Histogram equalization techniques for enhancement of low radiance retinal images for early detection of diabetic retinopathy

Abstract

Images are sometimes affected by improper illumination and are dark. This happens usually in medical images or the images acquired in low light conditions. This paper focuses on retinal imaging and proposes two techniques, RIHE-RVE (Radiance indicator based histogram equalization for retinal vessel enhancement) and RIHE-RRVE (Radiance indicator based histogram equalization for recursive retinal vessel enhancement) to address the problem of low light radiance. The techniques separate the histogram into sub-histograms at the split value determined by the tuneable parameter, ψ. RIHE-RVE recursively performs histogram integration after each split followed by equalization whereas in RIHE-RRVE histogram split can be done to any level (which is decided by the parameter,r) followed by equalization and integration. It has been observed from a comprehensive literature survey that very few algorithms exist that enhance the quality of retinal images. The proposed methods efficiently address the low light radiance problem. Performance evaluation of the techniques is done in terms of Information content (Entropy), PSNR (Peak signal to noise ratio), SSIM (Structure similarity index measurement), Euclidean distance and visual quality inspection. To demonstrate the robustness of the proposed methods, the techniques are not only applied specifically to publicly available retinal databases DRIVE, STARE and CHASE_DB1 but also to some of the MRI images taken from publicly available OASIS database. Results show that the proposed techniques outperform the state of the art techniques especially in low radiance images.