Modeling how people extract color themes from images

Abstract

Color choice plays an important role in works of graphic art and design. However, it can be difficult to choose a compelling set of colors, or color theme, from scratch. In this work, we present a method for extracting color themes from images using a regression model trained on themes created by people. We collect 1600 themes from Mechanical Turk as well as from artists. We find that themes extracted by Turk participants were similar to ones extracted by artists. In addition, people tended to select diverse colors and focus on colors in salient image regions. We show that our model can match human-extracted themes more closely compared to previous work. Themes extracted by our model were also rated higher as representing the image than previous approaches in a Mechanical Turk study.

Quantifying Aesthetics of Visual Design Applied to Automatic Design

“…To extract color palettes from images we used techniques from [57], saliency map [164], and segmentation [165]. These color palettes are used for visualizing the inferred color combinations.…”

“…These palettes are drawn from a pool of 5-color palettes in the dataset. As mentioned before, this pool is created by using techniques from [57].…”

“…This type of query is mapped to the word topics, and from them to their corresponding color topics. We then map these color topics to a set of color palettes in a pool of color palettes, created from the magazine dataset using techniques in [57]. The user can then choose his/her preferred color palettes from the recommended set.…”

A study of image colourfulness

“…Results are impressive, but they do not take into account how hue manipulation affects the colourfulness or contrast of the image. O'Donovan et al [2011] look at creating harmonious colour templates by harnessing data from online communities and Amazon Mechanical Turk, while Lin and Hanrahan [2013] model the way people extract colour themes from images.…”

Autonomous color theme extraction from images using saliency

“…That is, they cannot effectively capture the various and yet fine colors that might occur in an artwork. Recently, Lin and Hanrahan [4] introduced a different solution * ; a regression model trained on 1600 collected color palettes for 40 images from 160 human participants. Their model includes six types of features among which saliency is reported to be the main feature.…”