Texture based method for automated detection, localization and evaluation of the exudates in retinal images

Abstract

The detection, localization and evaluation of exudates can contribute to help to diagnosis and to automatic create a specific databases. In this paper, we propose a new method for detection of exudates using adaptive selection of features based on matching score. For image processing two algorithms were proposed. The first uses sliding box method for optic disc detection and the second uses non-overlapping boxes for exudates detection. The selected features combines spatial distribution, spectral properties and fractal characteristics of the region of interest and are based on sorted values. The algorithm has three steps: detection and removing of the optic disc, detection and localization of exudates and finally the evaluation of exudates size. In order to evaluate of algorithm efficiency we used 70 images with exudates. The experimental results prove the efficiency of the proposed method due to accurately detection of the exudates and the size evaluation.