结合绘画先验的线稿上色方法  

Sketch Colorization Method with Drawing Prior

作　　者：窦智 王宁 王世杰 王智慧[1] 李豪杰[1] DOU Zhi;WANG Ning;WANG Shi-jie;WANG Zhi-hui;LI Hao-jie(College of Software Technology,Dalian University of Technology,Liaoning 116000,China)

机构地区：[1]大连理工大学软件学院,辽宁大连116000

出　　处：《计算机科学》2022年第4期195-202,共8页Computer Science

基　　金：国家自然科学基金(61772108,61932020,61976038)。

摘　　要：自动线稿上色近年来已成为计算机视觉领域的研究热点之一。现有方法旨在通过改进网络架构或图像生成流程来提升上色的质量,但生成结果多存在色相集中、饱和度和明度分布不合理等现象。为此,提出一种结合绘画先验的线稿上色方法。该方法受插画师上色过程的启发,学习其广泛使用的绘画先验知识(如色相变化、饱和度对比和明暗对比)来提升自动线稿上色的质量。具体来讲,该方法在HSV色彩空间上增加了像素级损失,引导网络生成异常纹理较少的结果。同时,提出的三项启发式损失函数分别引入了色相变化、饱和度对比和明暗对比等绘画先验,引导网络生成具有合理色彩组成的上色结果。在真实线稿构建的测试数据集上,采用FID和MOS两项指标对所提方法和现有方法从生成结果与真实数据的分布相似度及视觉质量方面进行了比较。实验结果表明,相比性能第二的模型,所提方法的FID指标降低了21.00,MOS指标提高了0.96,因此所提线稿上色方法有效提升了自动线稿上色的视觉质量。Automatic sketch colorization has become an important research topic in computer vision.Previous methods intent to improve the colorization quality with advanced network architecture or innovative pipeline.However,they usually generate results with concentrated hue,unreasonable saturation and gray distribution.To alleviate these problems,this paper proposes a sketch colorization method with drawing priors.Inspired by the actual coloring process,this method learns the widely used drawing priors(such as hue variation,saturation contrast,and gray contrast)to improve the quality of automatic sketch colorization.Specifically,it incorporates pixel-level loss in the HSV color space to gain more natural results with less artifacts.Meanwhile,three heuristic loss functions that introduce the drawing priors such as hue variation,saturation and gray contrast are used to train our method to generate results with harmonious color composition.We compare our method with current state-of-the-art methods on test dataset constructed by real sketch images.Fréchet inception distance(FID)and mean opinion score(MOS)are adopted to measure the similarity between the distribution of real and generated images and the visual quality,respectively.Compared to the second-best method,the experimental results show that the FID of our method decreases by 21.00 and the MOS increases by 0.96,respectively.All the experimental results prove that the proposed method effectively improves the visual quality of automatic sketch colorization.

关 键 词：自动线稿上色 深度学习 对抗生成网络(GAN) 绘画先验 HSV色彩空间 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]