based on deep learning to achieve accurate segmentation of retinal vessels.In view of the problem of retinal blood vessel segmentation in color fundus images,this paper focuses on two aspects of the fundus image vessel enhancement and improved U-Net network model.The following contents are mainly studied.Firstly in view of the characteristics of retinal blood vessels,retinal blood vessel enhancement and segmentation technology is in-depth studied in this paper.By combining the characteristics of different image enhancement algorithms to complete the enhancement of retinal blood vessels,this paper first uses a method based on histogram equalization to improve the quality of the image and maintain local details,and enhance the contrast between the blood vessel and the background.Then the MSR algorithm is used to enhance most of the details of the blood vessel image.At the same time,in order to reduce the influence of noise,the wavelet algorithm is used to select the appropriate threshold and threshold function to effectively reduce the larger amplitude noise in the retinal image and obtain a clearer retinal blood vessel image.Secondly aiming at the problem of confusing blood vessels and surrounding environment information in the process of retinal image blood vessel extraction,which leads to a decrease in extraction accuracy,an improved U-shaped convolution neural network model with low-dimensional feature information enhancement is proposed.Because the low-dimensional detailinformation is gradually weakened during the network propagation process,before the fusion of the feature map of the expansion path and the feature map at the same level of the contraction path in the U-Net network,it is first fused with its previous feature map to further optimize the edge extraction accuracy of the extraction results.At the same time,a cascaded hole convolution module is added at the bottom of the U-Net network to achieve the expansion of the receptive field of the retinal image feature map without adding additional parameters,retaining multi-scale vascular features and more image details,especially for Segment the smaller blood vessels in the fundus retina,which will also effectively prevent the over-fitting of the fundus retinal blood vessel images during training.Experimental results based on the standard image set DRIVE show that the average accuracy,sensitivity,and specificity of the proposed algorithm are higher than those of the traditional U-Net algorithm,,respectively,as well as outperform other segmentation algorithms.