An Adaptive Scheme for Two-stage Multi-hypothesis Prediction

Existing two-stage multi-hypothesis reconstruction (2sMHR) schemes deploy a pixel-domain multi-hypothesis (MH) prediction after a measurement-domain recovery to break the limit of flexibility in reconstruction. However, these schemes for 2sMHR are sensitive to different types of videos because they select reference frames non-adaptively in pixel-domain recovery. To address this problem, we propose an adaptive scheme for 2sMHR in which we hold the assumption that the most accurate reference frame has the most significant weight. Specifically, we firstly integrate candidate reference frames into the same hypothesis set and utilize L1 norm to select the preferred frame as the reference to execute a secondary reconstruction. Simulation results demonstrate that the proposed scheme outperforms the state-of- the-art schemes in terms of stability.