Fascinating link, thank you! On that note though: How are the horizontal bars (first one right below the introduction) made? My guess is that they gathered all colours from all pixels, made a histogram, and then converted that into a horizontal stacked bar. However, the ordering seems weird, as some colours look like they appear multiple times (say for example the almost black-ish regions), disproving my histogram hypothesis. Then again, that may be due to image compression. And ordering the colour vectors would highly depend on the used colour space & the used partial order. I similarly stumbled over the image-version of these bars in section "The colours within a single object" further down, which I just don't quite understand how they're made.

Does someone have insight on the methodology or maybe a link to a paper? The present methodology section seems to focus on object similarity.