Collaborative personalization of image enhancement

This paper presents methods for personalization of image enhancement, which could be deployed in photo editing software and also in cloud-based image sharing services. We observe that users do have different preferences for enhancing images and that there are groups of people that share similarities in preferences. Our goal is to predict enhancements for novel images belonging to a particular user based on her specific taste, to facilitate the retouching process on large image collections. To that end, we describe an enhancement framework that can learn user preferences in an individual or collaborative way. The proposed system is based on a novel interactive application that allows to collect user's enhancement preferences. We propose algorithms to predict personalized enhancements by learning a preference model from the provided information. Furthermore, the algorithm improves prediction performance as more enhancement examples are progressively added. We conducted experiments via Amazon Mechanical Turk to collect preferences from a large group of people. Results show that the proposed framework can suggest image enhancements more targeted to individual users than commercial tools with global auto-enhancement functionalities.