Automated Tessellated Fundus Detection in Color Fundus Images

In this work, we propose an automated tessellated fundus detection method by utilizing texture features and color features. Color moments, Local Binary Patterns (LBP), and Histograms of Oriented Gradients (HOG) are extracted to represent the color fundus image. After feature extraction, a SVM classifier is trained to detect the tessellated fundus. Both linear and RBF kernels are applied and compared in this work. A dataset with 836 fundus images is built to evaluate the proposed method. For linear SVM, the mean accuracy of 98% is achieved, with sensitivity of 0.99 and specificity of 0.98. For RBF kernel, the mean accuracy is 97%, with sensitivity of 0.99 and specificity of 0.95. The detection results indicate that color features and texture features are able to describe the tessellated fundus.