Pixel Color Component Processing with Luma Matching for HDR Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Highly non-linear transfer functions in HDR video processing combined with chroma subsampling and non-constant luminance ordering result in severe artifacts, particularly in saturated colors, due to sharp shifts in luma components without corresponding changes in chroma components.
Innovation Solution
A method that adjusts the luma value (Y') by linearizing the transfer function to match the original luminance (Yo) while avoiding color channel saturation, using different functions or look-up tables based on whether color components are within allowed ranges, ensuring accurate luminance values without artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a highly non-linear transfer function is used in HDR video processing, then the distribution of codewords is improved (more codewords in dark regions, fewer in bright regions), but severe artifacts appear in saturated colors due to sharp shifts in luma components
Solution Approach 1:
The patent applies preliminary action by adjusting the luma component Y' before chroma subsampling and color space conversion. By pre-adjusting Y' based on the original luminance Yo and the relationship between luma and chroma components, the method prevents sharp luma shifts that would otherwise cause artifacts in saturated colors after processing.
Solution Approach 2:
The patent implements feedback by using the original luminance Yo (calculated from original linear RGB values) as a reference to guide the adjustment of luma component Y'. The adjusted Y' is determined to maintain the correct luminance relationship, creating a feedback loop that ensures accurate color representation while preserving the benefits of non-linear transfer functions.
2Measurement precision
If binary search is used to find the correct luma value Y' that gives the correct luminance Yo, then accurate luminance matching is achieved, but the number of iterations varies and can be up to ten iterations for 10-bit values
Solution Approach 1:
The patent applies parameter changes by transforming the problem from searching for Y' in the full range [0, 1023] to calculating it directly using a derived formula. By changing the approach from iterative search to direct calculation based on the relationship between luminance and luma components, the method achieves the same accuracy with constant, minimal computation.
Solution Approach 2:
The patent uses a simplified calculation model that provides sufficient accuracy without requiring complex iterative algorithms. By using an approximate but efficient calculation based on luminance-luma relationships, the method achieves acceptable results with minimal computational resources, effectively replacing the more expensive binary search approach.
3Ease of manufacture
If linearization of the transfer function is used to calculate Y' approximately, then computation is simplified, but the result can sometimes lead to failures and severe artifacts
Solution Approach 1:
The patent applies preliminary action by adjusting the luma component Y' before chroma subsampling and color space conversion. By pre-adjusting Y' based on the original luminance Yo and the relationship between luma and chroma components, the method prevents sharp luma shifts that would otherwise cause artifacts in saturated colors after processing.
Solution Approach 2:
The patent implements feedback by using the original luminance Yo (calculated from original linear RGB values) as a reference to guide the adjustment of luma component Y'. The adjusted Y' is determined to maintain the correct luminance relationship, creating a feedback loop that ensures accurate color representation while preserving the benefits of non-linear transfer functions.
Data Source
Figure 1
Figure 2~4
Figure 3~6
AI summary
A processing of pixels comprises checking whether a color component of a pixel in a color space resulting in a smallest error between at least one color component in another color space determined based on the color component and at least one original color component of the pixel in the another color space causes any color channel of the pixel to fall outside of an allowed range. A value of the color component is obtained using a first function/LUT if the color component resulting in the smallest error does not cause any color channel of the pixel to fall outside of said allowed range. However, a value of the color component is obtained using a second, different function/LUT if the color component resulting in the smallest error causes any color channel of the pixel to fall outside of the allowed range.