Segmentation of underwater video objects using Extended Markov Random Field Model

Tracking of Underwater object has been a challenging task because of uncontrolled illumination condition. The problem is compounded due to the movement of both camera and object. In this paper, we address this incomplete data problem while simultaneously estimating the camera position and image labels. This has been achieved using Expectation Maximization (EM) algorithm and Extended Markov Random Field (E-MRF). We have extracted oriented weighted features from different frames in different scales of the image. The estimation of the camera parameters have been achieved based on the notion of pipelining. The proposed scheme has been tested with the underwater video sequence obtained from www.worldnaturevideo.com data base. This has also been compared with the algorithm proposed by Stolkin et al. (2008).