Image-difference prediction: from color to spectral.

Area of Science:

• Computational imaging
• Color science
• Human-computer interaction

Background:

• Evaluating multi and hyperspectral image quality is crucial for applications like remote sensing and digital archiving.
• Existing metrics often fail to capture human perceptual differences, especially under varying viewing conditions.
• Chromatic adaptation plays a significant role in how spectral distortions are perceived.

Purpose of the Study:

• To propose a novel strategy for evaluating multi and hyperspectral image quality from a human perception standpoint.
• To define and analyze spectral image difference under various illuminants.
• To develop computationally efficient metrics that align with human visual perception.

Main Methods:

• Analysis of seven image-difference features for stability across different illuminants using an information-theoretic strategy.
• Investigation of chromatic and achromatic feature variations under common spectral distortions (gamut mapping, compression, reconstruction).
• Development and validation of two new spectral image difference metrics through subjective visual experiments.

Main Results:

• Chromatic features exhibit greater variability than achromatic features across illuminants, even with chromatic adaptation.
• The proposed spectral image difference metrics demonstrate significant improvement over the root-mean-square error (RMSE).
• The new metrics show better correlation with subjective visual assessments of image quality.

Conclusions:

• Human perception is a critical factor in assessing spectral image quality.
• The developed metrics offer a more perceptually relevant and efficient way to evaluate multi and hyperspectral images.
• This work provides a foundation for improved image quality assessment in spectral imaging.