Approximate Optimization of Gabor Filter Parameters in Application to Blood Vessel Segmentation in Retinal Images

Abstract

The Gabor filter responses at different orientations are useful for blood vessel segmentation (BVS) in retinal images. Instead of using all possible Gabor filter responses, the use of maximum (optimum) Gabor filter response at every orientation is a better solution to reduce computational complexity of BVS methods. To achieve maximum Gabor filter response, one has to perform filtering with all possible Gabor filters by changing parameters of the filter and from that, a Gabor filter that provides the maximum response has been selected. The existing methods calculated the maximum response by applying the hundreds of Gabor filters (filter bank) on the input retinal image. This increases the computational complexity of BVS methods. Hence, this paper provides a single set of values for the parameters of the Gabor filter to get an approximate optimum response instead of optimum response. The proposed method facilitates a single set of approximate optimum parameters to Gabor filter at every orientation and thus reduces the computational complexity. This concept is described by using a proposed BVS method, which achieves 95.05% average accuracy on DRIVE dataset. The percentage of error in accuracy between optimum and approximate optimum responses is 0.23%. The experimental results show superiority as compared to the existing methods in the literature.