Shadow Patching: Guided Image Completion for Shadow Removal

Removing unwanted shadows is a common need in photo editing software. Previous methods handle some shadows well but perform poorly in cases with severe degradation (darker shadowing) because they rely on directly restoring the degraded data in the shadowed region. Image-completion algorithms can often replace severely shadowed regions with appropriately simulated content, and perform well with smaller-scale textures, but often fail to reproduce larger-scale macrostructure still visible in the shadowed region. This paper provides a general framework that leverages degraded (shadowed) data in a region to guide image completion by extending the objective function commonly used in patch-based image completion to include not only visual realism but consistency with the shadowed content. This approach achieves realistic-looking shadow removal even in cases of severe shadowing where precise recovery from the degraded signal may not be possible.