Gradient orientation based feature detection: an application for extracting retinal blood vessels

A computer-assisted system for diagnosis of eye diseases has been in great demand. This paper presents a robust and efficient method for the detection of anatomical features in retinal images. The method is based on the analysis of gradient orientation and is not directly affected by image intensity. For this, the method performs very well despite the inherent problems of the retinal images, such as low contrast and non-uniform illumination. The method is particularly suitable for detecting features with circular and linear structures. A multi-scale approach is employed to detect various sizes of features, especially blood vessels with varying diameters. The blood vessel network is finally extracted from the detected features by global thresholding with some morphological operations.