Multi-directional Texture Feature Extraction for Glaucoma Classification from Color Retinal Images

Abstract

Glaucoma is a type of eye disease in which there are no symptoms. As the disease progresses for a long time, the patients lose their visions permanently. An automatic glaucoma classification is essential for identifying early glaucoma diagnosis from retinal images to reduce the vision loss risk. Image analysis and machine learning techniques are applied to create an automatic method for glaucoma classification from retinal images. To increase the effectiveness of glaucoma classification, a multi-directional texture feature extraction (MTFE) is proposed. This method extracts texture features on segmented red, green, and blue images without background. Its extraction is based on 2D Gabor filters with SURE entropy to efficiently provide appropriate texture features with low dimensions for classifying glaucoma. For 455 images of the RIM-ONE R2 database, the MTFE method yields 90.44%, which is higher than other methods. The MTFE proposed method can extract proper texture features, thus improving the accuracy of glaucoma classification from retinal images.