MODELING LIGHT-REFLECTION FOR COMPUTER COLOR-VISION

In computer vision applications, analysis of shading information requires a proper model of light reflection from object surfaces. The most often used model so far relates the image intensity I to the light source intensity E by a simple relation: I = Eph (θ, ø) cos θ, where θ is the incident angle, ø is the viewing angle with respect to the surface normal, ρ is the so-called surface reflectance, and h (θ, ø), normalized to one at the maximum, is called the reflectivity function, representing the angular dependence of light reflection. Although the terminology is not of standard use, this model seems to be empirically true for a collimated monochromatic incoherent light source. However, the optical geometry and the sensor spectral responsivity of the vision (imaging) system are not explicitly considered in the model. Furthermore, it is not clear that one can always model the light reflection in a product form for a polychromatic light source. To overcome the above shortcoming and to extend the reflection model for computer color vision, we examine the light reflection problem using the bidirectional spectral-reflectance distribution function (BSRDF) to specify both incident- and reflected-beam geometry. It is shown that the product form can still be retained for a polychromatic light source under two lighting conditions: 1) the light source is collimated; or 2) the spectral factor and the geometric factor can be separated for both the light source and the BSRDF of the surface. The reflection model is then applied to the formulation of a neutral-interface-reflection model, which is tested experimentally. The results show the adequacy of this type of model for surfaces of some material compositions, e.g., plastics, plant leaves, painted surfaces, orange peels, and some glossy cloth, but not for others, e.g., colored paper and some ceramics. © 1990 IEEE