多尺度卷积神经网络的噪声模糊图像盲复原 | |
Alternative Title | Multi-scale Convolutional Neural Network of Blind deblurring of Noisy and Blurry Images |
刘鹏飞1,2,3,4; 赵怀慈1,2,4![]() | |
Department | 光电信息技术研究室 |
Source Publication | 红外与激光工程
![]() |
ISSN | 1007-2276 |
2019 | |
Volume | 48Issue:4Pages:1-9 |
Indexed By | EI ; CSCD |
EI Accession number | 20192807171724 |
CSCD ID | CSCD:6481309 |
Contribution Rank | 1 |
Funding Organization | 装备预研领域基金(No.61400010102) |
Keyword | 多尺度卷积神经网络 多重损失函数 生成式对抗网络 噪声模糊图像 |
Abstract | 图像盲复原是从一幅观测的模糊图像恢复出模糊核和清晰图像,传统盲去卷积算法采用简化模型估计模糊核,导致预测模糊核与真实值误差较大,最终复原结果不理想。针对此问题提出一种基于改进残差模块的多尺度卷积神经网络模型,采用端到端模式,输入模糊图像,输出即为复原图像,无需估计模糊核。为了打破Wasserstein GAN (WGAN)参数的限制,设计了稀疏自动编码器,提出了一种基于限制网络输入的改进WGAN,有利于参数朝着最优解方向迭代,避免网络陷入局部最优解,提高了网络收敛速度。设计了多重损失函数,融合了基于多尺度网络的感知损失和基于条件式生成对抗网络的对抗损失。实验结果表明:所提方法在定量和定性评价指标上优于已有的代表性方法,并且运行速度比相近算法快了4倍。 |
Other Abstract | The purpose of image blind deconvolution is to estimate the unknown blur kernel from an observed blurred image and recover the original sharp image. Conventional methods used simple models estimating blur kernel, meaning mistakes were inevitable between estimated blur kernel and the real one. It would cause the final deblurred image unpredictable. A multi-scale convolutional neural network was presented based on the novel residual network which was designed skillfully. And it restored sharp images in an end-to-end manner, the input was blurred image and the output was the restored image without estimating blur kernel. In order to break the restriction of weights of Wasserstein GAN, domain constraint layer was designed to the WGAN, it can restrict domain of input latent variables, which it would benefit for the parameters iterating towards the optimal solutions and accelerating convergence. A total loss function was designed including perception loss which was based on the multi-scale network and adversarial loss which was based on conditional GAN. Extensive experiments manifest the superiority of the proposed method over state-of-the-art techniques, both qualitatively and quantitatively. The method is 4 times faster than the closest competitor. |
Language | 中文 |
Citation statistics | |
Document Type | 期刊论文 |
Identifier | http://ir.sia.cn/handle/173321/24463 |
Collection | 光电信息技术研究室 |
Corresponding Author | 刘鹏飞 |
Affiliation | 1.中国科学院沈阳自动化研究所 2.中国科学院机器人与智能制造创新研究院 3.中国科学院大学 4.中国科学院光电信息处理重点实验室 |
Recommended Citation GB/T 7714 | 刘鹏飞,赵怀慈,曹飞道. 多尺度卷积神经网络的噪声模糊图像盲复原[J]. 红外与激光工程,2019,48(4):1-9. |
APA | 刘鹏飞,赵怀慈,&曹飞道.(2019).多尺度卷积神经网络的噪声模糊图像盲复原.红外与激光工程,48(4),1-9. |
MLA | 刘鹏飞,et al."多尺度卷积神经网络的噪声模糊图像盲复原".红外与激光工程 48.4(2019):1-9. |
Files in This Item: | ||||||
File Name/Size | DocType | Version | Access | License | ||
多尺度卷积神经网络的噪声模糊图像盲复原.(3255KB) | 期刊论文 | 作者接受稿 | 开放获取 | CC BY-NC-SA | View Application Full Text |
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment