针对图像盲去模糊的可微分神经网络架构搜索方法

doi:10.19678/j.issn.1000-3428.0058575

摘要/Abstract

摘要： 为了解决设计图像去模糊神经网络依赖大量手工调参的问题，提出一种面向图像盲去模糊的可微分神经网络架构搜索方法。通过设计U型残差搜索空间，将去模糊网络的搜索过程分为9个搜索单元的搜索过程，降低了搜索的复杂度，并设计出一个基于随机游走和最近邻插值的算法，通过模拟相机运动轨迹的方式生成模糊核，进而生成足够的模糊图像用于训练。实验结果表明，该方法明显减少了人工调参的工作量，在GOPRO和Kohler数据集上搜索得到的网络，峰值信噪比相对于基准网络UNet分别提升3.10 dB和1.17 dB，并接近UNet的推理速度。

关键词: 卷积神经网络, 可微分神经网络架构搜索, 图像去模糊, 图像复原, 数据扩增

Abstract: The design of neural networks for image deblurring requires heavy work of manual parameter tuning. To address the problem, a differentiable neural network search method for image deblurring is proposed. By designing a U-shaped residual search space, the search of image deblurring networks is simplified into the search of 9 search units. Besides, this paper presents an algorithm based on random walk and nearest-neighbor interpolation. The algorithm can generate blur kernels by simulating the camera motion trajectory, and thus provide enough blur images for training. The experimental results show that the proposed method significantly reduces the workload of manual tuning. The PSNR of the network obtained by the proposed method is 3.10 dB higher than that of UNet on the GOPRO data set and 1.17 dB on the Kohler dataset, while the inference speed of the network is close to that of UNet.

Key words: convolutional neural network, differentiable neural architecture search, image deblurring, image restoration, data augmentation

中图分类号:

TP18

缪斯, 祝永新. 针对图像盲去模糊的可微分神经网络架构搜索方法[J]. 计算机工程, 2021, 47(9): 313-320.

MIAO Si, ZHU Yongxin. Differentiable Neural Architecture Search Method for Blind Image Deblurring[J]. Computer Engineering, 2021, 47(9): 313-320.

https://www.ecice06.com/CN/Y2021/V47/I9/313

图/表 15

20210917193201

20210917193204

20210917193208

20210917193212

20210917193216

20210917193220

20210917193224

20210917193228

20210917193237

20210917193241

20210917193245

20210917193248

20210917193252

20210917193256

20210917193300

参考文献

[1] XU L, ZHENG S, JIA J.Unnatural l0 sparse representation for natural image deblurring[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2013:1107-1114.
[2] PAN J, SU Z.Fast l0-regularized kernel estimation for robust motion deblurring[J].IEEE Signal Processing Letters, 2013, 20(9):841-844.
[3] HE K, SUN J, TANG X.Single image haze removal using dark channel prior[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(12):2341-2353.
[4] PAN J, SUN D, PFISTER H, et al.Deblurring images via dark channel prior[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40(10):2315-2328.
[5] 王俊芝, 玉振明.基于LMS自适应算法的图像去模糊研究[J].计算机工程, 2012, 38(17):226-231. WANG J Z, YU Z M.Research on image debluring based on adaptive least mean square algorithm[J].Computer Engineering, 2012, 38(17):226-231.(in Chinese)
[6] SCHULER C J, HIRSCH M, HARMELING S, et al.Learning to deblur[J].IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 38(7):1439-1451.
[7] HRADIŠ M, KOTERA J, ZEMCIK P, et al.Convolutional neural networks for direct text deblurring[EB/OL].[2020-05-11].http://www.bmva.org/bmvc/2015/papers/paper006/paper006.pdf.
[8] NAH S, HYUN K T, MU L K.Deep multi-scale convolutional neural network for dynamic scene deblurring[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:3883-3891.
[9] TAO X, GAO H, SHEN X, et al.Scale-recurrent network for deep image deblurring[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:8174-8182.
[10] HOCHREITER S, SCHMIDHUBER J.Long short-term memory[J].Neural Computation, 1997, 9(8):1735-1780.
[11] KUPYN O, BUDZAN V, MYKHAILYCH M, et al.Deblurgan:blind motion deblurring using conditional adversarial networks[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2018:8183-8192.
[12] GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al.Generative adversarial nets[C]//Proceedings of Advances in Neural Information Processing Systems.Montreal, Canada:MIT Press, 2014:2672-2680.
[13] ZOPH B, LE Q V.Neural architecture search with reinforcement learning[EB/OL].[2020-05-10].https://arxiv.org/pdf/1611.01578.pdf.
[14] LIU H, SIMONYAN K, YANG Y.Darts:differentiable architecture search[EB/OL].[2020-05-11].https://arxiv.org/pdf/1806.09055.pdf.
[15] HE K, ZHANG X, REN S, et al.Deep residual learning for image recognition[C]//Proceedings of IEEE Cconference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2016:770-778.
[16] PENG W, HONG X, ZHAO G.Video action recognition via neural architecture searching[C]//Proceedings of IEEE International Conference on Image Processing.Washington D.C., USA:IEEE Press, 2019:11-15.
[17] WENG Y, ZHOU T, LIU L, et al.Automatic convolutional neural architecture search for image classification under different scenes[J].IEEE Access, 2019, 7:38495-38506.
[18] HUNDT A, JAIN V, HAGER G D.sharpDARTS:faster and more accurate differentiable architecture search[EB/OL].[2020-05-15].https://arxiv.org/pdf/1903.09900.pdf.
[19] WU B, DAI X, ZHANG P, et al.Fbnet:hardware-aware efficient convnet design via differentiable neural architecture search[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:10734-10742.
[20] SIMONYAN K, ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[EB/OL].[2020-05-16].https://arxiv.org/pdf/1409.1556.pdf.
[21] RONNEBERGER O, FISCHER P, BROX T.U-net:convolutional networks for biomedical image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention.Munich, Germany:Springer, 2015:234-241.
[22] KINGMA D P, BA J.Adam:a method for stochastic optimization[EB/OL].[2020-05-16].https://arxiv.org/pdf/1412.6980.pdf.
[23] CHEN X, FANG H, LIN T Y, et al.Microsoft coco captions:data collection and evaluation server[EB/OL].[2020-05-17].https://arxiv.org/pdf/1504.00325.pdf.
[24] KÖHLER R, HIRSCH M, MOHLER B, et al.Recording and playback of camera shake:benchmarking blind deconvolution with a real-world database[C]//Proceedings of European Conference on Computer Vision.Florence, Italy:Springer, 2012:27-40.
[25] WANG Z, BOVIK A C, SHEIKH H R, et al.Image quality assessment:from error visibility to structural similarity[J].IEEE Transactions on Image Processing, 2004, 13(4):600-612.
[26] LEDIG C, THEIS L, HUSZÁR F, et al.Photo-realistic single image super-resolution using a generative adversarial network[C]//Proceedings of IEEE Conference on Computer Vision And Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:4681-4690.
[27] SUN J, CAO W, XU Z, et al.Learning a convolutional neural network for non-uniform motion blur removal[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2015:769-777.
[28] ZHANG H, DAI Y, LI H, et al.Deep stacked hierarchical multi-patch network for image deblurring[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:5978-5986.
[29] LIU S, PAN J, YANG M H.Learning recursive filters for low-level vision via a hybrid neural network[C]//European Conference on Computer Vision.Amsterdam, Netherlands:Springer, 2016:560-576.

选择文件类型/文献管理软件名称

选择包含的内容