Investigating human color harmony preferences using unsupervised machine learning

Przemysław Skurowski, Michał Kozielski
Silesian University of Technology (Poland)
Download paper

Play (24min)

Download: MP4 | MP3

Color harmony patterns are relationships between coexisting colors where human psycho-perceptual visual pleasantness is the judging criterion. They play pivotal role in visualization, digital imaging and computer graphics. As a reference we assumed Itten model where harmony is expressed in terms of hue. The paper demonstrate investigation on color harmony patterns using clustering techniques.

Our source data was Adobe Kuler database consisting of hundreds of thousands of color palettes prepared for creative purposes. For the color palettes dissimilarity measurement we propose to use Jaccard distance additionally treating colors as the elements of a fuzzy set. Then, in the next step, separate colors are grouped within each group of palettes to specify each scheme of relations. The results are schemes of relationships between color within palettes.