摘要
提出了一种结合深度学习的图像降噪算法。采用尺度感知边缘保护滤波器对噪声图像进行多尺度分解,利用其尺度感知和边缘保持的特性对图像噪声等小结构信息进行移除,并保持边缘细节不变;将训练好的卷积神经网络模型用于学习图像细节信息,并用于指导被尺度感知边缘保护滤波器处理后的图像进行细节恢复。结果表明,本文降噪算法能够有效降噪,并保持较好的高频信息,融合结果更利于人类视觉观察。
This study proposed an image denoising algorithm based on deep learning. The scale-perception edge-protection filter was used to decompose the noise image in multiple scales. Small features, such as the image noise, were removed via scale sensing and edge preserving, and the edge details were kept unchanged. A trained convolutional neural network model was used to gather detailed information about the image, and the image was then processed using the scale-perception edge-protection filter for detail recovery. The results show that the proposed denoising algorithm can effectively reduce noises and well retain high-frequency information. Moreover, the fusion results correlate well with human visual observations.
作者
陈欢
陈清江
Chen Huan;Chen Qingjiang(Department of Fundamentals, Shaanxi Institute of International Trade & Commerce,Xianyang, Shaanxi 712046, China;School of Science, Xian University of Architecture and Technology, Xi'an , Shaanxi 710055, China)
出处
《激光与光电子学进展》
CSCD
北大核心
2019年第9期101-107,共7页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61403298)
陕西省自然科学基金(2015JM1024)
关键词
图像处理
图像识别
尺度感知
边缘保护
图像降噪
卷积神经网络
细节恢复
image processing
image recognition
scale-perception
edge-protection
image denosing
convolutional neural network: detail recovery