Deep Image Harmonization in Dual Color Spaces

Image harmonization is an essential step in image composition that adjuststhe appearance of composite foreground to address the inconsistency betweenforeground and background. Existing methods primarily operate in correlated$RGB$ color space, leading to entangled features and limited representationability. In contrast, decorrelated color space (e.g., $Lab$) has decorrelatedchannels that provide disentangled color and illumination statistics. In thispaper, we explore image harmonization in dual color spaces, which supplementsentangled $RGB$ features with disentangled $L$, $a$, $b$ features to alleviatethe workload in harmonization process. The network comprises a $RGB$harmonization backbone, an $Lab$ encoding module, and an $Lab$ control module.The backbone is a U-Net network translating composite image to harmonizedimage. Three encoders in $Lab$ encoding module extract three control codesindependently from $L$, $a$, $b$ channels, which are used to manipulate thedecoder features in harmonization backbone via $Lab$ control module. Our codeand model are available at\href{https://github.com/bcmi/DucoNet-Image-Harmonization}{https://github.com/bcmi/DucoNet-Image-Harmonization}.