Light Field Analysis for Non-Lambertian Scenes

Abstract

The world around us consists of objects that are composed of several different types of materials. One of the methods through which we can examine the environment is to analyse how objects visually react to incident light. One dominant category is diffuse materials, which radiate light in all visible directions. If the light is further radiated uniformly, then the object's material is called Lambertian. The visual properties of Lambertian materials are popular for numerous image processing approaches, largely due to their ease of handling. However, it is a very limiting category. For instance, many objects are fabricated from metals or plastic, which are often glossy or even specular and control the distribution of light based on rules different from those of Lambertian materials. Therefore, non-Lambertian materials have many undesired properties, particularly for inverse problems. One of the most signifcant of these is that the non-Lambertian materials suffer from viewpoint variance. However, this can also be highly informative about the scene, especially when the scene has been sampled from several different viewpoints. Plenoptic devices provide angular sampling, which can effciently complement the geometric properties of non-Lambertian materials and the undesired viewpoint dependency. Thus, they can reveal additional scene information, unlike the traditional 2D images. In this thesis, we show several inverse problems and use the dense angular sampling of a plenoptic camera to extract information about the scene using the viewpoint variance of non-Lambertian materials.