Accurate localization of the optic disc based on LBP descriptors

The paper presents a new methodology for the detection and accurate localization of the optic disc in retinal images. Different color channels are investigated and, as a result, LBP histogram on Hue channel is combined with the information of intensity level on Green channel. The proposed algorithm is applied on overlapped patches, having a corresponding dimension, extracted from the retinal images using a gliding box algorithm. The patches are discriminated for optic disc detection by computing similarity features based on the resulted histogram vectors and the mean intensity feature. A number of 100 images from MESSIDOR database were tested and comparing with other works, the proposed algorithm gave good results in terms of accuracy.