Retinal Vessel Segmentation using Gabor Filter and Textons

This paper presents a retinal vessel segmentation method that is inspired by the human visual system and uses a Gabor filter bank. Machine learning is used to optimize the filter parameters for retinal vessel extraction. The filter responses are represented as textons and this allows the corresponding membership functions to be used as the framework for learning vessel and non-vessel classes. Then, vessel texton memberships are used to generate segmentation results. We evaluate our method using the publicly available DRIVE database. It achieves competitive performance (sensitivity (TPF) =0.7673, specificity (1-FPF) =0.9602, accuracy=0.9430), and especially its considerably higher sensitivity means it is better in detecting vessels than the other methods. These figures are particularly interesting as our filter bank is quite generic and only includes Gabor responses.