White Balance Algorithm Boundary Pixel Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing white balance algorithms fail to accurately compensate for changing illumination spectra, especially in scenes with large monochrome objects, leading to incorrect color rendition and white balance.
Innovation Solution
A method that selectively gathers white balance statistics by identifying pixels on the boundaries of monochrome image segments, avoiding inner areas and pixels with small chrominance values, and employing both coarse and fine modes of correction to distinguish between illumination effects and object colors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If gray world model is used to estimate white balance, then color shift compensation is simplified, but accuracy deteriorates when large monochrome objects are present
Solution Approach 1:
The image is segmented into different regions based on chrominance values and spatial relationships. Pixels are classified as belonging to monochrome objects or non-monochrome objects based on their color characteristics and neighborhood relationships. This segmentation allows the algorithm to treat different regions differently, maintaining accuracy even when large monochrome objects are present.
Solution Approach 2:
Different quality criteria are applied to different regions of the image. For pixels identified as belonging to non-monochrome objects, standard gray world statistics are used. For pixels in monochrome object regions, alternative statistics gathering methods are applied, such as restricting statistics to only non-monochrome pixels or using spatial filtering to exclude monochrome regions from white balance calculation.
2Device complexity
If all pixels are used for white balance statistics, then statistics gathering is simplified, but reliability deteriorates due to contamination by monochrome objects
Solution Approach 1:
Pixels belonging to monochrome objects are extracted and removed from the set of pixels used for white balance statistics gathering. The algorithm identifies monochrome object pixels through chrominance thresholding and spatial analysis, then excludes these pixels from the statistics calculation. This extraction ensures that only pixels representing actual scene content (non-monochrome objects) contribute to white balance estimation.
Solution Approach 2:
Instead of including all pixels and then filtering out problematic ones, the algorithm inverts the approach by proactively identifying and excluding monochrome object pixels from the statistics gathering process. This inversion prevents contamination before it affects the results, maintaining reliability without requiring complex post-processing.
3Measurement precision
If chrominance-based filtering is applied to exclude monochrome pixels, then white balance accuracy improves, but processing complexity increases
Solution Approach 1:
The algorithm applies chrominance-based filtering selectively rather than universally. Instead of processing every pixel with complex monochrome detection, the algorithm first identifies regions with high chrominance values that are likely to be monochrome objects, then applies more sophisticated filtering only to those regions. This partial action reduces overall processing complexity while maintaining precision where needed.
Data Source
AI summary
A method for determining a correction for a spectra of scene illumination based on analysis of color content of a video image, in which average image colors are obtained by considering only those image pixels that have significantly different color from any of its neighboring pixels resulting in selection of only those pixels that lay on the boundaries of monochrome image segments, while avoiding their inner areas. Therefore, the white balance method is not only tolerant to the presence of large non-gray monochrome areas in the image, but also avoids the image areas that have small chrominance values and appear to be gray due to specific illumination spectra, while not actually corresponding to the grey areas of the scene.