Segmenting video images based on joint region and contour information

摘要

Segmenting video sequences based on joint region and contour information uses a motion segmentation algorithm to separate moving objects from background, and then generates a contour around the object. The contour is initially determined binarily using minimization of an energy function having continuity, rigidity and balloon terms to shrink the contour to the separated objects from an initial position, to eliminate sharp corners and to push/pull contour nodes to the boundaries of the object. The resulting binary contour is further refined using a luminance gradient term to take into account the luminance of the image, and using a chrominance gradient term to provide final adjustment of the contour. A shape template for the object is generated from the contour by filling the interior of the contour with a constant value, such as white. <IMAGE>