Coral Identification Information System

Abstract

We propose the Coral Identification Information System (CIIS) that addresses the need of marine experts and scientists for a semi-automatic large scale analysis of coral reefs from images taken by underwater cameras. Our information system aims to provide these users important statistics on the spatial size and distribution of coral types from image data in order to rapidly assess the health of coral beds. The system uses texture classification algorithms to identify Acropora and Porites which are the most abundant types of corals in the Philippines. CIIS has three components, namely, the texture classifier, the expert sourcing mobile application, and the web application. The classifier identifies the types of corals present in an image using a texture-based recognition algorithm. The mobile application is used as a tool for coral labeling by experts. The web application serves as a repository for coral images and as interface to the classifier engine. Images uploaded through the web application will first undergo segmentation process involving superpixelization and superpixel merging prior to texture analysis. Texture classification is then performed on the merged superpixels. In order to obtain very high quality labels for classifier training, we employ expert sourcing methodology where coral experts use an Android application on mobile phones to label the corals. For selected images, the experts identify the coral type and express the level of certainty of their answers. The web application is used by marine scientists for coral health assessment. This application will pass uploaded images to the image analysis engine for processing. When the processing of the images is done, reports such as types of corals present and percentage of coral cover will be generated.