Two-Dimensional Variational Mode Decomposition with Texture Feature Extraction for Glaucoma Classification from Retinal Images

Abstract

Image decomposition is very important for glaucoma classification from retinal images. Conventional methods can extract features, but the performance of those methods is insufficient because of loss information from the decomposition step. In this paper, 2D-VMD with texture feature extraction is proposed for classifying glaucoma. It decomposes a retinal image into different frequency sub-images by means of two-dimensional variational mode decomposition due to adaptive decomposition according to its data. Texture features are extracted by using GLCM with statistical approaches. Significant texture features are selected with high t-test values. From 1,544 retinal images in the Harvard dataverse dataset, the proposed method achieves 98.19%, which is higher than the conventional methods. Our method can extract the significant texture features with high accuracy, improving the performance of glaucoma classification.