History-Based Affine Merge Motion Vector Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies face challenges in efficiently compressing video data due to high bandwidth and storage requirements, particularly in reducing redundancy and managing motion vectors effectively.
Innovation Solution
The implementation of a history-based affine motion model for inter-picture prediction in video coding, which involves storing affine motion information in a buffer and constructing a motion candidate list using this information to predict motion vectors, thereby reducing data requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If motion vectors are directly coded without prediction, then coding simplicity is maintained, but bitrate efficiency deteriorates due to high data requirements
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing affine motion information candidates in an HMVP buffer during the encoding of previous blocks. This historical motion data is prepared in advance and made available for prediction of current block motion vectors, reducing the need to transmit full motion vector data and thereby improving bitrate efficiency without significantly increasing coding complexity
Solution Approach 2:
The patent implements feedback by using the HMVP buffer to store and reuse motion information from previously encoded blocks. The encoder feeds back historical motion data that can be referenced for predicting motion vectors of subsequent blocks, creating a feedback loop that improves compression efficiency by exploiting temporal and spatial correlations in motion patterns
2Measurement precision
If affine motion model is used for inter-picture prediction, then prediction accuracy is improved, but data requirements increase due to additional motion information
Solution Approach 1:
The patent applies parameter changes by dynamically selecting affine motion information candidates from the HMVP buffer based on matching criteria. Instead of always using full affine models with multiple parameters, the system adapts by selecting candidates with varying levels of parameter detail, reducing the quantity of motion data that needs to be coded while maintaining prediction accuracy when appropriate
Solution Approach 2:
The patent uses partial action by selectively applying affine motion prediction only when beneficial. The system evaluates available HMVP candidates and applies affine modeling only when it provides sufficient improvement, otherwise falling back to simpler prediction methods, thus avoiding the excessive data requirements of full affine coding while maintaining accuracy where needed
3Productivity
If historical motion data is stored in HMVP buffer, then compression efficiency is improved, but memory requirements increase
Solution Approach 1:
The patent applies discarding and recovering by implementing a buffer with limited capacity that discards older motion information candidates when full. The HMVP buffer maintains only the most relevant recent motion data, discarding less useful historical information, thereby balancing compression efficiency improvements with constrained memory requirements through selective retention and recovery of motion candidates
Data Source
AI summary
A method for video coding includes encoding or decoding a coding block in a current picture with an affine motion model based inter-picture prediction method in a video coding system, storing affine motion information of the coding block in a history-based motion vector prediction (HMVP) buffer that is configured for storing affine motion information candidates each including affine motion information of a processed affine-coded coding block, and constructing a motion candidate list for a current block that includes at least one candidate selected from the affine motion information candidates stored in the HMVP buffer or derived from one of the affine motion information candidates stored in the HMVP buffer.