Stereoscopic imaging for coral segmentation and classification
M. Johnson-Roberson · Suresh Kumar · O. Pizarro · Stefan Willams
TL;DR
Adding stereo-derived surface shape and roughness to a 2D pipeline improves coral classification and moves toward real-time underwater survey processing.
BRIEF
Identifying coral in underwater photographs is hard when different species look alike in two-dimensional images. This work adds range data from stereo cameras to an existing 2D classification pipeline. From the segmented coral, it estimates rugosity (surface roughness) and surface topography, which helps tell apart species with similar appearance but different shapes. The approach supports automatic processing of survey datasets and a faster path to scientific results. The stated aim is a real-time system that can segment and classify coral.
Abstract
Stereoscopic imaging as a means of aiding the segmentation and classification of coral in underwater imagery. This work presents a framework for integrating range data into a formerly two dimensional classification pipeline. By taking an estimate of the rugosity and surface topography of segmented coral greater accuracy can be achieved in morphological discrimination of visually similar species. Utilizing these techniques allows for automatic processing of datasets and quicker turnaround in achieving the scientific goals of a survey. This work strives to produce a real-time system that is capable of segmenting and classifying coral