Combining blood vessel segmentation and texture analysis to improve optic disc detection

The paper presents a methodology and algorithm to improve the optic disc detection and localization in retinal images. The image is decomposed in sub-images which have dimensions adapted to image resolution using gliding-box algorithm. The proposed method combines texture analysis of the sub-images, as first stage, with blood vessel detection, as second stage. This combination of techniques is useful in the presence of exudates. Taking into account the analysis of the feature efficiency on different color channels and also the pixel distribution it was establish an optimal set of features to perform the optic disc detection. The method was tested on 100 images using the publicly available STARE and DIARET datasets. Comparing the accuracy of different methods, this paper shows that the proposed algorithm is more efficient.