An Improved Framework of Affine Motion Compensation in Video Coding.

Area of Science:

• Video Compression
• Digital Signal Processing
• Computer Vision

Background:

• Affine Motion Compensation (AMC) is a key tool in Joint Exploration Model (JEM) for video coding.
• Current AMC in JEM uses a 4-parameter affine model to derive sub-block Motion Vectors (MV) from control points.

Purpose of the Study:

• To present a practical framework for improving AMC in JEM.
• To enhance coding efficiency and performance of AMC in video compression.

Main Methods:

• Introduced a multi-model AMC approach with adaptive selection between 4- and 6-parameter affine models.
• Improved affine inter-mode by developing an efficient affine motion coding method with a second-order MVP and enhanced motion estimation.
• Unified affine merge-mode and normal merge-mode into a single merge candidate list.

Main Results:

• Proposed methods demonstrated an average of 1.67% BD rate savings for random access configurations.
• Some proposed techniques have been incorporated into the Versatile Video Coding (VVC) standard.

Conclusions:

• The developed framework significantly improves AMC performance in JEM.
• The enhancements contribute to more efficient video compression standards like VVC.