Zero MV-based AMVP improvement considering block size and insertion position

Following the standardization of Versatile Video Coding (VVC) in July 2020, the Joint Video Experts Team (JVET) initiated the "Beyond VVC" exploration and is developing the Enhanced Compression Model (ECM) as reference software. Advanced Motion Vector Prediction (AMVP) is an inter-prediction tool designed to reduce transmitted motion information. ECM introduces modifications to AMVP, including non-adjacent MVPs and template matching-based refinement and reordering. However, AMVP in ECM has limitations, as zero motion vector (MV) is excluded from the refinement and reordering process. To overcome this limitation, this paper investigates an enhanced AMVP candidate list construction method that explicitly inserts zero MV between its temporal and non-adjacent candidates and applies a refined candidate pruning strategy. Experimental results demonstrate a 0.06% gain in the Y channel by inserting Zero MV between its temporal and non-adjacent candidates and a 0.07% gain by Zero MV inserted under restricted block size, compared to ECM-15.0 as an anchor. These results confirm that incorporating the Zero MV into the AMVP candidate list improves the accuracy of motion vector prediction and enhances compression efficiency without increasing encoding complexity.