LI Quanxue,NIU Mengchen,CHEN Ruilin,et al.Image Style Conversion Algorithm based on Generative Confrontation Network[J].Journal of Chengdu University of Information Technology,2021,36(06):629-633.[doi:10.16836/j.cnki.jcuit.2021.06.008]
基于生成对抗网络的图像风格转换算法
- Title:
- Image Style Conversion Algorithm based on Generative Confrontation Network
- 文章编号:
- 2096-1618(2021)06-0629-05
- Keywords:
- image style conversion; GAN; Cycle-GAN; U-Net residual network
- 分类号:
- TP302.4
- 文献标志码:
- A
- 摘要:
- 针对图像样式转换产生的图像质量不高的问题,提出一种基于生成对抗网络的高质量图像样式转换方法。借鉴循环GAN网络结构上的发电网络相结合的方法采用跳层结构和U-Net网络中的残差网络,增强网络的多尺度不变性; 其次,在判别网络方面,提出一种多尺度扩展卷积判别器,以改善图像样式的空间几何变换。实验证明,与Cycle-GAN算法相比,该算法在图像样式转换中的效果有很大提升,图像样式转换的质量也得到了提高。
- Abstract:
- Aiming at the problem of low image quality generated by image style conversion, a high-quality image style conversion method based on generative confrontation network is proposed. This paper draws on the Cycle GAN network structure on the generation network, the method of combining the jump layer structure and the residual network in the U-Net network is used to increase the multi-scale invariance of the network; Secondly, in terms of discriminating network, a multi-scale dilated convolution discriminator is proposed to improve the spatial geometric transformation in image style conversion and generate high-resolution images.It has been verified that compared with Cycle GAN, the effect of this algorithm in image style conversion has been greatly improved, and the quality of image style conversion has been improved.
参考文献/References:
[1] Xiang Li,Yasushi Makihara,Chi Xu,et al.Gait recognition invariant to carried objects using alpha blending generative adversarial networks[J].Pattern Recognition,2020,105.
[2] Yanbiao Zou,Xianzhong Wei,Jiaxin Chen.Conditional generative adversarial network-based training image inpainting for laser vision seam tracking[J].Optics and Lasers in Engineering,2020,134.
[3] Robert Skilton,Yang Gao.Combining object detection with generative adversarial networks for in-component anomaly detection[J].Fusion Engineering and Design,2020,159.
[4] Gaffari Çelik,Muhammed Fatih Talu. Resizing and cleaning of histopathological images using generative adversarial networks[J]. Physica A: Statistical Mechanics and its Applications,2020,554.
[5] Xiaohuang Qin,Ziyang Wang,Jiepeng Yao,et al.Using a one-dimensional convolutional neural network with a conditional generative adversarial network to classify plant electrical signals[J]. Computers and Electronics in Agriculture,2020,174.
[6] Robertas,Scherer Rafal. HEMIGEN: Human Embryo Image Generator Based on Generative Adversarial Networks[J].Sensors(Basel,Switzerland),2019,19(16).
[7] Gao Hongmin,Yao Dan,Wang Mingxia,et al.A Hyperspectral Image Classification Method Based on Multi-Discriminator Generative Adversarial Networks[J].Sensors(Basel, Switzerland),2019,19(15).
[8] Yao Shanjian. Research on pedestrian recognition technology based on cycle Gan and twin network [D].Jinan:Shandong University of science and technology,2020.
[9] Deng Yongcang. Digital breast cancer recognition method based on convolution neural network[D].Quan zhou:Huaqiao University,2020.
备注/Memo
收稿日期:2021-07-09
基金项目:四川省科技计划资助项目(2020YFH0122)