A Comparative Analysis of Application of Niblack and Sauvola Binarization to Retinal Vessel Segmentation

Abstract

This paper demonstrates a comparative analysis of two binarization techniques- Niblack and Sauvola algorithm in context with their applications to retinal vessel segmentation. Preprocessed images are applied with Niblack’s binarization. Sauvola’s binarization is also applied as modification to Niblack’s algorithm. Drawbacks of Sauvola algorithm are addressed by incorporating some changes in the original algorithm and by setting experimentally the value of its Dynamic Range and the constant k in an application oriented way. Some post processing steps are needed to get rid of background noise pixels, The result is compared with the ground truth images of DRIVE database. The method achieves 93.23% accuracy in case of Niblack’s algorithm application and 93.31% (without post processing) and 94.34% (with post processing) accuracy in case of Sauvola algorithm. The accuracy obtained is very encouraging as far as the simplicity of the method is concerned.