Automatic Localization of the Optic Disc Based on Iterative Brightest Pixels Extraction

Abstract

The identification of the optic disc (OD) is necessary for the computer aided diagnosis of retinal diseases, but most methods for the OD detection often fail in the existence of retinal lesions and imaging artifacts. This paper proposes a new method based on iterative brightest pixels extraction (IBPE) for OD localization, which is designed to overcome the presence of large exudates or bright artifacts. The iterative algorithm integrates brightest pixels extraction and discrimination of geometric features. At each iteration step, the brightest pixels are extracted and a binary candidate map is constructed. In order to distinguish the OD region from candidates, two geometric features of each candidate are employed. Then, a checking process is performed. The algorithm deletes the false positive OD region and re-gets the new brightest pixels, iteratively. Hence, the real OD region rises finally though it is not bright at the first. The proposed method is evaluated on STARE retinal database. The experimental results show that the performance achieves a success rate of 95%.