Photometric-based recovery of illuminant-free color images using a red-green-blue digital camera

Abstract

Albedo estimation has traditionally been used to make computational simulations of real objects under different conditions, but as yet no device is capable of measuring albedo directly. The aim of this work is to introduce a photometric-based color imaging framework that can estimate albedo and can reproduce the appearance both indoors and outdoors of images under different lights and illumination geometry. Using a calibration sample set composed of chips made of the same material but different colors and textures, we compare two photometric-stereo techniques, one of them avoiding the effect of shadows and highlights in the image and the other ignoring this constraint. We combined a photometric-stereo technique and a color-estimation algorithm that directly relates the camera sensor outputs with the albedo values. The proposed method can produce illuminant-free images with good color accuracy when a three-channel red-green-blue (RGB) digital camera is used, even outdoors under solar illumination.1 Introduction Our visual system allows us to distinguish elements such as color, texture, and surface shape, and our brains use this information to recognize objects under different lighting and observation conditions. Computational vision uses imaging devices to simulate the way the human visual system analyzes color images. Reproducing color appearance in color-based image displays depends on several factors such as the imaging device itself, the geometry of the lighting conditions, and the surface properties of the objects in question. When dealing with textured objects these factors become more evident; for example, the position of the camera and the object must be kept fixed because the color appearance of the object can change with the direction of illumination, which can cause a problem for object characterization. Thus, extracting information from textured images independently of lighting and imaging geometries poses a challenge for many computational vision systems. The use of the albedo rather than raw sensor response data can help to solve this issue. Albedo is usually defined as the ratio of scattered flux to that scattered and absorbed by a monodisperse or relatively uniform suspension, i.e., the ratio of flux scattered to the quantity of incident flux. 1 As opposed to intensity, albedo does not depend on the illumination conditions, and has been used in computer graphics to derive illuminantinvariant properties. 2 But the light reflected from a surface depends on the surface's spectral characteristics, the illumination geometry, the spectral content of the incident light, the viewing direction, and the roughness of the surface. This means that albedo recovery should not only be insensitive to the intensity of illumination but also to spectral changes in the reflected light.