Elimination of Specular Surface-Reflectance Using Polarized and Unpolarized Light

Abstract

Highlights are an unwanted phenomenon in computer vision, they may severely hamper the use of standard image processing algorithms. Highlights are caused by specular reflectance, thus the objective is to eliminate this type of reflectance. In this paper, a new polarization-based method is introduced to separate the diffuse and specular component of reflection. A polarisation filter can reduce the intensity of higlights but can not completly eliminate this troublesome effect. Two images input with different orientation of a polarisation filter in front of a camera provide the necessary information to calculate the intensity of specular reflectance on plane surfaces. A third image with different orientation is necessary to determine specular reflectance in three-dimensional scenes. The specular reflectance is substracted from the original image, providing an image without highlights. Both polarized and unpolarized light is used in this paper. Polarized light has the advantage, that it reduces the degrees of freedom of reflected light, thus no knowlegde about surface properties and image aquisition geometry is required, when specular reflectance shall be removed. The method introduced in this paper can be applied to all kinds of dielectrics including textured surfaces. Highlights can be removed both from greylevel images and colour images, in the latter case yielding colour constancy. If the inspected objects are placed at a definite position, it is possible to aquire most of the required data for highlight-elimination during a training phase. In this way, highlights can be removed with just one image input, even if the source of illumination is changing. Thus the hardware-requirements to the vision system are considerably reduced, narrowing the gap between research and practical applications.