Focusing on nonphotorealistic rendering techniques of digital art entertainment, this paper investigates a method of color transformation from a reference famous picture to a target picture. A spatial filter based on an artificial neural network is proposed to transform the color information of the target picture. The learning of the neural network is undertaken in order to minimize the difference of histograms between the reference and target pictures by using genetic algorithm. Experimental results show the feasibility and the characteristics of the proposed method.