A Morphological Method to Detect Red Lesions in Retinal Images

摘要

In this paper, we address the problem of detecting red lesions in the color retinal images. Fundus photos are always affected by the illumination variations across the image and lead to misclassification of red lesions from the background and the blood vessels. In this research, the green-plane of the original RGB color image is shade-corrected in the pre-processing step. We introduced a multiscale-windowed enhancement method to the image before the morphological operations and the following thresholding step. A series of morphological operations are applied to separate the red lesions and the vasculature from the background of the image. An improved Fuzzy C-Means (FCM) algorithm is used to find the optimal threshold. We obtain the candidate set of red lesions by removing the large vessels using a length filter. Finally, region growing technique is applied to determine the actual shape the red lesions and remained small vessel segments. Shape analysis is applied to the region grown image to identify the red lesions. The results showed that the proposed method gave a satisfactory performance in the detection of red lesions.