Real-time highlight removal using intensity ratio



In this paper, we propose an efficient method to separate the diffuse and specular reflection components

from a single image. The method is built on the observation that, for diffuse pixels, the intensity ratios

between the maximum values and range values (maximums minus minimums) are independent

of surface geometry. The specular fractions of the image pixels can then be computed by using the

intensity ratio. For textured surfaces, image pixels are classified into clusters by constructing a pseudochromaticity

space, and the intensity ratio of each cluster is robustly estimated. Unlike existing

techniques, the proposed method works in a pixel-wise manner, without specular pixel identification

and any local interaction.