Novel and adaptive contribution of the red channel in pre-processing of colour fundus images

Abstract

A new pre-processing method for colour fundus images with adaptive contribution of the red channel is proposed. Based on a condition that is developed in this paper, this method utilises the intensity information from both red and green channels instead of using only the green channel as in the usual practice. The histogram matching is used to modify the histogram of the green channel by using the histogram of the red channel (of the same retinal image) to obtain a new processed image having the advantages of both channels. This method can be used to correct non-uniform illumination in colour fundus images or as a pre-processing step in the automatic analysis of retinal images. Results show that the use of histogram matched (HM) image give better performance than using the green channel image when employing the two-dimensional matched filter to detect retinal blood vessels. At specificity of 90%, in case of abnormal images, sensitivity increased from 76% when using the green channel image to 82% when using the HM image compared with 81% when using the piece-wise threshold probing method. In case of normal images, at the same specificity, the sensitivity obtained when using green channel image or HM image was 87% compared with 88% for the piece-wise threshold probing method.