抄録
This paper describes a method for estimating the color vectors of a light source and an object surface by means of line detection from a histogram of the color image. The object is supposed to be an inhomogeneous dielectric object such as plastics with a smooth surface. First, we suggest the characteristics of the color histogram in a RGB space. The histogram forms a combination of two straight clusters of highlight pixels by specular reflection, and matte pixels by diffuse reflection. The estimation of two color vectors is then reduced to finding the gradients of straight lines for the two clusters. Next, basic algorithms for extracting a line from a cluster are shown using the principal component analysis and Hough transform. Moreover, an effective algorithm is proposed based on iteratively splitting and merging the histogram on the color signal plane. Finally, the usefulness of the method is shown in an experiment.
