Spatial-temporal segmentation scheme for object-oriented video coding based on wavelet and MMRF

In this paper, we introduce an object segmentation algorithm that can be used in object oriented video coding, computer vision, target tracking or video sequence analysis. The algorithm is based on two kinds of information from a video sequence: the color segmentation which provides the spatial domain information and the motion estimation which provides the temporal information. The DWT based MRF segmentation algorithm, which can avoid oversegmentation problems, is used for color segmentation. The quadtree: motion estimation algorithm which employs the color segmentation results and DWT coefficients is used for motion segmentation. Finally, an object extracting procedure will be used to combine the spatial-temporal information and extract an object out.