Efficient video object segmentation based on Gaussian mixture model and Markov random field

This paper proposes an efficient video object segmentation approach based on Gaussian mixture model (GMM) and Markov random field (MRF). The user-interested video objects are interactively extracted in the first frame of the video sequence, and each video object and the remaining background are represented by individual GMM, which is initialized based on the region segmentation result used for interactive object extraction. For each following frame, two GMM classification results are respectively generated based on only color feature, and both color feature and position feature, which is compensated by the estimated average position change to adapt to fast moving regions. Based on the initial pixel classification result generated from the two GMM classification results and the corresponding confidence measures, the pixel classification result is refined to obtain a reliable video object segmentation result under the MRF framework. Experimental results on several MPEG-4 test sequences demonstrate the good segmentation performance of the proposed approach.