Incorporating phase information for efficient glaucoma diagnoses through hyper-analytic wavelet transform

Abstract

A technique for the automated glaucoma detection, from the fundus images, using hyper-analytic wavelet transform is presented. The hyper-analytic wavelet transform of the images is done to enhance the directional selectivity and to preserve the phase information. Before the transformation, the adequate preprocessing steps such as grayscale conversion and histogram equalization are carried out. Subsequently, the magnitude and phase spectra of the images are evaluated. Mean, energy, entropy, entropy-square, and entropy-cube are the parameters extracted from both the magnitude and phase spectra. Also, the circular mean parameter for relative phase features (within relevant sub-bands) is extracted. The magnitude and phase feature matrices are the features reduced by measuring every feature’s individuality among the feature set and the differentiating capability of every feature between the two classes. The feature matrices are evaluated in terms of accuracy. The evaluated metric is compared between magnitude feature matrix, and magnitude cum phase feature matrix. The classification accuracy is raised to 5% by incorporating the phase, relative phase information. The training of neural network is done using the conjugate gradient descent algorithm for classification.