Affine Motion Derivation from Refined Subblock Vectors in Video Coding

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing video coding techniques face inefficiencies in inter prediction, particularly in deriving accurate affine motion models, leading to suboptimal coding efficiency and increased distortion.

Innovation Solution

The method involves partitioning a coding block into subblocks, generating initial subblock motion vectors based on affine merge candidates, refining these vectors using decoder-side motion vector refinement, and applying linear regression to derive a more accurate affine motion model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional inter prediction methods are used without affine motion modeling, then device complexity is reduced, but coding efficiency deteriorates and distortion increases

Engineering Contradiction:

Improvecoding efficiencyVSAvoidcomplexity of motion model derivation

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The coding block is divided into multiple subblocks, each with its own motion vector. This segmentation allows the affine motion model to be derived from multiple local motion vectors rather than requiring a complex global model, thereby improving coding efficiency while keeping the derivation process manageable through systematic partitioning

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Initial subblock motion vectors are derived first using affine merge candidates from neighboring blocks before the linear regression process. This preliminary derivation provides a starting point that captures basic motion information, and the subsequent regression refines this information to obtain the final affine motion model, improving accuracy without requiring the entire process to be complex from the start

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If initial subblock motion vectors are used without refinement, then processing time is reduced, but measurement precision of motion vectors deteriorates

Engineering Contradiction:

Improveprecision of motion vectorsVSAvoidtime for motion vector refinement

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The linear regression process uses the refined subblock motion vectors to derive control point motion vectors, which then serve as feedback to improve the initial motion vector estimates. The regression analyzes the relationship between subblock positions and their motion vectors to generate corrected control point motion vectors, thereby improving measurement precision through iterative refinement

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The motion vectors are refined by changing parameters through the linear regression process. The initial motion vectors serve as one set of parameters, and the regression process transforms these into improved parameters (control point motion vectors) that better represent the actual motion, achieving higher precision through parameter transformation rather than requiring additional time-consuming search processes

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If a simple motion model is used, then device complexity is reduced, but manufacturing precision of motion compensation deteriorates

Engineering Contradiction:

Improveprecision of motion compensationVSAvoidcomplexity of affine motion model

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

By segmenting the block into subblocks and deriving motion vectors for each subblock independently, the system achieves more precise motion compensation through local motion modeling. The affine motion model is then constructed from these segmented motion vectors, allowing high precision without requiring an overly complex monolithic model

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent replaces complex mechanical motion estimation processes with a mathematical linear regression approach. Instead of using complex iterative optimization or multiple search processes, the system uses linear regression to transform initial motion vectors into refined control point motion vectors, achieving high precision through mathematical transformation rather than complex computational machinery

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12563198B2Regression-based decoder side affine motion derivation for video coding

Publication Date: 2026.02.24 QUALCOMM INC

AI summary

A video coder may be configured to partition a coding block into subblocks, and generate initial subblock motion vectors for a plurality of the subblocks. The video coder may further refine the initial subblock motion vectors for the plurality of the subblocks using decoder side motion vector refinement to produce refined subblock motion vectors for the plurality of the subblocks, and perform a linear regression on the refined subblock motion vectors and coordinates of the plurality of the subblocks to derive an affine motion model. The video coder may then code the coding block using the derived affine motion model.