Retinal Vessel Segmentation Algorithm Based on Improved Gaussian Matched Filter

Retinal images can assist doctors in diagnosis, and their segmentation algorithms have received widespread attention. This article proposes a retinal segmentation algorithm based on an improved Gaussian matched filter. Firstly, the pre-processing step enhances the contrast of blood vessels, obtaining results of equalization and morphological filtering . Compared to the former result, the latter one weakens the spot interference and noise interference while it also loses a lot of detail information. Therefore, the above two preprocessing results are respectively applied matched filtering in this paper. Considering the characteristics of blood vessel distribution , this paper improves the Gaussian kernel function in two dimensions. Then change the Gaussian kernel scale in horizontal and vertical directions, and apply matched filters to the above preprocessed results separately. Finally, use detail growth algorithm and selectively fuse the results . Experimental results on DRIVE show that the average accuracy, sensitivity and specificity are 95.4%, 88.21%, 95.8%, respectively. Compared with traditional matched filtering methods, this method segments blood vessel structures with better continuity, less interference from light spots, and significantly improved average sensitivity, which is beneficial for detecting small blood vessels.