Vessel centerlines detection by Neuro-Fuzzy and Wavelet multiscale product in digital ratinal images

摘要

This paper presents a new approach of retinal centerlines detection with a combination of Neuro-Fuzzy (NF), Multi-scale Product (MSP) of Wavelet transform (WT) and Mathematical morphology (MM) tools. The proposed approach has an accuracy of 94,14% obtained on DRIVE Database. We started by operating a supervised Neuro-Fuzzy of the image and edges extraction by multi-scale product (MP) of three bands wavelet transform. Local Binary Patterns is operated on the supervised Neuro-Fuzzy. These attributes are used for a morphological reconstruction to impose minima in the NF attribute. Morphological processing with an alternating sequential filter is applied to minima. Edge reconnections is realized with the NF attribute by adding pixels having the characteristic of the same belonging degree as minima pixels in a 5×5 neighborhood window. Finally skeleton is performed on image of minima reconnected.