Barbell Lifting Wavelet Transform for Highly Scalable Video Coding

This paper proposes a generic lifting technology (Barbell Lifting), where each predicting or updating signal is generated with a Barbell function instead of always a sample from single direction or bi-direction, to incorporate various motion alignment methods into temporal wavelet transform. The proposed lifting technology embraces fractional pixel motion alignment, variable block size motion alignment and overlapped block motion alignment into a framework. It also guarantees the perfect reconstruction in various Barbell functions even with fractional precision. Furthermore, the proposed lifting technology also solves several problems that extensively exist in the current 3D wavelet coding schemes, such as many-to-one mapping for the area with covered scenes, non-referred pixels for the area with uncovered scenes, ambiguities of inversed mapping in the updating stage. Replacing the conventional lifting with the proposed Barbell lifting, the 3D wavelet coding scheme achieves a high coding efficiency.