Inter Prediction Image Coding Using Symmetric Motion Vector Differences

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The increasing demand for high-resolution, high-quality images and immersive media such as VR and AR content has led to a need for more efficient image/video compression technologies to reduce transmission and storage costs, particularly in handling motion vector differences and inter prediction processes.

Innovation Solution

The implementation of inter prediction methods that include signaling information on motion vector differences, specifically L0 and L1 motion vector differences, using SMVD flags and deriving L1 motion vector differences based on SMVD flags, and utilizing short-term reference pictures for improved efficiency in video/image coding systems.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If inter prediction with motion vector difference signaling is used to improve compression efficiency, then compression ratio increases, but computational complexity increases

Engineering Contradiction:

Improvecompression efficiencyVSAvoidcomputational complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The motion vector difference signaling is segmented into two components: L0 MVD and L1 MVD. The L1 MVD is derived from the L0 MVD through a symmetric relationship, which divides the computational workload and reduces the overall complexity while maintaining compression efficiency.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of independently signaling both L0 MVD and L1 MVD, the patent inverts the traditional approach by deriving L1 MVD from L0 MVD using a symmetric motion vector difference relationship. This reduces the signaling overhead and computational complexity.

Inventive Principle:

Principle #13The other way round (Inversion)

2Measurement precision

If symmetric motion vector difference with L0 and L1 motion vector differences is used to improve prediction accuracy, then prediction quality improves, but signaling overhead increases

Engineering Contradiction:

Improveprediction accuracyVSAvoidsignaling overhead

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The patent inverts the traditional symmetric motion vector difference approach by deriving L1 MVD from L0 MVD rather than independently signaling both. This maintains prediction accuracy through the symmetric relationship while significantly reducing signaling overhead.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The patent extracts only the essential L0 motion vector difference information and derives the L1 motion vector difference from it using the symmetric relationship. This extraction approach maintains prediction quality while minimizing the information that needs to be signaled.

Inventive Principle:

Principle #2Taking out (Extraction)

3Productivity

If bi-prediction is applied to current block to improve compression efficiency, then compression ratio increases, but processing complexity increases

Engineering Contradiction:

Improvecompression efficiencyVSAvoidprocessing complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The bi-prediction process is segmented into two independent prediction directions (L0 and L1), where each direction can be processed separately. The L1 prediction utilizes the derived L1 MVD from the L0 MVD, which simplifies the overall processing while maintaining compression efficiency.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250211778A1Inter prediction-based image coding method and apparatus

Publication Date: 2025.06.26 LG ELECTRONICS INC

AI summary

An image decoding method according to the present document may comprise the steps of: deriving an inter prediction mode from encoded information; configuring reference picture lists, deriving motion information comprising reference picture indexes for symmetric motion vector differences (SMVD); and generating prediction samples on the basis of the motion information, wherein the reference picture indexes for SMVD can be derived on the basis of short-term reference pictures included within the reference picture lists.