Gauss-Lorenz hybrid prior super resolution reconstruction with mixed sparse representation

Ma Zijie; Zhao Xijun; Ren Guoqiang; Lei Tao; Yang Hu; Liu Dun

doi:10.12086/oee.2021.210299

Article navigation > Opto-Electronic Engineering > 2021 Vol. 48 > No. 11 > 210299

Next Article Previous Article

Ma Z J, Zhao X J, Ren G Q, et al. Gauss-Lorenz hybrid prior super resolution reconstruction with mixed sparse representation[J]. Opto-Electron Eng, 2021, 48(11): 210299. doi: 10.12086/oee.2021.210299

Citation:

Ma Z J, Zhao X J, Ren G Q, et al. Gauss-Lorenz hybrid prior super resolution reconstruction with mixed sparse representation[J]. Opto-Electron Eng, 2021, 48(11): 210299. doi: 10.12086/oee.2021.210299

Gauss-Lorenz hybrid prior super resolution reconstruction with mixed sparse representation

1.
Institute of Optics and Electronics, Chinese Academy of Sciences, Chengdu, Sichuan 610209, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China

Fund Project: National Key R&D Program of China (2016YFB0500200) and National Natural Science Foundation of China(61905254)

More Information

^*Corresponding author: Ren Guoqiang, E-mail: renguoqiang@ioe.ac.cn

Received Date 15 September 2021

Revised Date 26 November 2021

Published Date 30 November 2021

Abstract

Abstract

In order to obtain a super-resolution prior model with higher confidence and balance the reconstructed results between noise and details, this paper establishes a Gauss-Lorenz hybrid prior model based on the mixed sparse representation framework. This prior model's advantages and specific application schemes are studied. Firstly, according to the type of prior information, the advantages and problems of some traditional algorithms are introduced. Next, the statistical characteristics of different components of the image are modeled separately. Then, based on the analysis of the mixed sparse framework, the Gauss-Gibbs prior and the Lorenz prior, the super-resolution algorithm based on the Gauss-Lorenz hybrid prior under the group sparse framework is illustrated. Finally, the implementation and the final iteration scheme are introduced. The aim of noise suppression while maintaining details in the reconstruction process has been completed, which can be used for in more complex environments with super-resolution resconstruction.
- super-resolution algorithm /
- prior model /
- Gauss-Lorenz model /
- mixed sparse representation

FullText(HTML)

References

[1]	周春平, 宫辉力. 实用卫星遥感超分辨率理论及应用[M]. 北京: 科学出版社, 2016: 6. Google Scholar Zhou C P, Gong H L. Theory and Application of Superresolution for Practical Satellite Remote Sensing[M]. Beijing: Science Press, 2016: 6. Google Scholar
[2]	赵圆圆, 施圣贤. 融合多尺度特征的光场图像超分辨率方法[J]. 光电工程, 2020, 47(12): 54–64. doi: 10.12086/oee.2020.200007 CrossRef Google Scholar Zhao Y Y, Shi S X. Light-field image super-resolution based on multi-scale feature fusion[J]. Opto-Electron Eng, 2020, 47(12): 54–64. doi: 10.12086/oee.2020.200007 CrossRef Google Scholar
[3]	于淑侠, 胡良梅, 张旭东, 等. 彩色图像多尺度引导的深度图像超分辨率重建[J]. 光电工程, 2020, 47(4): 42–51. doi: 10.12086/oee.2020.190260 CrossRef Google Scholar Yu S X, Hu L M, Zhang X D, et al. Multi-scale guided depth image super-resolution reconstruction[J]. Opto-Electron Eng, 2020, 47(4): 42–51. doi: 10.12086/oee.2020.190260 CrossRef Google Scholar
[4]	朱福珍, 刘越, 黄鑫, 等. 改进的稀疏表示遥感图像超分辨重建[J]. 光学精密工程, 2019, 27(3): 718–725. Google Scholar Zhu F Z, Liu Y, Huang X, et al. Remote sensing image super-resolution based on improved sparse representation[J]. Opt Precis Eng, 2019, 27(3): 718–725. Google Scholar
[5]	Fu B, Li Y, Wang X H, et al. Image super-resolution using TV priori guided convolutional network[J]. Patt Recognit Lett, 2019, 125: 780–784. doi: 10.1016/j.patrec.2019.06.022 CrossRef Google Scholar
[6]	Xu T, Huang T Z, Deng L J, et al. Hyperspectral image superresolution using unidirectional total variation with tucker decomposition[J]. IEEE J Select Top Appl Earth Observat Remote Sens, 2020, 13: 4381–4398. doi: 10.1109/JSTARS.2020.3012566 CrossRef Google Scholar
[7]	Farsiu S, Robinson M D, Elad M, et al. Fast and robust multiframe super resolution[J]. IEEE Trans Image Process, 2004, 13(10): 1327–1344. doi: 10.1109/TIP.2004.834669 CrossRef Google Scholar
[8]	代少升, 崔俊杰, 张德洲. 基于自适应阈值HMRF的红外超分辨率重建[J]. 半导体光电, 2017, 38(4): 577–579, 613. Google Scholar Dai S S, Cui J J, Zhang D Z. Infrared super-resolution reconstruction based on adaptive threshold HMRF[J]. Semicond Optoelectr, 2017, 38(4): 577–579, 613. Google Scholar
[9]	Rezayi H, Seyedin S A. Huber Markov random field for joint super resolution[C]//2017 10th Iranian Conference on Machine Vision and Image Processing (MVIP), Isfahan, Iran, 2017: 93–98. Google Scholar
[10]	Shao Z F, Wang L, Wang Z Y, Deng J. Remote sensing image super-resolution using sparse representation and coupled sparse autoencoder[J]. IEEE J Select Top Appl Earth Observat Remote Sens, 2019, 12(8): 2663–2674. doi: 10.1109/JSTARS.2019.2925456 CrossRef Google Scholar
[11]	Li J, Zhou T, Wang W X, et al. Enhancement method for aviation image based on improved NSCT[C]//2020 Cross Strait Radio Science & Wireless Technology Conference (CSRSWTC), Fuzhou, China, 2020: 1–3. Google Scholar
[12]	Li F, Xin L, Guo Y, et al. A framework of mixed sparse representations for remote sensing images[J]. IEEE Trans Geosci Remote Sens, 2017, 55(2): 1210–1221. doi: 10.1109/TGRS.2016.2621123 CrossRef Google Scholar
[13]	鲍东星, 李晓明, 李金. 基于洛伦兹范数的医学图像超分辨率重建研究[J]. 计算机仿真, 2020, 37(4): 205–208. doi: 10.3969/j.issn.1006-9348.2020.04.043 CrossRef Google Scholar Bao D X, Li X M, Li J. Lorentzian norm based super-resolution reconstruction of medical image[J]. Comput Simul, 2020, 37(4): 205–208. doi: 10.3969/j.issn.1006-9348.2020.04.043 CrossRef Google Scholar
[14]	孙超. 基于混合方法的卫星遥感图像自动配准技术的研究[D]. 长春: 吉林大学, 2018. Google Scholar Sun C. The research on automatic registration technology of satellite remote sensing images based on hybrid methods[D]. Changchun: Jilin University, 2018. Google Scholar
[15]	Liu L, Huang W, Wang C, et al. SAR image super-resolution based on TV-regularization using gradient profile prior[C]//2016 CIE International Conference on Radar (RADAR), Guangzhou, China, 2016: 1–4. Google Scholar
[16]	李志明. 基于L1范数的全变分正则化超分辨重构算法[J]. 计算机工程与应用, 2016, 52(15): 212–216. doi: 10.3778/j.issn.1002-8331.1511-0286 CrossRef Google Scholar Li Z M. Super resolution reconstruction algorithm based on L1 norm of total variation regularization[J]. Comput Eng Appl, 2016, 52(15): 212–216. doi: 10.3778/j.issn.1002-8331.1511-0286 CrossRef Google Scholar
[17]	薛翠红, 于明, 杨宇皓, 等. 基于学习的马尔科夫超分辨率复原[J]. 吉林大学学报(工学版), 2013, 43(S1): 406–409. Google Scholar Xue C H, Yu M, Yang Y H, et al. MRF reconstruction based on the markov network[J]. J Jilin Univ (Eng Ed), 2013, 43(S1): 406–409. Google Scholar
[18]	Liu X H, Tanaka M, Okutomi M. Noise level estimation using weak textured patches of a single noisy image[C]//2012 19th IEEE International Conference on Image Processing, Orlando, FL, USA, 2013. Google Scholar

Overview

Overview

Overview: The optimization function of a super-resolution algorithm consists of two parts: a data consistency term that is dependent on input images and a prior term that is derived from a prior model. The data consistency term introduces non-redundant information between images, but it also exacerbates the noise and causes overfitting. Therefore, the selection of the prior model is crucial for optimizing the performance of the image reconstruction. In order to obtain a higher confidence super-resolution prior model and balance the reconstructed results between noise and details, this paper establishes a Gauss-Lorenz hybrid prior model based on the mixed sparse representation framework. This prior model's advantages and specific application scheme are studied. Firstly, according to the type of prior information, the advantages and problems of some traditional algorithms are introduced. Next, the statistical characteristics of different components of the image are modeled separately. Then, based on the analysis the mixed sparse framework, the Gauss-Gibbs prior and the Lorenz prior, the super-resolution algorithm based on the Gauss-Lorenz hybrid prior under the group sparse framework is illustrated. Finally, the implementation and the final iteration scheme are introduced. The aim of noise suppression while maintaining details in the reconstruction process has been completed, which can be used for more complex environments with super-resolution reconstruction requirements. This paper has three main innovations. 1) We use the group sparse framework as the basic framework of the prior model. In this paper, different components of the natural images are projected into spatial, wavelet, and curved domains respectively. The image is divided into three components: point, line, and surface. We get constraints that are closer to natural images and improve the prior confidence. 2) Line component, that is, the edge component of an image, is closer to Lorenz distribution in statistical derivation than Gaussian distribution, so Lorenz model is used to model the edge component in the curved domain; Gaussian-Gibbs model is used to model the point and area components, which can suppress noise. The three components are continuously and alternately optimized in iteration to achieve a balance. 3) The pixel residuals threshold in the convex set projection iteration method is used as the iteration termination condition to solve the problem that the optimization progress and the optimization step are not uniform in different domains. We use four detection methods: structural similarity, maximum absolute error, noise level detection and peak signal-to-noise ratio to evaluate the results of the algorithm. Whether it is the evaluation method with or without reference, the algorithm proposed in this paper obtains better evaluation results than other algorithms.