Offset Vector Identification for Video Coding Efficiency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video coding techniques face inefficiencies in motion vector prediction, particularly due to the signaling overhead required to cover all blocks in reference frames, which can reduce coding efficiency and limit the area of reference pictures that can be effectively utilized.

Innovation Solution

A method is introduced where a video coder signals offset vectors from a limited set of non-uniformly distributed values, allowing for the identification of blocks in larger areas of reference pictures, thereby improving coding efficiency by reducing residual data needed for current blocks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If offset vectors are signaled from a limited set of uniformly distributed values, then signaling overhead is reduced, but the area of reference pictures that can be effectively utilized is limited

Engineering Contradiction:
Improvesignaling overheadVSAvoidarea of reference pictures
Core Design Contradiction:
Loss of informationVSArea of stationary object

Solution Approach 1:

The patent changes the distribution pattern of offset vector values from uniform to non-uniform. By concentrating more offset values in certain directions (e.g., horizontal and vertical directions where motion is more common), the system achieves better motion prediction accuracy without increasing the total number of offset values, thus maintaining low signaling overhead while improving reference picture utilization

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces asymmetric distribution of offset vector values, where certain directions (horizontal and vertical) have more available offset values compared to diagonal directions. This asymmetric design matches the typical motion patterns in video sequences, allowing more effective utilization of reference pictures in directions where motion is more likely to occur, while keeping the overall offset set limited

Inventive Principle:
Principle #4Asymmetry

2Measurement precision

If offset vectors cover all blocks in reference frames, then prediction accuracy is improved, but signaling overhead increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidsignaling overhead
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent changes the parameter of offset vector distribution from uniform to non-uniform, concentrating precision where it matters most. By having more offset values in horizontal and vertical directions (where motion is more common), the system achieves better prediction accuracy for typical motion patterns without needing to provide offset values for all possible block positions, thus reducing signaling overhead

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies local quality by providing higher offset vector resolution in specific directions (horizontal and vertical) where motion is more likely to occur, while using fewer offset values in diagonal directions. This localized precision improvement achieves better overall prediction accuracy without the cost of uniformly high precision across all directions

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3513560B1Offset vector identification of temporal motion vector predictor
Publication Date: 2020.10.21 QUALCOMM INC
  • EP3513560B1 patent drawingFigure 1
  • EP3513560B1 patent drawingFigure 2
  • EP3513560B1 patent drawingFigure 3

AI summary

In one example, a method includes receiving, in a coded video bitstream and for a current prediction unit (PU) of a current picture of video data, a representation of an offset vector that identifies a block of a reference picture, wherein the offset vector is within a set of values that are not uniformly distributed; determining, based on the block of the reference picture identified by the offset vector, motion information for the current PU; and reconstructing the current PU based on the determined motion information.