首页 > 论文 > 光学 精密工程 > 26卷 > 4期(pp:951-961)

超低照度下微光图像的深度卷积自编码网络复原

Deep convolutional autoencoder networks approach to low-light level image restoration under extreme low-light illumination

  • 摘要
  • 论文信息
  • 参考文献
  • 被引情况
  • PDF全文
分享:

摘要

微光/红外图像彩色融合是目前国内外夜视技术的重要发展方向, 在超低照度下(环境照度小于2×10-3 lux), 由于成像器件限制, 微光图像具有低信噪比、低对比度等特点, 导致目标难以辨识, 成为制约彩色夜视技术的关键。为了提高目标的探测和识别率, 提出了一种基于卷积自编码网络的微光图像复原方法, 利用卷积自编码网络从微光图像训练集中学习超低照度下微光图像特征, 实现去噪和对比度增强。实验结果表明, 本文提出的方法得到的峰值信噪比(Peak Signal to Noise Ratio, PSNR)较经典的BM3D算法平均提高1.67 dB, 结构相似度(Structural Similarity Index, SSIM)的值平均提高0.063, 均方根对比度的值(Root Mean Square Contrast, RMSC)平均提高0.19。对微光图像复原具有很好的效果, 能够有效地提高信噪比和对比度水平。

Abstract

LLL (Low-Light Level) / infrared image color fusion is an important development direction of night vision technology in the world. Under extreme low light level (environment illumination less than ), LLL image has low signal to noise ratio and low contrast features, and the target is difficult to identify as imaging device limitations, which has been a key constraint to color night vision technology. In order to improve target detection and recognition rate, LLL image restoration method based on the deep convolutional autoencoder network was proposed, which learning LLL image characteristics from the LLL image training set by using the convolutional autoencoder, and implementing de-noising and contrast enhancement. The experiment results show that, compared with the classical BM3D algorithm, the proposed method improves the peak signal to noise ratio (PSNR) with value of 1.67 dB and reduces the RMSE with value of 0.098, improves the structural similarity (SSIM) with the value of 0.063 and improves the root mean square contrast (RMSC) with value of 0.91. It has a very good effect on the restoration of low light level images, and improves the signal-to-noise ratio and contrast level effectively.

Newport宣传-MKS新实验室计划
补充资料

中图分类号:TN223

DOI:10.3788/ope.20182604.0951

所属栏目:信息科学

基金项目:军内重点科研项目(No.427210843)

收稿日期:2017-08-18

修改稿日期:2017-09-26

网络出版日期:--

作者单位    点击查看

刘超:海军工程大学 兵器工程系, 湖北 武汉 430033
张晓晖:海军工程大学 兵器工程系, 湖北 武汉 430033

联系人作者:刘超(generaladolph@163.com)

备注:刘超(1989-), 男, 陕西周至人, 博士研究生, 2012年于西安理工大学获得学士学位, 2014年于海军工程大学获得硕士学位, 主要从事机器视觉及彩色夜视方面的研究。E-mail: generaladolph@163.com

【1】曾桂林, 苏秉华, 牛丽红, 等. 基于神经网络的微光图像增强与复原[J]. 兵工学报, 2006, 27(4): 652-654.
ZENG G L, SU B H, NIU L H, et al.. Neural network-based low-light-level image enhancement and reconstruction[J]. Acta Armamentarii, 2006, 27(4): 652-654. (in Chinese)

【2】金伟其, 陶禹, 石峰, 等. 微光视频器件及其技术的进展[J]. 红外与激光工程, 2015, 44(11): 3167-3176.
JIN W Q, TAO Y, SHI F, et al.. Progress of low level light video technology[J]. Infrared and Laser Engineering, 2015, 44(11): 3167-3176. (in Chinese)

【3】鲁佳颖, 谷小婧, 顾幸生. 面向微光/红外融合彩色夜视的场景解析方法[J]. 红外与激光工程, 2017, 46(8): 0804002.
LU J Y, GU X J, GU X SH. Scene parsing method toward low-light-level/infrared color night vision[J]. Infrared and Laser Engineering, 2017, 46(8): 0804002. (in Chinese)

【4】ROUDOT P, KERVRANN C, BOULANGER J, et al.. Noise modeling for intensified camera in fluorescence imaging: Application to image denoising[C]. Proceedings of 2013 IEEE 10th International Symposium on Biomedical Imaging (ISBI), IEEE, 2013: 600-603.

【5】QUAN J. Image denoising of Gaussian and Poisson noise based on wavelet thresholding[D]. Cincinnati: University of Cincinnati, 2013.

【6】VINCENT P, LAROCHELLE H, BENGIO Y, et al.. Extracting and composing robust features with denoising autoencoders[C]. Proceedings of the 25th International Conference on Machine Learning, ACM, 2008: 1096-1103.

【7】XIE J Y, XU L L, CHEN E H. Image denoising and inpainting with deep neural networks[C]. Proceedings of the 25th International Conference on Neural Information Processing Systems, ACM, 2012: 341-349.

【8】BURGER H C, SCHULER C J, HARMELING S. Image denoising: Can plain neural networks compete with BM3D?[C]. Proceedings of 2012 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE, 2012: 2392-2399.

【9】AGOSTINELLI F, ANDERSON M R, LEE H. Adaptive multi-column deep neural networks with application to robust image denoising[C]. Advances in Neural Information Processing Systems, NIPS, 2013: 1493-1501.

【10】WU Y W, ZHAO H H, ZHANG L Q. Image denoising with rectified linear units[C]. Proceedings of the 21st International Conference on Neural Information Processing, Springer, 2014: 142-149.

【11】LORE K G, AKINTAYO A, SARKAR S. LLNet: A deep autoencoder approach to natural low-light image enhancement[J]. Pattern Recognition, 2017, 61: 650-662.

【12】DONG CH, DENG Y B, LOY C C, et al.. Compression artifacts reduction by a deep convolutional network[C]. Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV), IEEE, 2015: 576-584.

【13】MAO X J, SHEN CH H, YANG Y B. Image restoration using very deep convolutional encoder-decoder networks with symmetric skip connections[C]. Advances in Neural Information Processing Systems, NIPS, 2016.

【14】肖龙, 徐超, 刘广荣. 应用于可穿戴微光成像系统的嵌入式平台设计[J]. 红外与激光工程, 2016, 45(1): 0118006.
XIAO L, XU CH, LIU G R. Design of embedded platform applied in wearable low-light level imaging system[J]. Infrared and Laser Engineering, 2016, 45(1): 0118006. (in Chinese)

【15】DABOV K, FOI A, KATKOVNIK V, et al.. Image denoising by sparse 3-D transform-domain collaborative filtering[J]. IEEE Transactions on Image Processing, 2007, 16(8): 2080-2095.

引用该论文

LIU Chao,ZHANG Xiao-hui. Deep convolutional autoencoder networks approach to low-light level image restoration under extreme low-light illumination[J]. Optics and Precision Engineering, 2018, 26(4): 951-961

刘超,张晓晖. 超低照度下微光图像的深度卷积自编码网络复原[J]. 光学 精密工程, 2018, 26(4): 951-961

您的浏览器不支持PDF插件,请使用最新的(Chrome/Fire Fox等)浏览器.或者您还可以点击此处下载该论文PDF