Spatial correlation based contrast enhancement for retinal images

Abstract

A Novel Contrast Enhancement Approach is proposed in this paper to enhance the quality of low contrast retinal images such that the automatic diagnosis of Diabetic Retinopathy results in more accurate results. Unlike the conventional approaches, this approach considers the spatial correlations between image pixels along with image pixel intensities. Due to the consideration of spatial correlations, the pixel intensities in the output image are not only linear combinations of input pixels but also relate to the neighboring pixel intensities. Extensive simulations are carried out over the proposed approach through the standard fundus image datasets such as DRIVE and DIARETDB1. Further the proposed approach is compared with conventional approaches and the performance enhancement is measured through the performance metrics like contrast improvement index and linear index of fuzziness and observed that the proposed approach outperforms the conventional approaches.