Automatic fundus image field detection and quality assessment

Abstract

Fundus images are an important diagnostic tool for many retinal diseases. Sometimes the images captured are of low quality and cannot be used for diagnosis requiring repeat image acquisition. So, it is efficient to have an automatic system to assess the quality of the fundus image during the time of image capture. We have developed an automatic approach to assess the quality of the acquired fundus image based upon the inherent symmetry of retinal vessels. We approach the problem of quality assessment in two ways-individual quality assessment of a single fundus image and comprehensive quality assessment of a set of three fundus images of different fields of an eye. Our method also detects the field and side of the fundus image using the position of optic disc and the intensity information in two local windows.