Underwater Object Detection by Combining the Spectral Residual and Three-Frame Algorithm

In this paper, a hierarchical system, in which each level is composed by a specialized processor, is proposed to detect objects in underwater videos. The system is designed to assist the underwater monitor system survey operations, specialized to the task of object detection. The input image is firstly transformed into the spatial frequency domain to detect the saliency regions including the object region and parts of the background. Then, the saliency maps are further analyzed by the frame-difference algorithm which contributes to the background noise reduction. Experimental results, which have been performed on a set of real underwater images acquired in different environments, demonstrate the robustness and the accuracy of the proposed system.