Visible Light Spectrum Extraction from Diffraction Images by Deconvolution and the Cepstrum.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning

Background:

• Accurate color determination is challenging under variable lighting, often requiring specialized equipment.
• Consumer-grade cameras lack the capability for precise spectral analysis.

Purpose of the Study:

• To develop a low-cost method for extracting visible light spectra using standard camera sensors.
• To enable color measurements with consumer equipment by reconstructing spectral data.

Main Methods:

• Utilized a diffractive element attached to a standard camera to capture diffraction images.
• Developed and applied two machine learning algorithms based on deconvolution and cepstrum operations to form the light spectrum.
• Trained and evaluated methods on data from multiple cameras and illuminants, comparing against hyperspectral camera ground truth.

Main Results:

• The proposed machine learning methods successfully reconstructed light spectra and color with good accuracy across various conditions.
• Performance varied slightly depending on the specific camera and lighting, but remained robust.
• The approach demonstrated generalizability, performing well even with novel illuminants not used during training.

Conclusions:

• The developed computational imaging approach enables accurate, low-cost spectral and color measurements using ordinary cameras.
• The machine learning algorithms provide a versatile solution for spectral reconstruction, adaptable to different camera and lighting scenarios.
• This method significantly broadens the accessibility of precise color determination for consumer applications.