The new video&audio coding standard MPEG-4 is enabling content-based functionalities. It takes advantage of a decomposition of sequences into semantically meaningful video object (VO). A snapshot of a VO is named video object plane(VOP) at a given time and a series of VOPs represent one moving object. This is a very challenging task

because physical objects are normally not homogeneous with respect to low-level features such as color

luminance

or optical flow. Hence

conventional segmentation algorithms will fail to obtain meaningful partitions. In this paper

a new automatic VO segmentation algorithm based on modified Hausdorff object tracker is presented. A binary model for moving object is automatically derived and tracked in subsequent frames using the modified Hausdorff distance. First initial model is extracted and tracked based on the proposed object tracker against subsequent frames in the sequence. Then the model is updated every frame to accommodate for rotation and changes in shape. The video object is extracted by a series of binary models in the end. Furthermore

to improve the quality of segmentation and to reduce the computational complexity

stationary background is filtered by a novel technique. Experimental results demonstrate the performance of our proposed algorithm.