Affine Motion Vector Prediction for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as HEVC, face challenges in further improving video compression efficiency without degrading picture quality, particularly in handling complex motions that are not translational.
Innovation Solution
A motion vector prediction method based on an affine motion model is introduced, where motion vectors of control points from neighboring blocks are used to interpolate motion vectors for subblocks of the current block, reducing storage needs and improving prediction accuracy by using only subblock motion vectors for motion compensation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If motion vectors of control points are stored and used for motion compensation, then motion compensation can be performed, but storage requirements increase and prediction accuracy decreases for non-translational motions
Solution Approach 1:
The patent extracts only the necessary motion information from control points by using their motion vectors solely for deriving subblock motion vectors through affine transformation, rather than storing and using control point motion vectors directly for motion compensation. This extraction principle reduces storage requirements while maintaining the essential motion prediction functionality.
Solution Approach 2:
The patent segments the motion prediction process into two distinct stages: first deriving control point motion vectors from neighboring blocks, then using these to calculate subblock motion vectors through affine transformation. This segmentation allows efficient use of motion information without storing redundant control point motion vectors for compensation.
2Reliability
If control point motion vectors are used for motion compensation, then compensation can be performed, but inconsistent motion compensation occurs leading to reduced prediction accuracy
Solution Approach 1:
The patent applies dynamic motion modeling by using affine transformation to calculate different motion vectors for each subblock based on control point motion vectors. This dynamic approach adapts motion compensation to local motion variations within each block, improving prediction accuracy for non-translational motions while maintaining systematic processing.
Solution Approach 2:
The patent performs preliminary calculation of control point motion vectors from neighboring blocks before deriving subblock motion vectors through affine transformation. This preliminary action ensures consistent motion compensation by establishing a systematic derivation chain that eliminates inconsistencies in motion vector application.
Data Source
AI summary
A motion vector prediction method based on an affine motion model and a device are provided. The method includes: obtaining one spatial reference block of a to-be-processed picture block; determining a plurality of preset subblock locations of the spatial reference block; obtaining motion vectors corresponding to preset pixel locations of the to-be-processed picture block that are extrapolated from motion vectors corresponding to the preset subblock locations; and obtaining motion vectors corresponding to a plurality of subblock locations of the to-be-processed picture block that are interpolated from the motion vectors corresponding to the preset pixel locations. According to this application, prediction accuracy in coding can be improved, and coding efficiency can be improved.