Computational Design Driven by Aesthetic Preference
Yuki Koyama · 2016
Tweaking design parameters is one of the most fundamental tasks in many design domains. In this paper, we describe three computational design methods for parameter tweaking tasks in which aesthetic preference---how aesthetically preferable the design looks---is used as a criterion to be maximized. The first method estimates a preference distribution in the target parameter space using crowdsourced human computation. The estimated preference distribution is then used in a design interface to facilitate interactive design exploration. The second method also estimates a preference distribution and uses it in an interface, but the distribution is estimated using the editing history of the target user. In contrast to these two methods, the third method automatically finds the best parameter that maximizes aesthetic preference, without requiring the user of this method to manually tweak parameters. This is enabled by implementing optimization algorithms using crowdsourced human computation. We validated these methods mainly in the scenario of photo color enhancement where parameters, such as brightness and contrast, need to be tweaked.