Efficient video coding based on backward coding of wavelet trees

The authors present a wavelet tree-based video coding system, which is referred to as three-dimensional (3-D) backward coding of wavelet trees (BCWT). The 3-D BCWT algorithm encodes the wavelet trees from the lowest wavelet level to the highest wavelet level. All the coding steps for each wavelet coefficient are finished during one coding pass, which makes this algorithm very efficient. The coding process of the 3-D BCWT algorithm is subband by subband, so that a coefficient will be encoded only after all the coefficients in the next lower-level wavelet subbands have been encoded. The output bitstream of the encoder is therefore layered according to the wavelet level. The implementation results show that the 3-D BCWT algorithm has almost identical rate distortion performance to the well-known 3-D set partitioning in hierarchical trees (SPIHT) algorithm. However, it is computationally more efficient and uses significantly less memory than the 3-D SPIHT algorithm.