Abstract
This paper proposes a method to correct the color-mismatch images of paintings based on artists' color features, and reproduce the exact color images under a specific illumination condition. First, we describe a standard image database consisting of oil paintings and an algorithm for extracting artists' color features. The color distribution is analyzed by PCA and described with an ellipsoid to represent the standard color features. Next, the color correction is based on the coordinate transformation of pixel values in a color-mismatch image so that the color features are fitted to the artist's color features. We present the correction algorithms using two color spaces of the sRGB and the CIELAB. Moreover, we describe a procedure for rendering the exact color images under a favorite illuminant. The correlated color temperature of scene illuminant is estimated and the color image under D65 is further transformed to a target image. The feasibility of the proposed method to recover the exact color images is examined with color-mismatch image samples of famous oil paintings.
