Flexible bi-directional triadic MCTF with multi-hypothesis update

In this paper, we present a flexibly adaptable lifting scheme of motion-compensated temporal filtering (MCTF) that is based on a triadic decomposition structure and allows for temporal scalability by a factor of 3. In contrast to previous publications on this topic, our proposed framework employs bi-directional prediction and update operations, and rate-distortion optimized, locally adaptive motion estimation. Additionally, we propose a multi-hypothesis update step, discuss the optimization of the filter coefficients, and compare our results to the well-known dyadic lifting architecture using 5/3 filters.