Two-Stage Color ink Painting Style Transfer via Convolution Neural Network
Chengyu Zheng, Yuan Zhang · 2018
Despite the rapid progress in style transfer, it is challenging of the color ink painting style transfer, because of the ambient characters of plane sense, freehand brushwork and local line sense. In this paper, we explore how to transfer flower photo to color ink painting. Unlike traditional image processing methods, we treat it as a generative problem by taking the advantages of CNN and GAN. Different from common neural style transfer methods, we propose a method that imitates the creation process of color ink painting. Specifically, we divide the task into two subtasks - line drawing extraction and image colorization. Instead of using edge detection algorithms, we take the line drawing as a kind of style and exploit CNN-based neural style transfer method to obtain line drawing. As for image colorization, we use the GAN-based neural style transfer method. Experimental results show that our method performs better than the one-stage neural style transfer methods.