Towards Image Ambient Lighting Normalization

Lighting normalization is a crucial but underexplored restoration task withbroad applications. However, existing works often simplify this task within thecontext of shadow removal, limiting the light sources to one andoversimplifying the scene, thus excluding complex self-shadows and restrictingsurface classes to smooth ones. Although promising, such simplifications hindergeneralizability to more realistic settings encountered in daily use. In thispaper, we propose a new challenging task termed Ambient Lighting Normalization(ALN), which enables the study of interactions between shadows, unifying imagerestoration and shadow removal in a broader context. To address the lack ofappropriate datasets for ALN, we introduce the large-scale high-resolutiondataset Ambient6K, comprising samples obtained from multiple light sources andincluding self-shadows resulting from complex geometries, which is the first ofits kind. For benchmarking, we select various mainstream methods and rigorouslyevaluate them on Ambient6K. Additionally, we propose IFBlend, a novel strongbaseline that maximizes Image-Frequency joint entropy to selectively restorelocal areas under different lighting conditions, without relying on shadowlocalization priors. Experiments show that IFBlend achieves SOTA scores onAmbient6K and exhibits competitive performance on conventional shadow removalbenchmarks compared to shadow-specific models with mask priors. The dataset,benchmark, and code are available at https://github.com/fvasluianu97/IFBlend.