Motion Vector Predictor Derivation for Image Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional image coding methods face challenges in achieving high coding efficiency due to difficulties in accurately deriving spatial motion vector predictors, particularly when reference pictures are long-term or short-term, leading to reduced prediction accuracy and efficiency.

Innovation Solution

An image coding method that determines whether reference pictures are long-term or short-term and adjusts the derivation scheme for motion vector predictors accordingly, using scaling based on temporal distance only when both are short-term, thereby avoiding extreme values and improving prediction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If motion vector scaling based on temporal distance is applied to all reference picture combinations, then prediction accuracy may be maintained across different scenarios, but computational complexity increases and coding efficiency deteriorates

Engineering Contradiction:
Improvemotion vector prediction accuracyVSAvoidcoding complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies different motion vector derivation schemes based on the specific combination of reference picture types. When both current and neighboring reference pictures are long-term, no scaling is applied. When both are short-term, scaling based on temporal distance is applied. This localized adaptation of the derivation approach optimizes prediction accuracy for each scenario while avoiding unnecessary computational complexity in other scenarios.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the derivation parameter (scaling factor based on temporal distance) conditionally based on the reference picture types. The scaling parameter is applied only when both reference pictures are short-term, and omitted when both are long-term or when they differ in type, thus adapting the parameter to local conditions to balance accuracy and complexity.

Inventive Principle:
Principle #35Parameter changes

2Ease of manufacture

If uniform motion vector derivation scheme is used for all reference picture types, then implementation is simpler, but prediction accuracy decreases for specific long-term reference picture scenarios

Engineering Contradiction:
Improveimplementation simplicityVSAvoidmotion vector prediction accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent introduces dynamic selection of derivation schemes based on reference picture type combinations. The system dynamically adjusts whether to apply temporal distance scaling by checking the types of current and neighboring reference pictures, allowing the derivation process to adapt to different scenarios rather than using a fixed uniform approach.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

Different derivation quality levels are applied locally based on reference picture types. For long-term reference picture combinations, a simpler derivation without scaling is used. For short-term reference picture combinations, a more accurate derivation with temporal distance scaling is applied. This localized quality adjustment balances implementation complexity with prediction accuracy.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP2773112B1Image encoding method, image decoding method, image encoding apparatus, and image decoding apparatus
Publication Date: 2018.08.08 SUN PATENT TRUST
  • EP2773112B1 patent drawingFigure 1
  • EP2773112B1 patent drawingFigure 2
  • EP2773112B1 patent drawingFigure 3

AI summary

An image coding method includes: deriving a candidate for a motion vector predictor from a neighboring motion vector (S1201); adding the candidate to a list (S1202); selecting a motion vector predictor from the list; coding a current block (S1203); and coding a current motion vector (S1204). In the deriving (S1201), the candidate is derived according to a first derivation scheme when each of a current reference picture and a neighboring reference picture is determined to be a long-term reference picture, and the candidate is derived according to a second derivation scheme when each of a current reference picture and a neighboring reference picture is determined to be a short-term reference picture.