Interactive Portrait Harmonization

1Johns Hopkins University, 2Adobe Research
Composite Image Mask Generic Harmonization Interactive Harmonization
Red mask represents the composite foreground region and Blue mask represents reference region chosen by user to guide the harmonization.

Abstract

We introduce Interactive Harmonization and propose a new framework which has the flexibility to solve it achieving realistic portrait harmonization.

Current image harmonization methods consider the entire background as the guidance for harmonization. However, this may limit the capability for user to choose any specific object/person in the background to guide the harmonization. To enable flexible interaction between user and harmonization, we introduce interactive harmonization, a new setting where the harmonization is performed with respect to a selected region in the reference image instead of the entire background. A new flexible framework that allows users to pick certain regions of the background image and use it to guide the harmonization is proposed. Inspired by professional portrait harmonization users, we also introduce a new luminance matching loss to optimally match the color/luminance conditions between the composite foreground and select reference region. This framework provides more control to the image harmonization pipeline achieving visually pleasing portrait edits. Furthermore, we also introduce a new dataset carefully curated for validating portrait harmonization. Extensive experiments on both synthetic and real-world datasets show that the proposed approach is efficient and robust compared to previous harmonization baselines, especially for portraits.

Method

Overview of the proposed Interactive Harmonization Framework.
 
Overview of the Luminance Matching Loss.

Results

Slide left to view the harmonized Portrait!
/> />
/> />
/> />
/> />
/> />
/> />

BibTeX

@article{valanarasu2022interactive,
  title={Interactive Portrait Harmonization},
  author={Valanarasu, Jeya Maria Jose and Zhang, He and Zhang, Jianming and Wang, Yilin and Lin, Zhe and Echevarria, Jose and Ma, Yinglan and Wei, Zijun and Sunkavalli, Kalyan and Patel, Vishal M},
  journal={arXiv preprint arXiv:2203.08216},
  year={2022}
}
Website Template taken from Nerfies