Video object segmentation for content-based video coding

This paper deals with automatic object segmentation method towards object-oriented video encoding. Since mathematical morphology efficiently deals with geometrical features such as size, contrast, or connectivity which is important to segmentation-oriented features, intra-frame segmentation is first performed using morphological tools. After obtaining the different areas in intra-frame segmentation stage, some video object areas can be merged according to their motions in the merge stage. For this purpose, a new criterion is proposed to measure the difference between the estimated local motion and the global motion. A two-dimensional binary model can be obtained from the object of interest. At the refine stage, the border pixels can be absorbed into one of the neighboring objects which have the minimum prediction error computed from its model parameters. Experimental results demonstrate successful segmentation of objects in the case of moving or stationary background.