Detection of specularity using colour and multiple views

摘要

This paper presents a model and an algorithm for the detection of specularities from Lambertian reflections using multiple colour images from different viewing directions. The algorithm, called spectral differencing, is based on the Lambertian consistency that colour image irradiance from Lambertian reflection at an object surface does not change depending on viewing directions, but colour image irradiance from specular reflection or from a mixture of Lambertian and specular reflections does change. The spectral differencing is a pixelwise parallel algorithm, and it detects specularities by colour differences between a small number of images without using any feature correspondence or image segmentation. Applicable objects include uniformly or nonuniformly coloured dielectrics and metals, under extended and multiply coloured scene illumination. Experimental results agree with the model, and the algorithm performs well within the limitations discussed.