Saliency based shape extraction of objects in unconstrained underwater environment

Un-manned underwater exploration in unconstrained environment is a challenging and non-trivial problem. Manual analysis of large volume of images/videos captured by the underwater stations/vehicles is a major bottleneck for the underwater research community. Automated system for analyzing these videos is need of the hour for exploring the underwater space. In this paper, we present a method for extracting the shape of the objects present in the unconstrained underwater environment scenarios. The proposed method extracts the shape of the objects using saliency gradient based morphological active contour models. The uniqueness in the method is that the stopping condition for the active contour models is derived from the combination of saliency gradient with the gradient of the scene. As a result the proposed method is able to work in highly dynamic and unconstrained underwater environments. The results show that the proposed method is able to extract the shapes of the man-made as well as natural objects in these environmental conditions. The proposed method is able to detect shapes of multiple objects present in an underwater scene. The method is successful in extracting the shape of the occluded objects in such conditions. The results show that the proposed saliency gradient based morphological GAC extracts a minimum of 63% and average of 90% of the objects with misclassification rate of 4% whereas the saliency gradient based morphological ACWE extracts a minimum of 62% and average of 85% of the objects with misclassification rate of 4%.