Optimal Multispectral Imaging using RGB Cameras
A new physics-driven method uses off-the-shelf RGB cameras and filters to capture 12 spectral bands with optimal noise robustness.
A team of researchers has developed a novel framework that enables high-quality multispectral imaging using inexpensive, off-the-shelf components. The method, detailed in the arXiv paper 'Optimal Multispectral Imaging using RGB Cameras' by Tomislav Matulić, Ivan Škrabo, Dubravko Babić, and Damir Seršić, starts by formulating a linear measurement model. Here, camera responses are expressed as mixtures of unknown spectral components, with mixing coefficients determined by the overlap between camera spectral sensitivities and filter transmittances. For a multi-camera setup, these per-camera models are stacked into a single global system.
The core innovation is treating wavelength allocation as a deterministic design problem. The researchers select the configuration that minimizes the spectral condition number of the resulting system matrix—a frame-theoretic criterion that promotes numerical stability, maximizes the worst-case output signal-to-noise ratio (SNR), and improves reconstruction robustness. They demonstrated the method on a representative setup targeting 12 wavelengths using four triband filters, evaluating all feasible configurations within the finite design space to identify the most stable and noise-robust arrangement. The framework also supports redundant configurations, where individual wavelengths are measured by multiple cameras, providing additional degrees of freedom to further enhance noise robustness.
- Uses a physics-driven linear model to turn standard RGB cameras and optical filters into a multispectral imaging system.
- Formulates wavelength allocation as a design problem, minimizing the system's spectral condition number to maximize stability and noise robustness.
- Demonstrated on a setup with 12 target wavelengths and four filters, enabling evaluation of all feasible configurations for optimal performance.
Why It Matters
Dramatically lowers the cost and complexity of multispectral imaging, making advanced spectral analysis accessible for agriculture, medicine, and material science.