Wavelet-based video coding with early-predicted zerotrees

The coding efficiency of DCT-based standard video codecs is significantly improved by the use of skipped macroblocks. A similar concept is proposed for zerotree-based wavelet video coders. In a pyramidal wavelet transform, the wavelet coefficients are linked through spatial orientation trees. For each transformed frame, a method is proposed to identify those trees that are likely to be zerotrees throughout the coding process at a given bit budget. These trees are called the early-predicted zerotrees. These trees are then excluded from the subsequent quantisation and encoding processes. The tree classification is based on average energy of coefficients in the tree. The proposed technique is general and can be applied to any zerotree-based coding algorithm, but it is more advantageous to improve the performance of those zerotree-based video coding algorithms that use longer trees, such as virtual set partitioning in hierarchical trees (V-SPIHT). Simulation results demonstrate that the proposed technique can improve the performance of V-SPIHT-based video coder by up to 0.5 dB (on average) and can reduce the computational complexity by up to 80% (i.e. five times faster), for various test video sequences