A novel retinal vessel segmentation based on local adaptive histogram equalization

Abstract

The appearance and structure of blood vessels in retinal images has an important role in diagnosis of diseases. This paper proposes a method for segmentation of blood vessels in color retinal images. We present a method based on preprocessing of retinal images by enhancing the histogram of grayscale images. We research the effect of linear and non-linear intensity transformation in retinal vessel segmentation. In this paper we focus on two methods of retinal vessel segmentation including first derivative of Gaussian matched filter and Gaussian matched filter. These methods are used by many researchers [1],[6],[7],[8],[24] but still there are some problems in vessel detection. We make use of adaptive histogram equalization to overcome this drawback. The proposed technique is tested on publicly available DRIVE database. The effect of improved matched filter can be seen based on accuracy calculation. The accuracy of 0.9353 shows a raise of about 2 percent compared to existing methods.