Improved image-based coral reef component classification

Abstract

Three coral reef components, living corals, dead corals, and sand or rubble, are classified using both color and texture as features. Two modes of classification, sequential and parallel input, are tested. In sequential mode, if-then statements are used to eliminate the possibility of one component belonging to a class by classifying an image first based on texture then by color. In parallel mode, neural networks and support vector machines are used as classifiers. The best combination of features and classifier is mean hue and saturation for color, local binary patterns for texture, and a neural network for recognition which yielded 86% success rate.