Rank-based camera spectral sensitivity estimation.

Area of Science:

• Digital imaging and color science.
• Computational photography.
• Sensor characterization.

Background:

• Accurate spectral sensitivity functions are essential for predicting digital camera responses to spectral stimuli.
• Traditional methods involve difficult lab measurements or statistical inference, which assumes camera linearity.
• Non-linearities in rendered images, like JPEGs, significantly degrade the accuracy of existing spectral sensitivity estimation techniques.

Purpose of the Study:

• To develop a novel technique for estimating camera spectral sensitivity functions from both linear and rendered images.
• To address the limitations of regression-based methods when camera response linearity is violated.
• To enable spectral sensitivity estimation for devices lacking raw image output capabilities.

Main Methods:

• A novel rank-based method is proposed to estimate spectral sensitivity functions.
• The method leverages the rank order of camera responses to impose constraints on the sensor's spectral sensitivity curve.
• Feasible regions for the sensor space are determined by intersecting constraints from ranked response pairs.

Main Results:

• The rank-based method achieves estimation accuracy comparable to prior art for linear data.
• A significant improvement in estimation performance is observed for non-linear (rendered) image data.
• The method successfully estimates effective spectral sensitivities from rendered images, including those without raw mode.

Conclusions:

• The proposed rank-based approach provides a robust method for estimating camera spectral sensitivities.
• This technique overcomes the limitations of linearity assumptions, offering superior performance for rendered images.
• It broadens the applicability of spectral sensitivity estimation to a wider range of digital imaging devices.