Video compression using adaptive wavelet packet and DPCM

This paper presents an efficient temporal, spatial and frequency decomposition method for improving the performance of the existing methods for 3D subband coding of video signals. In this method, a given video sequence is first partitioned into constant-sized 'groups' of frames. Within each group, every two consecutive frames are decomposed into low and high temporal subbands by a two-tap temporal filter. The redundancy in the low temporal subbands is removed by a closed loop DPCM unit. The low and high subbands of the first pair of frames, the difference of the low subbands from the DPCM loop and the high subbands of the subsequent pairs of frames in the 'group' are then divided into image blocks of equal size, such that each block can be efficiently decomposed by an adaptive wavelet packet based on a rate-distortion criterion. The subbands of wavelet packets for all blocks are quantized by a hybrid scalar/pyramidal lattice vector quantizer. This scheme achieves minimum distortion under a rate constraint specified for the 'groups' of frames and under the structural constraints of the algorithm. In this scheme, the motion compensation is not used explicitly, but the effects of motion are accounted for through the low-high temporal subbanding and the DPCM procedure. The results compare favorably with those of traditional video coding techniques. Although the computational complexity of this scheme is higher than that of the existing 3D subband method, it is suitable for parallel implementation.