Color-texture image analysis of coral reefs

Abstract

Marine scientists assess the condition of coral reefs from estimates of the population of living corals, dead corals, co-existing animals in the reefs, algae, and rocks. Some popular methods used are Line Intercept Transect (LITR) and In-Situ Mapping (ISMP) where both employ a diver to note in situ the population in a reef of interest though of different techniques of sampling. A rapid and consistent method used since 1992 is the Video Recording (VIDR). The area of interest is filmed and then processed by a method called Video Point Sampling (VIDS). Expert individuals estimate percent cover through random or fixed points placed on the monitor screen while the film is paused at random or even intervals and identifying the items at the sampling points on an appropriate benthos category (e.g. alive coral, dead coral, rock). Software called PointCount'99  also uses the random point-count method but still requires user-intervention for classification. Assessment in the said methods is visual, requiring a trained eye and experience. By automating the assessment with a computer, the analysis is more precise, less subjective and less tedious.
In computer vision, several techniques for image classification using color and texture have been developed. Color is a property of one point in the digitized image. Pattern recognition techniques using color often operate on the color distribution alone, ignoring the spatial, black and white (tonal) property of regions in the image which can be provided by the texture. Recently, these two features were combined as one feature called color-texture, or the spatial distribution of colors in a region. Coral reefs have color and texture and regular 3D structures, which are cues used by scientists for classification. The objective now is to investigate existing color-texture paradigms and 3D texture classification techniques to classify images of coral reefs.