Angular Mode Derivation Using Log-Gradient Table Lookup
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing angular mode derivation process in decoder-side intra mode derivation involves excessive processing due to repeated comparisons and two-stage conversions of gradient magnitudes to angular modes, leading to increased computational load.
Innovation Solution
An angular mode derivation apparatus that utilizes a gradient derivation unit to derive first and second gradients, employing logarithm-based shifts and prescribed tables to determine angular modes, with an angular mode selection unit to select a representative value, thereby reducing processing overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If angular mode derivation is performed using repeated comparison of gradient magnitudes and two-stage conversion, then angular mode accuracy is improved, but computational load increases
Solution Approach 1:
The patent changes the parameter representation from gradient magnitude comparisons to logarithmic gradient ratios. By using log(gx) and log(gy) values directly from pre-computed tables, the system avoids repeated magnitude comparisons and reduces computational steps while maintaining angular mode accuracy through the preserved ratio information.
Solution Approach 2:
The patent performs preliminary computation of logarithmic gradient values and stores them in tables before angular mode derivation. The gradient magnitude ratios are pre-calculated and stored as log(gx) and log(gy), eliminating the need for real-time magnitude comparisons and reducing the computational load during actual angular mode derivation.
2Measurement precision
If two-stage conversion from gradient ratio to angular mode is used, then derivation accuracy is improved, but processing time increases
Solution Approach 1:
The patent merges the two-stage conversion process into a single-stage lookup operation. Instead of separately computing gradient ratios and then converting to angular modes in two distinct stages, the system combines both operations by directly retrieving pre-computed angular mode values from tables using logarithmic gradient ratios as indices, thereby reducing processing time while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary computation of the conversion relationship between gradient ratios and angular modes, storing the results in lookup tables. This pre-computation eliminates the need for time-consuming real-time conversions, allowing the system to quickly retrieve accurate angular mode values during video decoding.
3Measurement precision
If multiple gradient comparisons are performed for angular mode derivation, then mode selection accuracy is improved, but device complexity increases
Solution Approach 1:
The patent replaces the mechanical comparison process with a table lookup mechanism. Instead of performing multiple gradient magnitude comparisons using conditional logic and control flow, the system substitutes these operations with direct memory access to pre-computed tables, simplifying the processing logic and reducing device complexity while maintaining mode selection accuracy.
Solution Approach 2:
The patent introduces logarithmic gradient values as an intermediary representation between raw gradients and angular modes. This intermediary form allows direct indexing into lookup tables, eliminating the need for complex comparison logic and reducing processing complexity while preserving the information needed for accurate mode selection.
Data Source
AI summary
In a conventional angular mode derivation, comparison of a magnitude relationship is required multiple times to convert gradients of pixel values into an angular mode, and this hinders parallel processing. A gradient derivation unit configured to derive a first gradient being a gradient of a pixel value included in a gradient derivation target image, and an angular mode derivation unit configured to derive an angular mode, using the first gradient, a second gradient different from the first gradient, and two prescribed tables are included.