基于卷积神经网络的蜡染染色模拟

Batik dyeing simulation based on convolutional neural network

蜡染是中国少数民族的传统技艺,广泛流行于中国西南地区。蜡染模拟是针对布料染色进行模拟,主要有裂纹模拟和布料染色模拟2个关键工作。此前的蜡染染色模拟方法只能进行单色染色,且对晕染模拟效果不明显。因此,提出一种基于卷积神经网络的蜡染多色染色方法。首先,使用基于距离变换的方法生成裂纹,并对裂纹的形状进行修正;然后,使用图像标注软件Labelme对图像特定区域进行手动分割,以方便之后对其进行染色;最后,使用PhotoWCT算法对图像的特定区域进行染色,并利用内容图中的像素相似性来平滑染色后的结果,使结果更接近真实的蜡染图像。实验结果表明,该方法能针对图像进行多色染色模拟,且在晕染效果方面较之此前的方法有明显提升,更接近于真实蜡染图像。

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.