What is the Color of Chocolate?

What is the color of chocolate? - extracting color values of semantic expressions

A. Lindner; N. Bonnier; S. Süsstrunk

2012. IS&T 6th European Conference on Colour in Graphics, Imaging, and Vision (CGIV 2012). , University of Amsterdam, Amsterdam, Netherlands , May 6-9, 2012. p. 355 - 361.

We present a statistical framework to automatically determine an associated color for a given arbitrary semantic expression. The expression can not only be a color name but any word or character string. In addition to the color value, we are also able to compute the result's significance, which determines how meaningful defining the color is for the expression. To demonstrate the framework's strength we apply it to two well known tasks: assessing memory colors and finding the color values for a given color name (color naming). We emphasize that we solve these tasks fully automatic without any psychophysical experiment or human intervention. Further, we outline the potential of our automatic framework and in particular the significance for the imaging community.

Results: semantic expressions and associated colors

Using freely available Flickr images our algorithm estimates a color value for an arbitrary semantic expression. The top five rows of the figure below show results for color names, the bottom two rows for other semantic expressions related to color.

color estimates for 70 semantic expressions
[click on the figure for higher resolution]

 

Other color estimates for different shades of the three basic memory colors: vegetation, skin and sky.

color estimates different shades of memory colors

Method in brief

Using Flickr’s API we acquire for each semantic expression two hundred associated images (i.e. images that have been annotated with that expression). We then carry out a statistical test that assesses for each color bin whether the presence of a keyword makes it likely to have a significantly higher (lower) bin count. This is reflected by a positive (negative) standardized significance value as sgown in the figure below.

 

Left: Significance distribution in CIELAB color space for keyword ferrari. The predominant feature is the well localized maximum in the red region. Right: Same plot for keyword color. The negative values along the neutral gray axis indicate a significant absence of these color values.

Source code

Download the source code package for this publication.

A more recent version of the code for color value estimation is available on the webpage of our new research project: A large-scale multi-lingual color thesaurus.

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.

Contact

Questions and comments about the code and method in general are very welcome and can be sent to Albrecht Lindner.