Personalized Image Enhancement Using Global and Local Style Information

Image enhancement is a highly subjective task due to diverse and varying aesthetic preferences among individuals. However, most enhancement techniques pay little attention to the issue of personalization despite its importance. In this paper, we propose a transformer-based approach to personalized image enhancement, which enhances new images for a user by asking him or her to select a few preferred images from a random set of images. First, we represent various users’ preferences for enhancement as feature vectors — called preference vectors — in an embedding space. We construct the embedding space based on metric learning. Then, we develop a transformer-based enhancement network to enhance images adaptively according to each user’s preference vector. Experimental results demonstrate that the proposed algorithm is capable of achieving personalization successfully and outperforming conventional image enhancement algorithms.