MMVD Inter-Frame Prediction Method for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in optimizing inter-frame prediction, leading to low accuracy and distorted video images, which affects user experience and data compression efficiency.
Innovation Solution
An inter-frame prediction method based on the Merge with Motion Vector Difference (MMVD) mode is introduced, which constructs a candidate list of motion vectors, determines search amplitudes and directions, and uses these to find optimal motion vectors, expanding the search range and improving prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional merge mode is used for inter-frame prediction, then coding complexity is reduced, but prediction accuracy deteriorates leading to distorted video images
Solution Approach 1:
The motion vector search space is segmented into multiple candidate lists (first candidate list, second candidate list, third candidate list) with different priorities and search ranges. The MMVD mode further segments the motion vector difference search into multiple directions and amplitudes, allowing systematic exploration of candidate vectors without overwhelming complexity.
Solution Approach 2:
Different candidate motion vectors are assigned different weights and priorities based on their likelihood of being optimal. The first candidate list contains high-priority vectors with smaller search ranges, while the second and third lists contain lower-priority vectors with larger search ranges, allowing focused computation on the most promising candidates.
2Manufacturing precision
If larger motion search range is used, then prediction accuracy is improved, but coding amount increases
Solution Approach 1:
Instead of exhaustively searching all possible motion vectors in a large range, the patent performs partial search by limiting the search to multiple predefined directions and amplitudes. The MMVD mode searches only in 8 specific directions with 7 predefined amplitudes, which covers the likely motion ranges while avoiding unnecessary computations and coding overhead.
Solution Approach 2:
The patent changes the search parameters from a uniform exhaustive search to a directional search with discrete amplitude levels. By transforming the search space into 8 directions × 7 amplitudes = 56 discrete points, the system achieves adequate coverage of motion possibilities while maintaining manageable coding complexity and bitrate.
3Measurement precision
If multiple candidate motion vectors are evaluated, then optimal motion vector accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-defining the candidate list structure, search directions, and amplitude levels before the actual motion vector search. The first candidate list is constructed with high-priority vectors that are most likely to be optimal, allowing the encoder to quickly evaluate promising candidates without exhaustive searching.
Solution Approach 2:
The patent maintains continuity of useful action by systematically evaluating candidate vectors in order of priority. The encoder continuously refines the motion vector estimate by comparing candidates from the first, second, and third lists, ensuring that computation is always focused on the most promising candidates without idle or redundant processing.
Data Source
AI summary
An inter-frame prediction method based on a merge with motion vector difference (MMVD) mode may include: constructing a first candidate list of a current coding block in the MMVD mode, wherein the first candidate list includes a first preset number of first candidate motion vectors (MVs); determining a second preset number of motion search amplitudes and a third preset number of motion search directions; taking the first candidate MV as a starting point, and searching by different motion combinations of the motion search amplitudes and the motion search directions, to obtain a MV offset of the first candidate MV under the different motion combinations, for each of the first candidate MVs; determining to obtain an optimal MV of the current coding block based on the MV offset of each of the first candidate MVs in the different motion combinations.