Chroma Intra Prediction Using Reduced Line Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional chroma intra prediction methods require large line buffers for deriving parameters, leading to increased hardware costs and memory requirements, especially for on-chip implementations, due to the need for storing reconstructed luma samples from multiple lines.
Innovation Solution
The method reduces line buffer requirements by deriving chroma intra predictor parameters using only one reconstructed luma pixel line above the horizontal block boundary, employing horizontal decimation to match chroma resolution, and optionally using low-pass filtering for downsampling, thereby eliminating the need for vertical subsampling or downsampling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional chroma intra prediction methods use reconstructed luma samples from multiple lines to derive parameters, then prediction accuracy is improved, but line buffer memory requirements increase significantly
Solution Approach 1:
The patent segments the luma block into multiple regions (first region with sufficient samples, second region with insufficient samples) and applies different processing strategies to each region. For the first region, conventional multi-line sampling is used; for the second region, only the most recent luma line is used, thereby reducing overall buffer requirements while maintaining prediction accuracy.
Solution Approach 2:
The patent applies partial action by using only a subset of available luma lines (specifically, only the most recent luma line) for deriving chroma prediction parameters in certain regions, rather than using all available lines. This partial approach is sufficient for maintaining prediction accuracy while significantly reducing memory buffer requirements.
2Device complexity
If the line buffer size is reduced to minimize memory requirements, then hardware cost is reduced, but the ability to derive accurate prediction parameters may be compromised
Solution Approach 1:
The patent applies local quality by treating different regions of the luma block differently based on their local characteristics. Regions with sufficient causal luma samples use conventional processing, while regions with insufficient samples use a simplified approach using only the most recent line. This localized differentiation maintains overall prediction accuracy while reducing hardware complexity.
Solution Approach 2:
The patent uses a simplified copying approach where instead of copying and processing multiple luma lines, it copies only the most recent luma line for use in deriving prediction parameters. This reduces the amount of data that needs to be stored and processed, thereby reducing hardware complexity while maintaining sufficient prediction accuracy.
3Reliability
If multiple luma lines are stored for parameter derivation, then chroma prediction robustness is improved, but on-chip memory capacity requirements increase
Solution Approach 1:
The patent introduces dynamics by making the buffer usage adaptive rather than static. Instead of always allocating space for multiple luma lines, the system dynamically adjusts its behavior based on the available causal luma samples in different regions. When sufficient samples are available, multiple lines are used; when not available, only the most recent line is used, optimizing memory usage while maintaining robustness.
Solution Approach 2:
The patent applies discarding by selectively discarding older luma lines that are no longer needed for parameter derivation in certain regions. By discarding unnecessary historical data and retaining only the most recent luma line where needed, the system reduces on-chip memory requirements while maintaining prediction robustness through the retained relevant information.
Data Source
AI summary
A method and apparatus for chroma intra prediction for a current chroma block with reduced line memory requirement are disclosed. The chroma intra predictor is derived from reconstructed luma pixels of a current luma block using a model with parameters. In various embodiments according to the present invention, the derivation of the parameters relies on a reconstructed luma pixel set corresponding to neighboring reconstructed luma pixels from causal luma neighboring areas of the current luma block, wherein said causal luma neighboring areas include a first area corresponding to reconstructed luma pixels above a horizontal luma block boundary on a top side of the current luma block, and wherein the reconstructed luma pixels from the first area that are included in the reconstructed luma pixel set are from a luma pixel line immediately above the horizontal luma block boundary.