HoG Based Naive Bayes Classifier for Glaucoma Detection

Glaucoma is caused due to neuro degeneration of the optic nerve leading to vision loss. The Optic nerve that is responsible for transmission of visual information to the brain can be viewed in a fundus image. In this paper, a new approach for determination of the presence of glaucoma based on the features extracted from the fundus image is proposed. Histogram oriented features (HoG) are extracted from Region of Interest, that shows a significant difference between normal and glaucoma affected image. A Naive Bayes classifier is used to demonstrate the performance of extracted HoG feature against the transform domain features extracted from Wavelet, Contourlet, and Wavelet based Contourlet decomposition. The proposed HoG based decision support system is tested on publically available fundus image database. Experimental results indicate that HoG feature based Naive Bayes classifier outperforms the classifier based on transform domain features and yields an accuracy of 94%.