An Interactive System for Painterly Image Harmonization
Juwon Lee, Jung-Jae Yu, Wonyoung Yoo · 2023
Image harmonization is the process of adjusting the appearance of an object of interest that is copied from a source image and pasted onto a target image to make the composite image look seamless and natural. Painterly image harmonization represents a distinct category within image harmonization where the target image has a painterly style. To obtain the harmonized results, it involves several sub-tasks: segmenting the object of interest from a photograph, creating a naive composition that includes placing and scaling the object onto the target painting, generating a mask for the pasted object, and applying painterly image harmonization to the naive composite image. To facilitate these processes within a unified framework, we integrated object segmentation and painterly image harmonization methods, with a user-friendly graphical user interface. The proposed system enables users to easily create new artworks by placing objects from photos into paintings and harmonizing them according to their intention. In addition, we designed the harmonization pipeline employing a style transfer method and compared it with the conventional harmonization method, E2STN. Qualitative results show the effectiveness of our proposed approach.