Multi-Hypothesis Inter Prediction for Lower-Overhead Image Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image compression technologies face challenges in efficiently compressing high-resolution and high-quality images, particularly in accurately performing inter prediction and configuring motion information candidates for multi-reference block modes.
Innovation Solution
An image decoding method and device that generates a basic prediction block through bi-prediction, derives additional reference blocks, and combines them using weighted summation to improve prediction accuracy and reduce signaling overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-reference block mode is used to improve prediction accuracy, then prediction accuracy is improved, but device complexity increases due to additional reference block derivation and configuration
Solution Approach 1:
The patent applies preliminary action by pre-configuring motion information candidate lists with multiple reference blocks before the actual prediction process. The encoder prepares multiple candidate motion vectors and reference block combinations in advance, allowing the decoder to select from pre-computed options without performing complex derivations during decoding, thus improving prediction accuracy while managing device complexity
Solution Approach 2:
The patent utilizes parameter changes by dynamically adjusting the number and configuration of reference blocks based on prediction mode indicators. Different inter prediction modes (merge mode, AMVP mode, AMVP-merge combined mode) trigger different numbers of reference blocks to be used, allowing the system to adapt complexity to actual prediction needs rather than always using maximum complexity
2Measurement precision
If additional reference blocks are derived for multi-hypothesis prediction, then prediction accuracy is improved, but signaling overhead increases
Solution Approach 1:
The patent applies local quality by providing different levels of reference block information to different blocks based on their prediction mode. Instead of uniformly providing full reference block information to all blocks, the system selectively provides additional reference blocks only where needed (in merge mode, AMVP mode, or AMVP-merge combined mode), reducing overall signaling overhead while maintaining prediction accuracy where it matters most
Solution Approach 2:
The patent uses copying by reusing motion information from neighboring blocks and previously decoded blocks. The motion information candidate lists are constructed by copying motion vectors and reference block indices from spatial and temporal neighbors, reducing the need to signal completely new motion information for each block while maintaining prediction accuracy
Data Source
AI summary
An image encoding/decoding method and device, according to the present disclosure, may: generate a basic prediction block of a current block by performing bi-prediction on the basis of a first inter prediction mode; derive an additional reference block of the current block on the basis of a second inter prediction mode; and generate a final prediction block of the current block by weighted summing the basic prediction block and the additional reference block.