Novel method for automatic generation of fundus mask

TLDR: The proposed method uses a log operation on the image for mask generation on the fundus image for automated detection of features describing Diabetic Retinopathy, validated by experiments on different public databases like DIARETDB1 and MESSIDOR.

Abstract: Fundus imaging is a vital diagnostic tool in detection of retinal diseases. It is one of the routinely used method for screening Diabetic Retinopathy. Diabetic Retinopathy is one of the vision threatening and leading cause for blindness, early detection of which would prevent it Processing the fundus image for automated detection of features describing Diabetic Retinopathy includes extraction of foreground from background. This needs generation of a proper mask. Previously used techniques include thresholding and morphology operations. We have used image processing tools like ImageJ software [10] which also have tools to generate masks. All of these existing methods fail to give a perfect mask when the fundus image is not well exposed. In this paper we propose a simple yet effective method to overcome this drawback. The proposed method uses a log operation on the image for mask generation. The effectiveness of the proposed method is validated by experiments on different public databases like DIARETDB1 [11] and MESSIDOR [12]. The average error rate of our proposed method is found to be 0.4237%.