Handling Specularity in Intrinsic Image Decomposition

Abstract

Intrinsic image decomposition (IID), the process of separating an image into reflectance and shading components, is one of the fundamental problems in computer vision. Various approaches for IID have been proposed, but most assume Lambertian surfaces. In this paper, we propose a method that handles specularity while decomposing an input image into reflectance and shading components. The method first removes specularities from the image, and then it decomposes the image into reflectance and shading components. We propose a new algorithm for reconstruction of an image’s diffuse component and demonstrate the effectiveness of the method under specularity based on the extracted reflectance and shading images. Future work will focus on a more extensive empirical evaluation against ground truth and handling of shadows.