A comparison study to evaluate retinal image enhancement techniques

Abstract

Retinal vessels can show different states of several diseases, making the detection of vessels in retinal images very crucial. Retinal images can be used for other applications such as ocular fundus operations and human recognition. Due to the acquisition process, these images often have low grey level contrast and dynamic range that can seriously affect diagnosis procedure results. In this paper, we present an algorithm for retinal image contrast enhancement based on the second generation of new multi-resolution analysis tool called Curvelet Transform which is faster and simpler than the first version. Curvelet transform has favorable geometric features that provide better representation of edges compare to other widely used multi-resolution tools such as Wavelet. We use a nonlinear function to modify the Curvelet coefficients based on their statistic features. Results of applying this method to a known database, DRIVE, and comparing with previous approaches, proved this method to be effective and helpful for image segmentation.