基于michael jordanGAN的图像超分辨率方法研究
作者：王旺 徐俊武
来源：《软件导刊》2019翻译简历>持久化年第06期
canteen怎么读stacker 摘 要：图像超分辨率是一种采用软件算法提高图像空间分辨率的技术，由于传统超分辨率方法超清程度有限，基于深度学习的图像超分辨率方法成为研究者们近年来的研究重点。基于生成对抗网络（GAN）的图像超分辨率方法，在其网络模型结构基础上增加批处理归一化（BN）层以提升网络收敛速度、加强网络训练稳定性，更改上采样层网络并对损失函数作相应修改。在网络训练过程中，选择当前比较通用的Adam优化器。实验结果表
明，该方法具有很好的超分辨率图像重建能力，且在训练过程中训练稳定性得到提升，收敛速度也有所加快。
关键词：超分辨率;生成对抗网络;深度学习;图像重建
DOI：10. 11907/rjdk. 182450英语谜语带翻译
中图分类号：exclusive什么意思TP317.4
话语标记
文献标识码：A 文章编号：1672-7800（2019）006-0181-03爱尔兰留学费用
Abstract：Image super-resolution is a technology that us software algorithms to improve the spatial resolution of images. The traditional super-resolution methods can not avoid the limitation of blur or super-resolution. Image super-resolution method bad on deep learning has become the focus of rearchers. Generative adversarial networks （GAN） are the hottest deep neural networks in recent years. This paper explores the method bad on GAN. The super-resolution method which increas the u range of batch normalization and modifies the up-sampling method，国际夏令营 also modifies and optimizes
the loss function and its optimizer lection， optimizes the training instability of the image super-resolution method bad on GAN and greatly enhances the stability. The optimized method is still very good. Excellent images and super resolution capability verified the features in an experimental way.
