Abstract: Batik, a traditional art of the Chinese ethnic minority, is widely popular in southwest China. The Batik simulation mainly includes two major tasks: crack simulation and fabric dyeing simulation. In this paper, we mainly focus on fabric dyeing simulation. Previously, the method of Batik dyeing simulation can only perform monochromatic dyeing, and the simulation effect of self-dyeing is not obvious. Thus, a multicolor batik dyeing method based on convolutional neural network is proposed. To begin with, the crack was generated by the method based on distance transformation and then the shape of the crack is modified. Furthermore, we used Label me to segment the specific areas of the image manually, which facilitated the subsequent coloring. Finally, the PhotoWCT algorithm was applied to dye the specific regions of the image, and then the pixel affinities in the content map were used to smooth the dyed result. In this way, the result obtained is closer to the real Batik image. The experimental results indicate that, the method proposed in this article is applicable to multicolor dyeing simulation. In addition, this method is superior to the previous method in terms of sfumato effect.

Key words: batik dyeing, distance transform, image segmentation, PhotoWCT algorithm, sfumato technique