Motion Vector Refinement for Video Inter Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The increasing demand for high-resolution and high-quality video signals, such as HD and UHD, leads to larger amounts of data, resulting in higher costs for transmission and storage. Existing image compression technologies struggle to efficiently encode and decode video signals, particularly in handling stereographic image content.

Innovation Solution

A method and apparatus for efficiently performing inter prediction in video signal encoding/decoding by obtaining an initial motion vector for a current block, deriving refined motion vectors for search points based on the initial motion vector, and selecting a motion vector from these refined options. This approach includes selecting a refinement mode, such as bi-lateral matching or template matching, and determining a search pattern for the current block.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional image compression technologies are used, then transmission and storage costs are reduced, but inter prediction efficiency for high-resolution video signals deteriorates

Engineering Contradiction:
Improveinter prediction efficiencyVSAvoidtransmission and storage costs
Core Design Contradiction:
ProductivityVSLoss of energy

Solution Approach 1:

The patent applies preliminary action by deriving multiple refined motion vectors for different search points before final motion vector selection. The initial motion vector is refined through multiple search points (e.g., first search point, second search point) with different offset values, allowing the system to pre-compute multiple candidate motion vectors and select the optimal one, thereby improving prediction efficiency for high-resolution video

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements dynamics by making the motion vector refinement process adaptive. The system dynamically selects search points based on motion characteristics and prediction modes. Different search points are used for different motion scenarios (e.g., small offset values for certain modes, larger offset values for others), allowing the system to adapt to varying video content and optimize compression efficiency

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If motion information is updated for current blocks, then inter prediction accuracy is improved, but processing complexity increases

Engineering Contradiction:
Improvemotion information accuracyVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by updating motion information selectively for specific blocks rather than uniformly for all blocks. The system determines whether to update motion information based on local characteristics such as prediction mode (e.g., bidirectional prediction mode) and block position. This selective update approach improves motion accuracy where needed while reducing processing complexity in regions where motion information is already sufficient

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent implements segmentation by dividing the motion vector search process into multiple discrete search points around the initial motion vector. Instead of performing a comprehensive search, the system segments the search space into specific points (first search point with first offset value, second search point with second offset value) and evaluates motion information at each segment. This reduces processing complexity while maintaining accuracy

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS20250030886A1Method for refining a motion vector derived under a merge mode using a difference vector
Publication Date: 2025.01.23 KT CORP
  • US20250030886A1 patent drawing
  • US20250030886A1 patent drawing
  • US20250030886A1 patent drawing

AI summary

An image decoding method according to the present invention comprises: a step of acquiring an initial motion vector of a current block; a step of deriving a refined motion vector for each of a plurality of search points on the basis of the initial motion vector; and a step of acquiring a motion vector of the current block on the basis of the refined motion vector of any one of the plurality of search points.