Contour-based multi-source information fusion for motion segmentation

A motion segmentation framework that effectively exploited the multiple sources of image information and fused these sources of the information synergistically was proposed to serve the purpose of motion segmentation. A Markov process was formulated for motion segmentation in which two feature spaces were established to estimate the state transition Probability density function (PDF) and the initial state, respectively. An information fusion space was developed such that each motion structure was described as a single distribution in this space. The proposed framework can naturally embed the evolution equations of the active contour methods into the segmentation to achieve contour-based segmentation results. Extensive empirical evaluations demonstrate the robustness and the promise of this framework.