Efficient quantization of painting images by relevant colors

Abstract

Realistic images often contain complex variations in color, which can make economical descriptions difficult. Yet human observers can readily reduce the number of colors in paintings to a small proportion they judge as relevant. These relevant colors provide a way to simplify images by effectively quantizing them. The aim here was to estimate the information captured by this process and to compare it with algorithmic estimates of the maximum information possible by colorimetric and general optimization methods. The images tested were of 20 conventionally representational paintings. Information was quantified by Shannon's mutual information. It was found that the estimated mutual information in observers' choices reached about 90% of the algorithmic maxima. For comparison, JPEG compression delivered somewhat less. Observers seem to be efficient at effectively quantizing colored images, an ability that may have applications in the real world.

Figure 1

Montage of images of conventionally representational paintings in the Prado Museum, adapted from Ref., Fig. 1, licensed under CC BY 4.0. The color gamuts of each of these images are illustrated in Fig. 4 in Methods.

Figure 2

Information from relevant colors. Estimated mutual information between each of the 20 images in Fig. 1 and their quantized representations by relevant colors is shown for five clustering methods and six human observers, indicated by different symbols. Horizontal jitter has been added to reduce overlap between symbols. Estimates for individual images are not distinguished. Boxplots represent median (center line), mean (solid square), interquartile range (rectangle), and the 5th and 95th percentiles (whiskers).

Figure 3

Information and number of relevant colors. Estimated mutual information is plotted against logarithm to the base 2 of the number of relevant colors n for the five clustering methods and six human observers. Horizontal jitter has been added to reduce overlap between symbols. The dashed lines represent linear regressions on the unjittered data.

Figure 4

Color gamuts of painting images in Fig. 1. Each plot shows the CIELAB chromaticity coordinates (a^*, b^*) of pixels drawn uniformly from the corresponding image in Fig. 1, which has the same row-column layout. Only about 1% of pixels are plotted to more clearly reveal density variations.