Deep Image Harmonization in Dual Color Spaces

Abstract

Image harmonization is an essential step in image composition that adjusts the appearance of composite foreground to address the inconsistency between foreground and background. Existing methods primarily operate in correlated 𝑅𝐺𝐵 color space, leading to entangled features and limited representation ability. In contrast, decorrelated color space (e.g., 𝐿𝑎𝑏) has decorrelated channels that provide disentangled color and illumination statistics. In this paper, we explore image harmonization in dual color spaces, which supplements entangled 𝑅𝐺𝐵 features with disentangled 𝐿, 𝑎, 𝑏 features to alleviate the 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 harmonized image. Three encoders in 𝐿𝑎𝑏 encoding module extract three control codes independently from 𝐿, 𝑎, 𝑏 channels, which are used to manipulate the decoder features in harmonization backbone via 𝐿𝑎𝑏 control module. Our code and model are available at https://github.com/bcmi/DucoNet-Image-Harmonization.