Hyperspectral Image Processing for Light Source Type Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current auto white balancing algorithms in digital cameras struggle to accurately estimate light source type and spectral power distribution due to their reliance on RGB data, leading to metamerism issues where colors appear differently under varying illumination conditions, and fail to separate light sources with similar correlated color temperatures.
Innovation Solution
The method employs hyperspectral imaging sensors to acquire and process images, normalizing spectral power distribution to determine light source type estimates, which are then used to improve lens shading correction and color conversion matrices, enabling more accurate color quality and separation of differently illuminated regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RGB data is used for auto white balancing, then the processing complexity is low, but the light source type estimation accuracy is insufficient
Solution Approach 1:
The patent transitions from 3-channel RGB data to hyperspectral imaging with multiple spectral bands (e.g., 5-10 bands), adding the spectral dimension to enable accurate light source type estimation. This dimensional expansion allows differentiation between light sources with similar correlated color temperatures but distinct spectral signatures, resolving the metamerism problem while maintaining computational feasibility through structured spectral processing
2Reliability
If conventional AWB algorithms are used, then the color constancy is maintained, but metamerism issues occur under varying illumination
Solution Approach 1:
The patent replaces conventional RGB-based AWB algorithms with hyperspectral imaging processing that utilizes multiple spectral bands. This substitution enables direct measurement of spectral power distribution across the visible spectrum, allowing accurate differentiation between light sources and objects, thereby eliminating metamerism artifacts while maintaining color constancy through physics-based spectral analysis rather than heuristic algorithms
3Measurement precision
If HSI sensor with multiple spectral bands is used, then the spectral resolution is improved, but the device complexity increases
Solution Approach 1:
The patent segments the spectral response into multiple discrete bands (e.g., 5-10 wavelength regions) rather than using continuous spectral measurement. This segmentation approach achieves sufficient spectral resolution for light source identification while keeping the sensor design manageable. The hyperspectral sensor is divided into multiple channels, each capturing energy in specific wavelength ranges, enabling processed spectral data without requiring fully continuous spectral measurement hardware
4Productivity
If multiple light sources with similar CCT are treated as identical, then the processing is simplified, but the color quality is degraded
Solution Approach 1:
The patent applies local quality analysis by examining the spectral power distribution at different wavelength regions to distinguish between light sources. Instead of treating all light sources with similar correlated color temperatures as identical, the system analyzes localized spectral characteristics (e.g., specific wavelength peaks or valleys) to identify unique properties of each light source type. This enables differentiated processing for incandescent, fluorescent, LED, and other light sources, improving color quality while maintaining reasonable processing efficiency through targeted spectral analysis
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method for image processing, preferably implemented in a camera device, including acquiring a Hyperspectral Imaging, HSI, image by an HSI sensor of a field-of-view; normalizing of the HSI information to determine a normalized Spectral Power Distribution, SPD, of the field-of-view; determining from the normalized SPD a light source type, LST, estimate; and processing the image on the basis of the determined LST estimate.