3D adaptive wavelet packet for video compression

This paper presents an efficient temporal, spatial and frequency decomposition method that is designed for video coding. In this method, the given video sequence is decomposed into small volumes such that the signal in each volume is relatively homogeneous in frequency characteristics in all three dimensions. Volumes thus formed, therefore, adapt to the spatial nonstationarity and the motion in the video signals. For each volume, a best 3D wavelet packet is generated based on a rate-distortion criterion such that the subbands of this 3D wavelet packet are efficiently quantized. The partitioning of the total signal volume into smaller "homogeneous" volumes is realized according to the same rate-distortion criterion.