DIMD-Guided Transform Sets for Non-Directional Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding techniques face challenges in efficiently determining transform sets for non-directional intra prediction modes, leading to increased complexity in video encoding and decoding processes.
Innovation Solution
Implement techniques for deriving additional transform sets, such as matrix-based intra prediction (MIP), extrapolation filter-based intra prediction (EIP), and intra template matching prediction (IntraTMP) modes, using a decoder-side intra mode derivation (DIMD) process to reduce complexity by selecting transform sets based on intra prediction directions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple transform sets are used for non-directional intra prediction modes, then prediction accuracy is improved, but complexity of determining transforms increases
Solution Approach 1:
The patent pre-establishes multiple transform sets (first transform set and second transform set) with different transform kernels before the actual prediction process. This allows the system to avoid complex real-time determination of which transform to use, instead simply selecting from pre-configured options based on the prediction mode and block characteristics.
Solution Approach 2:
The patent dynamically selects between different transform sets based on the intra prediction mode and block characteristics. The system adapts its behavior by choosing the appropriate transform set (first or second) depending on whether the prediction mode is directional or non-directional, and which specific transform kernel performs best for each case, rather than using a fixed transform for all situations.
2Measurement precision
If decoder-side intra mode derivation process is applied, then transform selection accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs the intra mode derivation process on the prediction block samples before the actual transform selection. By preparing the prediction block and deriving the intra mode in advance, the system can efficiently select the appropriate transform set without repeating complex calculations during the transform determination step, thus reducing overall processing time while maintaining accuracy.
Solution Approach 2:
The patent uses the prediction block samples as a basis to derive intra modes that copy the characteristics of the actual block. This allows the system to infer appropriate transform parameters from the prediction block, which is already processed and available, rather than performing complex analysis on the original block during transform selection.
Data Source
AI summary
A video coder is configured to determine a first set of transform sets and a second set of transform sets, determine, for a current block of video data coded using a non-directional intra mode, a list of intra prediction directions determined from a decoder-side intra mode derivation (DIMD) process applied to prediction block samples of the current block, determine, for the current block, a transform set from the second set of transform sets based on a second highest intra prediction direction in the list of intra prediction directions, and encode or decode the current block using a transform in the transform set.