On Correcting Luminosity and Contrast of Retinal Images with Reflectance

Abstract

In this paper, we perform an automatic correction of luminosity and contrast of retina images. One hundred retina images with varying level of reflectance taken from custom and online databases are used to test the effectiveness of the proposed method. The approach is implemented in 4 stages namely pre-processing, lowpass filtering, luminosity equalization, and contrast stretching. In the pre-processing stage, the three components of a color retina image are separated and only the green channel is processed further as it contains the most information. Then the region of interest (which is the eye region) and its border are marked. After that, the eye region (ROI) of the green channel is subjected to lowpass filtering, row by row to create a smooth background luminosity surface without the foreground objects like the optic disc, exudates, blood vessels and blood spots. Three different types of lowpass filters are used and their performances are compared. The resulting background luminosity surface allows the estimation of the background illumination. Then, based on the background surface, the luminosity of the ROI is equalized so that every pixel experiences the same brightness. Finally, the contrast of the ROI is improved by histogram stretching so that the foreground objects appear more clearly. The proposed method was implemented using MATLAB R2021b running on AMD 5900HS processor and the average execution time was less than 1 second. The execution time can be further reduced if the codes are optimized and GPU is used. Overall, the proposed method improves the luminosity and contrast of the images greatly. This technique can be a useful tool to ophthalmologists who perform visual inspection of microaneurysm, exudates and other lesions.