Subblock-Based Template Matching for Accurate Motion Vector Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face inefficiencies in intra prediction and motion vector prediction, particularly in handling the increasing number of prediction directions and redundancy in video data, which affects compression ratios and decoding efficiency.
Innovation Solution
A subblock-based template matching motion vector prediction (SbTMVP) mode is employed, where each subblock in a current block is predicted using a collocated reference subblock determined by a displacement vector and motion vector offset, with template matching techniques to find the best reference blocks for reconstruction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional motion vector prediction is used, then decoding complexity is low, but compression ratio is insufficient
Solution Approach 1:
The current block is divided into multiple subblocks, and each subblock is independently predicted using template matching. This segmentation allows the system to achieve better compression ratios by capturing local motion variations while maintaining manageable decoding complexity through standardized processing for each subblock.
Solution Approach 2:
Templates are constructed from neighboring blocks before the actual motion vector prediction process. This preliminary action of creating reference templates from available data enables more accurate matching and improves compression ratios without significantly increasing the complexity during the main decoding phase.
2Measurement precision
If template matching is applied to each subblock, then motion vector prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
Template matching is applied to each subblock rather than the entire block, providing partial action that improves motion vector prediction accuracy for regions with different motion characteristics. This selective approach balances the computational complexity by avoiding full-block processing while still capturing essential motion variations.
Solution Approach 2:
Different templates are constructed and matched for each subblock based on local characteristics, allowing the system to adapt to local motion patterns. This local quality approach improves prediction accuracy for each region while maintaining overall computational efficiency through localized processing.
3Adaptability or versatility
If more prediction directions are supported, then adaptability to various video content is improved, but redundancy in prediction data increases
Solution Approach 1:
The system changes the parameter of prediction direction by deriving multiple candidate directions from template matching results rather than using fixed predefined directions. This dynamic parameter adjustment improves adaptability to various video content while reducing redundancy by only encoding the necessary direction information.
Solution Approach 2:
The template matching process provides feedback on the actual motion characteristics of each subblock, allowing the system to adaptively select the most appropriate prediction direction. This feedback mechanism improves versatility by responding to actual content characteristics while reducing redundancy by avoiding encoding of unnecessary directional information.
Data Source
AI summary
A coded video bitstream comprising a current block in a current picture is received. The current block includes a plurality of subblocks and is to be predicted by a subblock-based template matching motion vector prediction (SbTMVP) mode. A respective collocated reference subblock for each subblock is determined based on a combination of a displacement vector (DV) and a motion vector offset (MVO) that are associated with the respective subblock. A motion vector (MV) field in the respective collocated reference subblock of each subblock in the current block is determined. A respective reference template for each subblock is derived based on the determined MV field of the collocated reference subblock. The plurality of subblocks of the current block is reconstructed by predicting each subblock using the respective reference template in the SbTMVP mode.