Motion Vector Coding and Block Merging in the Versatile Video Coding Standard

Wei-Jung Chien, Li Zhang, Martin Winken, Xiang Li, Ru-Ling Liao, Han Gao, Chih-Wei Hsu, Hongbin Liu, Chun-Chi ChenIEEE TCSVT 2021 · DOI

Abstract

This paper provides a comprehensive analysis of the "whole-block" inter-prediction machinery within the Versatile Video Coding (VVC) standard. By moving beyond the rigid block structures of its predecessor, HEVC, VVC achieves significantly higher compression efficiency through a sophisticated suite of motion vector coding and block merging tools. These techniques alone provide an average bitrate reduction of 6.2% for random access video while maintaining high subjective visual quality.

Introduction

Video compression relies heavily on inter-prediction, which exploits temporal redundancy by predicting the current frame from previously coded ones. The High Efficiency Video Coding (HEVC/H.265) standard utilized a relatively rigid framework where motion was represented by rectangular blocks. However, real-world objects rarely move in perfect squares. As content moves toward 4K, 8K, and HDR, the bits spent describing motion vectors (overhead) become a significant bottleneck. The challenge addressed here is representing complex motion more accurately using fewer bits without compromising decoder speed for consumer hardware.

Methods

The paper focuses on whole-block inter-prediction, where a single set of motion information is applied to an entire coding unit (CU). The advancements are categorized into three main areas: Motion Vector (MV) Coding, Advanced Block Merging, and Hybrid Tools.

Motion Vector (MV) Coding Improvements

VVC introduces several tools to optimize the signaling of motion vectors:

  • AMVR (Adaptive MV Resolution): Instead of a fixed high-precision resolution, VVC allows switching between 1/16-pel and 4-pel resolution, saving bits on blocks with simple motion.
  • SMVD (Symmetric MV Difference): For bi-predictive coding, this signals only one motion vector difference and derives the other symmetrically, halving signaling overhead for linear motion.
  • AMVP (Advanced Motion Vector Prediction): This process derives candidates from spatial and temporal neighbors to predict the current block's motion.
Spatial and temporal candidate positions for AMVP list construction.

Advanced Block Merging

Merging tools allow blocks to inherit motion information from neighbors, significantly reducing bitrate:

  • HMVP (History-based MV Prediction): VVC maintains a FIFO table of recently used motion vectors, allowing blocks to "remember" motion from non-adjacent parts of the frame.
  • GPM (Geometric Partitioning Mode): GPM allows a block to be split into two non-rectangular parts using an angled line, better fitting the contours of moving objects.
  • MMVD (Merge with MV Difference): This adds a small refinement offset to a selected merge candidate, bridging the gap between low-bitrate Merge mode and high-precision AMVP mode.
Flowchart of the merge candidate list construction process in VVC.

Hybrid Tools

  • CIIP (Combined Inter-Intra Prediction): Blends inter-predicted and intra-predicted signals for regions where motion compensation is imperfect.
  • BCW (Bi-prediction with CU Weights): Allows for unequal weights in bi-prediction (e.g., 3/8 and 5/8) to handle fades and lighting changes more effectively.

Key Insights and Results

Evaluation using the JVET Common Test Conditions (CTC) on UHD and screen content (Classes A-F) via the VTM-9.0 reference software yielded several key findings:

  • Efficiency Gains: The combined whole-block tools achieved a 6.2% BD-rate saving in Random Access (RA) and 4.7% in Low-Delay (LDB) configurations compared to a baseline with these tools disabled.
  • Impactful Tools: AMVR and GPM were the most impactful individual tools, each contributing up to 1.6% in bitrate savings.
  • Visual Quality: GPM significantly reduces "sawtooth" distortion on moving diagonal edges, improving subjective quality.
  • Complexity Management: While encoding time roughly doubled, the increase in decoding complexity was minor, ensuring compatibility with battery-powered playback devices.

Conclusion

VVC's whole-block inter-prediction tools represent a major leap in compression efficiency by breaking the locality constraints of traditional codecs and moving toward more natural geometric representations. While the increased encoder complexity remains a challenge for real-time applications, the significant bitrate savings and maintained decoder efficiency solidify VVC's role in the future of high-resolution video delivery.

Summary updated

BibTeX
@article{chien2021motion,
  author = {Wei-Jung Chien and Li Zhang and Martin Winken and Xiang Li and Ru-Ling Liao and Han Gao and Chih-Wei Hsu and Hongbin Liu and Chun-Chi Chen},
  title = {{Motion Vector Coding and Block Merging in the Versatile Video Coding Standard}},
  journal = {IEEE Trans. Circuits Syst. Video Technol.},
  year = {2021},
  doi = {10.1109/tcsvt.2021.3101212},
}
Figure