Pattern recognition based color transfer

Color transfer is a recently proposed method for colorizing grayscale images. It transmits the appropriate chromatic "style" of a reference source color image to a target gray image. Compared with pseudocolor methods, color transfer provides more objective results. In this paper, the original color transfer method is reviewed briefly, and analyzed from the perspective of pattern recognition. In order to improve the quality of the colorized images and the computation efficiency of the colorizing process, several powerful techniques are combined to the original color transfer algorithm, including texture feature, sample reduction, median k-nearest-neighbor rule, KD tree and median filter. Experiments show that the new algorithm works well.