Image database clustering to improve exudate detection in color fundus images

In this paper a novel approach to improve exudate detection in color fundus images is proposed. Image databases usually contain images with different characteristics, thus determining an optimal parameter setting of an algorithm is a challenging task. To overcome this problem we cluster the image databases. For each cluster an optimal parameter setting is determined for the same algorithm. We extract Haralick features from the image, and apply k-means clustering to obtain the clusters. We tested our approach on a publicly available database, where the proposed approach improved the performance of a state-of-the-art exudate detector.