Understanding the effect of correlated colour temperatures on spatio-chromatic properties of natural images

Abstract

Despite the natural occurrence of global and local daylight changes in natural scenes, the human visual system typically adapts well to these changes and develops stable colour perception. In a previous study, the influence of daylight characterized by its Correlated Colour Temperatures (CCT) on different chromatic descriptors was analysed (Ojeda et al., 2017). The results showed that chromatic information is almost constant for CCT values above 14,000 K, with local extremes occurring in the range of low CCTs. The aim of this work is to extend the analysis of the CCT dependence of the illuminant to those that consider the spatio-chromatic structure, including second order descriptors (gradients, spectral slope, spectral signature, and PCA) and higher order descriptors (kurtosis, skewness, and number of relevant colours). Our results show that most of the descriptors exhibit horizontal asymptotic behaviour for CCTs above 15,000 K and local extremes in the range of 3,900 K-9,600 K. For those descriptors that could be analysed in CIELAB space, sufficient statistical evidence was obtained to consider skewness, kurtosis, and the independent spectral slopes of the L* channel as equal in the range of CCTs used. However, the slight variations in spectral signatures and the directions of the principal components when applying PCA to image patches are not statistically significant and cannot be considered equal under different illuminants. The number of relevant colours (NRC) exhibits sensitivity to temperature variations and behaves similarly to the other descriptors, due to its small number.