A Novel Retinal Vessel Segmentation Based On Histogram Transformation Using 2-D Morlet Wavelet and Supervised Classification

TL;DR

This paper introduces a new retinal vessel segmentation method combining histogram transformation, Morlet wavelet features, and Bayesian classification, achieving high accuracy on the DRIVE database.

Contribution

It presents a novel preprocessing and feature extraction approach using Morlet wavelet responses combined with Bayesian classification for improved retinal vessel segmentation.

Findings

Achieved 95.71% accuracy on DRIVE database

Effective noise filtering with Morlet wavelet

Improved vessel continuity with morphological transforms

Abstract

The appearance and structure of blood vessels in retinal images have an important role in diagnosis of diseases. This paper proposes a method for automatic retinal vessel segmentation. In this work, a novel preprocessing based on local histogram equalization is used to enhance the original image then pixels are classified as vessel and non-vessel using a classifier. For this classification, special feature vectors are organized based on responses to Morlet wavelet. Morlet wavelet is a continues transform which has the ability to filter existing noises after preprocessing. Bayesian classifier is used and Gaussian mixture model (GMM) is its likelihood function. The probability distributions are approximated according to training set of manual that has been segmented by a specialist. After this, morphological transforms are used in different directions to make the existing discontinuities uniform on the DRIVE database, it achieves the accuracy about 0.9571 which shows that it is an accurate method among the available ones for retinal vessel segmentation.