Non-separable 2-D and 3-D Discrete Wavelet Transform for image and video processing using lifting scheme

In this paper, lifting based Non-Separable 2D and 3D Discrete Wavelet Transform (DWT) is presented. Under the proposed lifting scheme with non-separable filters, several advantages are gained particularly in hardware implementation. The proposed algorithm results in a significant reduction in the number of computations as compared with convolution-based non-separable wavelet transform. This approach does not require any transposition and has a better SNR with a narrower bus bandwidth. It is compatible with applications requiring 2D scanning such as vector quantization. In addition to eliminating the transposition stage, it has a lower latency. Another attractive characteristic of lifting scheme with non-separable discrete wavelet transform is the simultaneous generation of all of four subband coefficients, approximations, diagonal details, vertical details, and horizontal details.