Intrinsic Properties of an Image with Highlights

Shading and reflectance images, which are commonly called intrinsic images, are useful in many computer vision applications. A number of methods have been proposed to extract those images. Unfortunately, all of these methods assume diffuse only reflections and deem highlights to be outliers. To overcome the presence of highlights, usually reflection components separation is applied before attempting to recover intrinsic images. However, it is well known that reflection components separation itself is an intractable task that could pose additional problems. In this paper, we present a method that not only extracts the intrinsic images from an image with highlights but also decomposes the image reflection components in a single integrated framework. By considering reflection components as part of image intrinsic properties, and then using our method, we can extract four type properties: shading image, reflectance image, diffuse reflection component and specular reflection component. The method requires only a single image, without knowledge of the 3D geometrical data of the objects. The method is based on the specular-free image, a diffuse component pseudo image that has a geometrical profile exactly identical to the diffuse component of the input image, and can be generated using a local (pixel-based) operation. The method is effective even if the input image has textured surfaces.