Neural Network Autoexposure and White Balancing via Illuminant Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing digital camera systems face challenges in accurately white balancing and autoexposure, particularly when dealing with scenes containing large areas of uniform color, as current methods rely on assumptions that fail in such cases, leading to incorrect color representation and exposure settings.
Innovation Solution
The use of neural networks trained to relate image exposure to illuminance parameters and automatic white balancing methods that extract color channel gains by comparing image prototype colors with reference colors at various color temperatures, along with sectoring in color space to reduce comparisons and adjust gains accordingly, and autoexposure methods that extract exposure settings from illuminance mean, variance, and maximum in subareas of an image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If the gray-world assumption is used for automatic white balancing, then the white balancing can be performed automatically without user intervention, but it fails when the scene contains large areas of uniform color leading to incorrect color representation
Solution Approach 1:
The patent divides the image into multiple regions and performs white balancing analysis on each region separately, then combines the results. This local analysis approach allows the system to handle scenes with large uniform color areas by not relying on the global average color assumption, thereby maintaining both automation and accuracy.
Solution Approach 2:
The image is segmented into multiple regions for independent analysis. By processing different regions separately and combining their white balancing results, the system avoids the pitfall of the gray-world assumption when large portions of the image have uniform color, thus resolving the contradiction between automation and precision.
2Measurement precision
If manual white balancing is performed by zooming into a pure white object, then accurate white balance can be achieved, but it requires user intervention and repeated adjustments when light source changes
Solution Approach 1:
The system performs white balancing automatically by analyzing the image content itself without requiring user intervention. It identifies reference regions and computes white balance parameters autonomously, eliminating the need for users to manually zoom into white objects or repeatedly adjust settings when lighting changes.
Solution Approach 2:
The system performs preliminary analysis of the image to identify suitable reference regions and pre-computes white balance parameters before final image processing. This preliminary action enables automatic white balancing that maintains accuracy without requiring subsequent user adjustments.
3Device complexity
If the average color of the scene is used to zero-out color bias, then white balancing can be performed using simple hardware circuitry, but it cannot properly compensate when there is a bias in the average color
Solution Approach 1:
The patent introduces an intermediary processing stage that analyzes image content to identify reliable reference regions, then uses these regions to compute white balance parameters. This intermediary step bridges the gap between simple hardware implementation and reliable white balancing by selecting appropriate reference areas that are not biased by dominant colors in the scene.
4Measurement precision
If neural networks are used for autoexposure and white balancing, then accurate color representation and exposure settings can be achieved in complex scenes, but the device complexity increases
Solution Approach 1:
The neural network processes segmented regions of the image independently, analyzing local characteristics to determine exposure and white balance parameters. This segmentation approach allows the complex neural network to focus on specific areas rather than the entire image, reducing computational complexity while maintaining high accuracy in complex scenes.
Data Source
AI summary
Automatic white balancing and/or autoexposure as useful in a digital camera extracts color channel gains from comparisons of image colors with reference colors under various color temperature illuminants and/or extracts exposure settings from illuminance mean, illuminance variance, illuminance minimum, and illuminance maximum in areas of an image with a trained neural network.