Intrinsic Image Decomposition

Intrinsic image decomposition studies how to separate an image into properties such as surface reflectance or albedo and shading or illumination, which helps explain appearance and enables relighting and material editing. Methods combine learned image priors, physical rendering constraints, and reconstruction or cycle-consistency losses; newer approaches also use multi-view geometry, video, and 3D scene representations to enforce consistency across viewpoints. Progress depends on better real-world and specialized data, including outdoor, polarized, and hyperspectral imagery, as well as evaluation that separates training and test scenes. A central challenge remains that different combinations of material and lighting can produce similar images, especially with shadows, specular reflections, and uneven illumination.