Chromatic differences between colours retrieved from RGB and hyperspectral images

Abstract

The common RGB digital camera (DC) is able to compress all the surrounding spectral information into a trichromatic system capable of represented the majority of the perceived colours. It follows the same compression philosophy as the human eye, where the spectral information is compressed into a three-dimensional color system. Despite the same apparent solution to compress the spectral information, the human eye and the DCs possess different spectral sensitivities and, in the end, provide different chromatic experiences. To which extend the human eye and the DCs provide the same chromatic capabilities, is still an open question. The spectral information of 50 hyperspectral images of natural scenes and 78 art paintings was used to estimate the chromatic differences between the human eye and a DC assuming the CIE D65 illuminant and the CIE1931 2° standard observer. The human eye chromatic diversity was estimated by computing the tristimulus values from the reflectance data. The DC chromatic diversity was estimated by assuming the spectral sensitivity of the DC to estimate the RGB colour from the spectral data and then converting it into the tristimulus values. The CIELAB chromaticity coordinates were then computed, as the pixel colour difference (∆E^*_ab) between the human eye and the DC. The ∆E^*_ab was assumed as a metric of the chromatic differences. It was found that the chromatic differences between the two systems were higher for paintings than for natural scenes. A Pseudo Voigt fit to the frequency distributions of the ∆E^*_ab between the human eye and the DC estimated the peak at about 6.70 (±0.06) and 3.56 (±0.07) ∆E^*_ab for art paintings and natural scenes, respectively. These results seem to indicate that DCs may provide better chromatic estimations on natural scenarios than on art paintings, when compared with the accuracy of hyperspectral imaging.

Meeting abstract presented at VSS 2018