Classification of Underwater Color Images with Applications in the Study of Deep Coral Reefs

Abstract

Coral reefs ecosystems have been impacted by natural and anthropogenic effects resulting in a decline of coral communities worldwide. This decline in coral reefs has an economical impact in tourist areas, and marine ecosystems. Coral reef scientists and resource managers monitor and map coral reefs communities manually. Automated techniques are nonexistence, especially in deep waters where the absorption and scattering properties of the water do not allow the use of satellites. In these cases, other imaging platforms like autonomous underwater vehicles (AUV) are needed. This work presents a prototype classification algorithm with applications in the study of deep coral reefs taken by the SeaBED (AUV) at the Hind Bank Marine Conservation District (MCD), south of Saint Thomas, U.S. Virgin Islands. Because of light conditions, the images acquired by this AUV have low contrast, are very noisy, and are extremely rich in both spatial variability and texture, making the automated classification a very difficult task [Rivera Maldonado Francisco J, 2004; River F.J. et al., 2004], The classification algorithm developing in this research use the local homogeneity coefficient segmentation algorithm [Rivera Maldonado Francisco J, 2004; River F.J. et al., 2004] as first stage to find the different regions of interest in the images like corals, bare substrate, and sand among others classes. Combining a pixel by pixel Euclidean classification in each region with some texture features the classification of each region is performed. Finally, the results of the algorithm are validated with the traditional manual classification done for this type of applications.