Automated retinal vessel segmentation using multiscale analysis and adaptive thresholding

Computer based analysis for automated segmentation of blood vessels in retinal images will help eye care specialists screen larger populations for vessel abnormalities. Because the width of retinal vessels can vary from very large to very small, and the local contrast of vessels is unstable especially in unhealthy ocular fundus, the automated retinal segmentation is difficult. We propose a novel method with the consideration of these problems. Our method includes 1) a multiscale analysis scheme using Gabor filters and scale production, 2) an adaptive thresholding scheme using adaptive tracking and morphological filtering. Our method is good for detecting large and small vessels concurrently. It is also efficient to denoise and enhance the responses of line filters so that the vessels with low local contrast can be detected.