Shape-Adaptive Intra Prediction Mode Selection in QTBT Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding systems, such as those using QTBT division, do not fully consider the shape differences of Coding Units (CUs), leading to suboptimal coding efficiency due to inadequate consideration of block shapes and their prediction directions.
Innovation Solution
An image decoding and coding apparatus that performs intra prediction by deriving and decoding a candidate list of intra prediction modes based on the shape of the block, adjusting the number of prediction modes and their associations to better match the block's aspect ratio and division history, thereby improving coding and decoding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If QTBT division is used to generate CUs of various shapes, then the flexibility and adaptability of block division is improved, but the coding efficiency deteriorates due to inadequate consideration of shape differences in prediction mode selection
Solution Approach 1:
The patent applies local quality by deriving different candidate prediction mode lists based on the specific shape characteristics of each CU. For vertically long blocks, vertical prediction modes are prioritized, while for horizontally long blocks, horizontal prediction modes are prioritized. This shape-adaptive approach ensures that the prediction mode selection is locally optimized for each block's geometry, resolving the contradiction between shape flexibility and coding efficiency.
Solution Approach 2:
The patent implements dynamics by making the candidate prediction mode list dynamic and adaptable to different CU shapes. The system dynamically adjusts which prediction modes are included in the candidate list based on the block's aspect ratio and orientation. This dynamic adaptation allows the system to efficiently handle various block shapes generated by QTBT division without using a fixed, one-size-fits-all prediction mode approach.
2Measurement precision
If the candidate list includes all intra prediction modes, then the prediction accuracy is improved, but the decoding complexity increases
Solution Approach 1:
The patent applies the taking out principle by extracting only the relevant prediction modes that are suitable for the specific CU shape into the candidate list. Instead of including all intra prediction modes, the system selectively extracts and includes only those modes that are geometrically appropriate for the current block shape. This extraction approach maintains prediction accuracy for the given shape while significantly reducing decoding complexity by limiting the search space.
Solution Approach 2:
The patent uses partial action by including a subset of prediction modes in the candidate list rather than all possible modes. The candidate list contains only the necessary prediction modes relevant to the current block shape, which is fewer than the complete set of intra prediction modes. This partial inclusion achieves sufficient prediction accuracy for each specific shape while reducing the overall computational burden.
Data Source
Figure 1
Figure 2A~2F
Figure 3A~3H
AI summary
An image decoding apparatus for using a binary tree division in addition to a quad tree division when dividing a picture includes an intra prediction parameter decoding control unit and a prediction image generation unit. The intra prediction parameter decoding control unit changes a first association between a plurality of prediction directions and a plurality of mode indices in an intra prediction mode into a second association based on a shape of a block. The prediction image generation unit generates a prediction image by the mode indices and the second association. The intra prediction parameter decoding control unit changes a number of first portions of the prediction directions based on the shape to generate a plurality of changed prediction directions for determining the second association based on the changed prediction directions. The prediction directions directed toward a first direction region are the first portions of the prediction directions.