Detection of specular highlights in color images using a new color space transformation

This paper presents a novel approach to detecting specular highlights in color images taken under non-white illumination conditions. Considering significance of the dichromatic reflection model, our new color representation scheme helps to locate dichromatic planes in which specular highlights can be efficiently detected. We first address how to approximate the color of illuminant and define a new axis in the RGB color cube - namely illumination axis - that conics from the notion of the brightness of the illuminant color. Based on the estimation of the illuminant color, the RGB color space is to he transformed into a new color representation with respect to the estimated color of illuminant. Our proposed specularity detection scheme that consists of the following three steps: (1) determine the dichromatic plane of the object in our new color space, (2) detect both non-specular and unsaturated specular pixels in the dichromatic plane, and (3) detect saturated (clipped) specular pixels which Ire not in the dichromatic plane.