Non-uniform channels mixing and template matching based denoising of retinal images for optic disk detection

Optic Disc (OD) recognition is a fundamental phase in the developing methods for automated analysis of different chronic ocular disorders. In this article, a novel technique has been introduced for OD detection and borderline extraction, which can be witnessed as a milestone in the advancement of a Computer-Assisted Diagnostic (CAD) framework for glaucoma detection in fundus photographs. A unique solution has been presented for removal of illuminated light spots due to refraction of camera light (so called fringe noise) on the edges of the retinal color images, which leads to false detection of OD. An intelligent non-uniform mixing of red and green channels is introduced for denoising of retinal photographs because of the dissimilar noise level in each channel. Template based matching technique is applied for OD localization of retinal image. The approach is thoroughly verified on openly accessible the DRIVE and the STARE datasets for OD detection and denoising of retinal images. This denoising technique greatly improved the accuracy of OD detection in contrast to other contending approaches.