Improved classification of glaucoma in retinal fundus images using 2D-DWT

Glaucoma is one of the leading causes of vision impairment. The screening of glaucoma in the earlier stage is crucial to avoid blindness. In this paper, an improved method is proposed for glaucoma screening in retinal fundus photographs using a two-dimensional discrete wavelet transform (2D-DWT). In this work, 2D-DWT is used for the decomposition of the preprocessed images into various sub-band images (SBIs). Further, the significant grey level co-occurrence matrix (GLCM) features have been computed from SBIs. Then, the ReliefF algorithm has been utilized to select the relevant features from the extracted feature set. Finally, these robust features have been used for classification using the K-nearest neighbor (KNN) classifier. The developed framework obtained the maximum accuracy of 92.10% with a tenfold cross-validation process.