Non-Square Intra Prediction Blocks for Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as HEVC, limit coding efficiency by restricting Intra prediction to square blocks, which does not fully exploit spatial correlations in non-square block structures, leading to suboptimal compression performance.
Innovation Solution
The method introduces non-square prediction modes and transforms for Intra prediction, allowing blocks to be partitioned into 2NxN and Nx2N shapes, and uses a combination of square and non-square transforms, along with truncated binary coding for improved mode signaling, to enhance coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If square blocks are used for Intra prediction, then the coding structure is simple and implementation is easier, but coding efficiency is limited and spatial correlations in non-square blocks are not fully exploited
Solution Approach 1:
The patent divides the coding unit into multiple prediction units with different shapes (square and non-square). This segmentation allows the system to adapt to different spatial correlation patterns in the video content, improving coding efficiency without requiring a complete redesign of the coding structure. The prediction units can be independently processed with appropriate transform sizes.
Solution Approach 2:
The patent introduces dynamic selection of prediction unit sizes and shapes based on the characteristics of the video content. The encoder can choose between square and non-square prediction units adaptively, allowing the system to optimize for different spatial correlation patterns dynamically rather than being constrained to a fixed square structure.
2Productivity
If non-square prediction modes are introduced, then spatial correlations in non-square blocks are better exploited and coding efficiency improves, but device complexity increases
Solution Approach 1:
The patent applies different prediction and transform modes locally to different prediction units based on their shape and content characteristics. Square prediction units use square transforms, while non-square prediction units use corresponding non-square transforms. This local adaptation improves coding efficiency without requiring the entire system to handle all possible modes simultaneously.
Solution Approach 2:
The patent changes the parameters of the transform operation based on the prediction unit shape. Instead of using a fixed transform size, the transform dimensions are adapted to match the prediction unit dimensions (e.g., 2NxN or Nx2N for non-square units). This parameter adaptation enables efficient processing of non-square blocks while maintaining a systematic approach.
3Productivity
If non-square transforms are used, then compression performance is enhanced, but implementation complexity and processing overhead increase
Solution Approach 1:
The patent introduces asymmetric transform sizes that match the prediction unit shapes. Instead of always using square transforms, the system uses transforms with different dimensions (e.g., 2N x N or N x 2N) for non-square prediction units. This asymmetric approach better matches the spatial correlations in the data while maintaining a regular and implementable structure.
Solution Approach 2:
The patent designs the transform framework to be universal, handling both square and non-square cases through a unified approach. The same transform algorithms can be applied with different dimension parameters, allowing the system to process various prediction unit types without requiring entirely separate processing paths for each case.
Data Source
Figure 1~2A
Figure 2B~3A
Figure 3B~4A
AI summary
A method and apparatus for Intra prediction of a block based on neighboring pixels around the block are disclosed. Embodiments according to the present invention use square blocks as well as non-square blocks for Intra prediction. For a 2Nx2N Luma CU (coding unit), the CU can be partitioned into 2NxN, Nx2N, 2Nx2N or NxN PUs. The 2NxN and Nx2N PUs can be further processed by either square transforms only or both non-square and square transforms. In one embodiment, the 2NxN PU or the Nx2N PU is processed as two NxN TUs (transform units) and each of the NxN TU is further split into smaller NxN TUs based on quad-tree split. In another embodiment, the 2NxN and Nx2N PUs are processed as two 2Nx0.5N and 0.5Nx2N TUs respectively.