Picture Prediction Using Spatial Adjacent Motion Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional video coding technologies face high computational complexity when using non-translational motion models for picture prediction, as they require selecting motion vector predictors from multiple optional sets, which increases calculation and bitstream transfer requirements.

Innovation Solution

A method and apparatus for picture prediction that determines motion vector predictors for a current picture block based on a non-translational motion model, where predictors are directly obtained from spatially adjacent blocks, reducing the need for complex selection processes and improving coding efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If non-translational motion model is used for picture prediction, then motion representation accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improvemotion representation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The current picture block is divided into multiple sub-blocks, and motion vector predictors are determined for each sub-block independently based on its spatially adjacent blocks. This segmentation allows the complex non-translational motion model to be applied in a distributed manner, improving motion representation accuracy while managing computational complexity through localized processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Motion vector predictors are determined in advance for multiple sub-blocks before performing the actual pixel value prediction. By pre-calculating the motion vector predictors using spatially adjacent blocks, the system prepares motion compensation data beforehand, which improves the accuracy of the non-translational motion model while organizing computations in a systematic sequence that manages complexity.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If motion vector predictors are selected from multiple optional sets, then prediction accuracy is improved, but bitstream transfer requirements increase

Engineering Contradiction:
Improveprediction accuracyVSAvoidbitstream transfer requirements
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent extracts and uses motion vector information directly from spatially adjacent picture blocks without requiring complex selection from multiple optional sets. By taking out the motion vector predictors from the adjacent blocks and applying them directly to the current block's sub-blocks, the system achieves prediction accuracy while avoiding the need to encode and transmit multiple candidate sets, thus reducing bitstream requirements.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The motion vector predictors from spatially adjacent blocks serve multiple purposes: they are used for multiple sub-blocks within the current picture block and can be derived from different adjacent blocks (above, below, left, right). This multi-functional use of motion vector predictors achieves accurate prediction without requiring separate predictor sets for each sub-block, reducing the information that needs to be transmitted.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3836546A1Picture prediction method and related apparatus
Publication Date: 2021.06.16 HUAWEI TECH CO LTD
  • EP3836546A1 patent drawingFigure 1-a~1-c
  • EP3836546A1 patent drawingFigure 2~3-a
  • EP3836546A1 patent drawingFigure 3-b~3-c

AI summary

Embodiments of the present invention disclose a picture prediction method and a related apparatus. The picture prediction method includes: determining motion vector predictors of K pixel samples in a current picture block, where K is an integer greater than 1, the K pixel samples include a first vertex angle pixel sample in the current picture block, a motion vector predictor of the first vertex angle pixel sample is obtained based on a motion vector of a preset first spatially adjacent picture block of the current picture block, and the first spatially adjacent picture block is spatially adjacent to the first vertex angle pixel sample; and performing, based on a non-translational motion model and the motion vector predictors of the K pixel samples, pixel value prediction on the current picture block. Solutions in the embodiments of the present invention are helpful in reducing calculation complexity of picture prediction based on a non-translational motion model.