Retinal Vessel Segmentation Using Matched Filter with Joint Relative Entropy

The matched filter is an effective method for the detection of retinal vessels when combined with other processing techniques. This paper presents a segmentation method to improve the extraction of retinal vessels based on the matched filter. The method combines a morphological approach to enhance retinal vessels before applying the matched filter and a modified joint relative entropy (MJRE) thresholding method to segment the matched filter response. The morphological approach is designed to suppress irregular bright regions and noise while preserving the information of vessel edges, and to improve the contrast of vessels, especially thin ones. The joint relative entropy thresholding is modified to provide an optimal threshold value for segmenting the retinal vessel tree properly. The proposed method is tested on the DRIVE dataset, yielding an average accuracy, specificity and sensitivity of 0.9546, 0.9742 and 0.7527 respectively. Experimental results demonstrate that the proposed method achieved better performance than the state-of-the-art methods.