摘要
为更好提取图像特征,增强图像清晰度变得极为重要,本文采用一个反复对抗互相优化的网络模型,与改进的生成网络主体架构和损失函数以及多尺度递归网络相结合,在GoPro数据集上采用相应的峰值信噪比、结构相似度和基于边缘特征的图像清晰度评价方法。算例结果表明,该方法对图像清晰度增强有较好的效果,可一定程度地解决由运动带来的清晰度不足的问题。
In order to better extract image features and enhance image clarity has become extremely important,this article uses a network model that repeatedly confronts each other and uses an improved generation network main structure and loss function combined with a multi-scale recursive network,which is based on the GoPro data set.Using the corresponding peak signal-tonoise ratio,structural similarity and image sharpness evaluation method based on edge features.The results of calculation examples show that this method has a good effect on image sharpness enhancement,and can solve the problem of insufficient sharpness caused by motion to a certain extent.
作者
王洋
陈朝新
张月光
郭磊
沈鹏
Wang Yang;Chen Chaoxin;Zhang Yueguang;Guo Lei;Shen Peng(Henan Qice Electronics Technology Co.,Ltd,Zhengzhou 453000;Zhengzhou University,Zhengzhou 450001)
出处
《现代计算机》
2022年第9期64-68,共5页
Modern Computer
关键词
生成对抗网络
图像特征
图像清晰度
多尺度递归网络
generative confrontation network
image features
image sharpness
multi-scale recurrent network