Image Decoding Block Sub-division for Prediction Candidate Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image encoding and decoding techniques using inter prediction do not achieve optimal coding efficiency.
Innovation Solution
An image decoding method that divides a current block into sub-blocks, derives prediction information candidates for each sub-block, and selects a candidate from these for decoding, while excluding neighboring blocks within the same CU from the merging block candidates to avoid redundancy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If all neighboring blocks are included as merging block candidates, then more prediction candidates are available, but redundant candidates increase and coding efficiency decreases
Solution Approach 1:
The patent extracts and removes redundant neighboring blocks from the merging block candidate list. Specifically, it identifies blocks that are already included in the current block as redundant and excludes them from the candidate list, thereby reducing the number of candidates without losing useful prediction information.
Solution Approach 2:
The patent applies different treatment to different neighboring blocks based on their relationship with the current block. Blocks outside the current block are included as candidates, while blocks inside the current block are excluded, creating a differentiated candidate selection strategy that optimizes coding efficiency.
2Measurement precision
If more merging block candidates are considered, then prediction accuracy improves, but processing complexity increases
Solution Approach 1:
The patent extracts only the necessary neighboring blocks for merging candidate consideration. By removing blocks that are already part of the current block, it reduces the set of candidates that need to be processed and evaluated, thereby reducing processing complexity while maintaining prediction accuracy for the relevant blocks.
Solution Approach 2:
The patent segments the set of neighboring blocks into two categories: those outside the current block (included as candidates) and those inside the current block (excluded). This segmentation simplifies the processing by clearly defining which blocks require detailed evaluation and which can be immediately discarded.
3Measurement precision
If reference picture information is lost, then decoding accuracy deteriorates, but error propagation affects subsequent blocks
Solution Approach 1:
The patent performs preliminary determination of merging block candidates using only locally available information from neighboring blocks, without requiring reference picture data. This preliminary action allows the decoding to proceed even when reference pictures are lost, maintaining error resistance while preserving decoding accuracy for the blocks that can be decoded with available information.
Solution Approach 2:
The patent enables each block to determine its merging candidates independently using only the information available in the current decoding context (neighboring blocks within the same picture). This self-service approach eliminates dependency on external reference pictures, making the decoding process resilient to reference picture loss and preventing error propagation.
Data Source
AI summary
An image decoding method includes: dividing a current block into sub-blocks; deriving, for each sub-block, one or more prediction information candidates; obtaining an index; and decoding the current block using the prediction information candidate selected by the index. The deriving includes: determining whether a neighboring block neighboring each sub-block is included in the current block, and when not included in the current block, determining the neighboring block to be a reference block available to the sub-block, and when included in the current block, determining the neighboring block not to be the reference block; and deriving a prediction information candidate of the sub-block from prediction information of the reference block; and when the number of prediction information candidates is smaller than a predetermined number, generating one or more new candidates without using the prediction information of the reference block till the number of prediction information candidates reaches the predetermined number.