Painterly Image Harmonization in Dual Domains

Image harmonization aims to produce visually harmonious composite images by adjusting the foreground appearance be compatible with background. When image has photographic and painterly background, task is called harmonization. There are only few works on this task, which either time-consuming or weak in generating well-harmonized results. In work, we propose a novel network consisting of dual-domain generator discriminator, harmonizes both spatial domain frequency domain. The performs using AdaIN modules our proposed ResFFT discriminator attempts distinguish inharmonious patches based feature each patch, can enhance ability an adversarial manner. Extensive experiments benchmark dataset show effectiveness method. Our code model available at https://github.com/bcmi/PHDNet-Painterly-Image-Harmonization.