Lifting-based wavelet transform for video coding

Abstract

In order to overcome the drawbacks of the conventional wavelet transform that does not provide a compact representation of edges which are not in horizontal and vertical directions, a number of approaches have been proposed in the literature. Lifting is an efficient algorithm to implement the discrete wavelet transform. It has been adopted in JPEG2000. The adaptive directional lifting (ADL)-based wavelet transform for image coding was proposed recently. The ADL scheme incorporates the directionally spatial prediction into the conventional lifting-based wavelet transform and forms a novel, efficient and flexible lifting structure. Although the ADL scheme is very general, it can significantly outperform the conventional 2D DWT scheme up to 2 dB as reported. The principle of ADL is very similar to the principle of motion compensated temporal filtering (MCTF) used in motion compensated 3D wavelet video coding. Therefore the problems existing in MCTF also exist in ADL. One of the issues is that there is an inconsistent problem between the predict step and update step. In order to resolve this inconsistent problem, we examine the performances of four update schemes to resolve the inconsistent problem between the predict step and update step. Experimental results show that energy distributed update consistently outperforms the update step using inverse directional angles. In the past few years, context-based adaptive binary arithmetic coding( CABAC ) is widely used in state-of-the-art image and video codec and usually a superior coding performance can be achieved, we propose a new directional angles coding scheme that uses CABAC algorithm to code differential directional angle by taking the direction information into account. We design a context model for encoding the bins resulting from binarizing differential directional angles. Experimental results show that the use of CABAC instead of the MAAC gives a bit-rate saving up to 16.67% for coding the first level directional angles. The bit-rate reduction by the use of CABAC instead of the MAAC for coding all levels of directional information is 12.5%. The CABAC scheme outperforms the MAAC up to 0.1 dB in terms of PSNR. In conventional motion compensated 3D wavelet video coding scheme, where the group of picture size and lowpass frames position are fixed, motion activities in video sequences are not considered. We propose a haar-like motion compensated 3D wavelet video coding scheme, which can adaptively select the group of picture size and low-pass frames based on mutual information of the video sequences and the types of motion activities. When the video sequence has high motion activity, the proposed low-pass frame selection has a better performance than that of the conventional scheme. The proposed adaptive approach for selecting the group of picture size is more suitable for situations with un-even variation of motion activities, e.g. with frequent shot cuts, and when the decoder only decodes frames at the low temporal level. A new motion compensated 3D wavelet video coding scheme is proposed. The proposed motion compensated 3D wavelet video coding scheme is called adaptive directional motion compensated 3D wavelet video coding scheme. It combines adaptive directional lifting(ADL) with motion compensated 3D wavelet video coding. The ADL can be considered as an efficient intra frame coding scheme to code the lowpass frame. This can significantly improve the overall coding efficiency of a video GOP because of the accurate spatial prediction ability inherent in ADL. Experimental results show that the proposed adaptive directional 3D DWT consistently outperforms the conventional MC 3D DWT for scalable video coding, and the improvement in PSNR is up to 1 dB.