DBSCAN聚类算法在图像风格迁移中的应用
Application of DBSCAN Clustering Algorithm in Image Style Transfer
DOI: 10.12677/CSA.2021.1112308, PDF,  被引量    科研立项经费支持
作者: 王飞龙, 田江波, 蒋博蓝, 曾文广, 叶忠勇：北京信息科技大学计算机学院，北京；黄宏博*：北京信息科技大学计算机学院，北京;北京信息科技大学计算智能研究所，北京
关键词: 图像风格迁移；DBSCAN；聚类算法；生成对抗网络；Image Style Transfer； DBSCAN； Clustering Algorithm； Generative Adversarial Network
摘要: 尽管图像风格迁移技术在不断发展，但现阶段对于图像风格迁移的效果而言仍有较大的提升空间。为了使图像在风格迁移后具有与目标风格图像更贴近的颜色特征，本文基于GAN (Generative Ad-versarial Network)模型提出了一种使用DBSCAN (Density-Based Spatial Clustering of Application with Noise)聚类算法定义颜色数量的损失函数并用于约束模型的训练，从而使生成的图像更趋近于目标风格图像。实验结果表明在引入由DBSCAN算法定义的颜色数量损失函数后图像风格迁移效果得到了更好的提升，生成图像在颜色上更具有原风格图像的特征，视觉上更具立体感。
Abstract: Although the image style transfer technology is constantly developing, there is still much room for improvement in the effect of image style transfer. In order to make the image have similar color characteristics with the target style image after the style transfer, this paper introduces a color-quantity-criteria defined by the DBSCAN (Density-Based Spatial Clustering of Application with Noise) clustering algorithm. The transfer model is designed based on the GAN (Generative Adversarial Network). The color-quantity-loss-function is used to constrain the training of the model, so that the generated image is closer to the style image in the number of colors. The experimental results show that the image style transfer effect of the proposed algorithm is better improved, and the generated image not only has the characteristics of the original style image in color but also has more stereoscopic visual effect.
文章引用：王飞龙, 田江波, 蒋博蓝, 曾文广, 叶忠勇, 黄宏博. DBSCAN聚类算法在图像风格迁移中的应用[J]. 计算机科学与应用, 2021, 11(12): 3051-3059. https://doi.org/10.12677/CSA.2021.1112308

参考文献

[1] Gatys, L.A., Ecker, A.S. and Bethge, M. (2015) A Neural Algorithm of Artistic Style. arXiv:1508. 06576.
[2] Li, C. and Wand, M. (2016) Combining Markov Random Fields and Convolutional Neural Networks for Image Synthesis. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, 27-30 June 2016, 2479-2486 [Google Scholar] [CrossRef] 
[3] Liao, J., Yao, Y., Yuan, L., et al. (2017) Visual Attribute Transfer through Deep Image Analogy. arXiv:1705. 01088.
[4] Mirza, M. and Osindero, S. (2014) Conditional Gener-ative Adversarial Nets. ArXiv:1411.1784.
[5] Zhu, J.Y. and Park, T., et al. (2017) Unpaired Image-to-Image Transla-tion using Cycle-Consistent Adversarial Networks.
https://arxiv.org/abs/1703.10593
[6] Chen, Y., Lai, Y.K. and Liu, Y.J. (2018) CartoonGAN: Generative Adversarial Networks for Photo Cartoonization. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 18-23 June 2018, 9465-9474. [Google Scholar] [CrossRef] 
[7] Chen, J., Liu, G. and Chen, X. (2019) AnimeGAN: A Novel Lightweight GAN for Photo Animation. In: Li, K.S., Li, W., Wang, H. and Liu, Y., Eds., Artificial Intelligence Algo-rithms and Applications, Springer, Singapore, 242-256. [Google Scholar] [CrossRef] 
[8]
https://github.com/TachibanaYoshino/AnimeGANv2
[9] 彼佳. 动漫人脸数据集[EB/OL].
https://www.kaggle.com/defileroff/comic-faces-paired-synthetic-v2, 2021-05-14.