Hierarchical segmentation-based software for cover classification analyses of seabed images (Seascape)

Abstract

An important aspect of marine research is to quantify the areal coverage of benthic communities. It is technically feasible to efficiently obtain images of marine environments at different depths and benthic habitats over large spatial and temporal scales. Currently, there is a large and growing library of digital images to analyze, representing a valuable benthic ecological archive. Benthic coverage is the basis of studies on biodiversity, characterization of communities and evaluation of changes over temporal and spatial scales. However, there is still a lack of automatic or semi-automatic analytical methods for deriving ecologically relevant data from these images. We introduce a software program named Seascape to obtain semi-automatically segmented images (patch outlines) from underwater photographs of benthic communities, where each individual patch (species/categories) is routinely associated to its area cover and perimeter. Seascape is an analog to the classical and better known discipline of landscape ecology approach, which focuses on the concept that communities can be observed as a patch mosaic at any scale. The process starts with a hierarchical segmentation, using a color space criteria adapted to the problem of segmenting complex benthic images. As an endproduct, we obtain a set of images ­segmented into classified homogenous regions at different resolution levels (hierarchical seg­mentation). To illustrate the versatility and capacity of Seascape, we analyzed 4 digital images from different habitats and depths: coral reefs (Pacific Ocean), coralligenous communities (NW Mediterranean Sea), deep-water coral reefs (NW Mediterranean Sea) and the Antarctic continental shelf (Weddell Sea). The development of this semi-automatic outline tool and its use for classification constitute an important step ­forward in the analysis and processing time of underwater seabed images at any scale