Affine Homographic Motion Field Rendering for Video Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video compression techniques struggle to accurately capture non-translational motion in video streams, leading to reduced motion compensated prediction accuracy.

Innovation Solution

The method involves generating a motion field-based reference frame by determining motion fields for blocks of a current frame using bi-directional reference frames, and then applying an affine homographic model to these motion fields to create a co-located reference frame that accurately represents the motion.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional motion compensated prediction is used, then encoding complexity is reduced, but motion prediction accuracy deteriorates for non-translational motion

Engineering Contradiction:
Improvemotion prediction accuracyVSAvoidencoding complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The current frame is divided into multiple blocks, and a motion field is determined for each block independently using bi-directional reference frames. This segmentation allows each block to capture its own non-translational motion characteristics, improving overall motion prediction accuracy while keeping the computational complexity manageable through localized processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies an affine homographic model with multiple parameters (including rotation and translation components) to represent motion, transitioning from simple translational motion vectors to a more comprehensive parametric motion representation. This parameter expansion enables accurate modeling of non-translational motion while the parameters are optimized through a defined objective function.

Inventive Principle:
Principle #35Parameter changes

2Manufacturing precision

If affine homographic model with optimization is applied, then motion field accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improvemotion field accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary determination of motion fields for multiple blocks using bi-directional reference frames before constructing the final motion field-based reference frame. This preliminary action organizes the computational work in a structured sequence, allowing the optimization process to work with pre-computed motion data rather than raw pixel data, reducing overall computational complexity.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The motion field-based reference frame serves as an intermediary structure that captures non-translational motion characteristics. Instead of directly applying complex optimization to generate prediction blocks, the patent uses this intermediate motion field representation, which simplifies subsequent prediction operations while maintaining high motion field accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3714601B1Motion field-based reference frame rendering for motion compensated prediction in video coding
Publication Date: 2025.03.05 GOOGLE LLC
  • EP3714601B1 patent drawingFigure 1
  • EP3714601B1 patent drawingFigure 2
  • EP3714601B1 patent drawingFigure 3

AI summary

A motion-field based reference frame is rendered that can be used to perform a prediction process for a first frame of a video sequence to be predicted. A first reference frame from the video sequence for forward inter prediction of the first frame is determined, as is a second reference frame from the video sequence for backward inter prediction of the first frame. A respective motion field determined for blocks of the first frame is used to determine parameters forming an affine homographic model. A current block of a reconstructed reference frame is rendered at a co-located position within a motion field-based reference frame by applying the affine homographic model to the current block. An affine homographic model may be determined for each reconstructed reference frame block to render the motion-field based reference frame.