Foreground segmentation of binocular video based on multiple cues

This paper proposes an automatic foreground segmentation algorithm. We develop an energy function that is composed of disparity, color and contrast cues to extract the foreground object accurately. Disparity is calculated to contribute the energy function with simple and effective methods. Background color model is constructed based on Gmms algorithm. Continuity characteristic is used to get the contrast cue. A binary label is assigned to each pixel by minimizing the designed energy function. A boundary smooth filter in frequency domain is presented in order to paste the foreground object into the virtual background seamlessly. Experimental results demonstrated that our segmentation algorithm based on multiple cues is efficient.