Selection of optimal wavelet packet transforms for object-based coding

An algorithm for optimal selection of wavelet packet transforms for object-based coding is presented in this paper. The method exploits the object statistics at each GOP or scene period and estimate an optimal wavelet packet tree and an optimal quantizer for each object so that a) objects of importance (e.g., foreground) are transmitted with higher bit rate (quality) than the background objects, while b) the total transmitted rate does not violate the constraints defined by the network requirements. Furthermore, since the proposed approach estimates object statistics over a GOP or a scene period (an optimal wavelet packet is assigned to each object), the algorithm does not result in a simple bit re-allocation from the background to foreground object but also increases object quality for the same bit rate.