Multiple video objects extraction for MPEG-4 application

MPEG-4 is an object-oriented video compression standard, which allows content-based representation of a video sequence. This requires a decomposition of scenes into semantic video objects (VOs). Many algorithms had been developed to extract VOs with accurate boundaries, however, the issue of multiple objects is less addressed. We proposed to use a simple clustering method based on distance transform to group moving pixels into different VOs. The objects will then be tracked by using Hausdorff distance and their boundaries are refined to form VOs with accurate boundaries. Experimental results show that the algorithm successfully finds and tracks multiple objects in different video sequences.