Retinal Image Segmentation Based on Texture Features

Vessels segmentation is one of the key issues in the study of retinal image lesions. In this paper, a segmentation method of blood vessels using texture feature theory is introduced. This method is based on the texture features quantification and it is different from other blood vessel segmentation methods. We have quantified a large number of texture features that can be used in image processing. There are 41 kinds of texture features in all. The method presented in this paper is applied to retinal images in DRIVE and STARE databases. Through quantitative analysis of vascular texture features of retinal images, we found that several texture features, like gray variance, can easily identify the foreground and background of blood vessels, and thus can obtain better visual effect than the original image. It can provide more information for vessels segmentation and optic disc location.