ADMM-DIPTV: combining Total Variation and Deep Image Prior for image restoration

In the last decades, unsupervised deep learning based methods have caught researchers attention, since in many applications collecting a great amount of training examples is not always feasible. Moreover, the construction of a good training set is time consuming and hard because the selected data have to be enough representative for the task. In this paper, we mainly focus on the Deep Image Prior (DIP) framework powered by adding the Total Variation regularizer which promotes gradient-sparsity of the solution. Differently from other existing approaches, we solve the arising minimization problem by using the well known Alternating Direction Method of Multipliers (ADMM) framework, decoupling the contribution of the DIP $L_{2}$-norm and Total Variation terms. The promising performances of the proposed approach, in terms of PSNR and SSIM values, are addressed by means of experiments for different image restoration tasks on synthetic as well as on real data.

[1]  Qinghua Hu,et al.  Neural Blind Deconvolution Using Deep Priors , 2019, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).

[2]  M. Moscoso.  Introduction to Image Reconstruction , 2008 .

[3]  Maximilian Schmidt,et al.  Computed tomography reconstruction using deep image prior and learned reconstruction methods , 2020, Inverse Problems.

[4]  Yves Chauvin,et al.  Backpropagation: theory, architectures, and applications , 1995 .

[5]  Andrea Vedaldi,et al.  Deep Image Prior , 2017, International Journal of Computer Vision.

[6]  W. Clem Karl,et al.  3.6 – Regularization in Image Restoration and Reconstruction , 2005 .

[7]  Ulugbek Kamilov,et al.  Image Restoration Using Total Variation Regularized Deep Image Prior , 2018, ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[8]  Stamatios Lefkimmiatis,et al.  Non-local Color Image Denoising with Convolutional Neural Networks , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[9]  Zhi-Quan Luo,et al.  On the linear convergence of the alternating direction method of multipliers , 2012, Mathematical Programming.

[10]  Zhi-Quan Luo,et al.  Convergence analysis of alternating direction method of multipliers for a family of nonconvex problems , 2014, 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[11]  Wotao Yin,et al.  Global Convergence of ADMM in Nonconvex Nonsmooth Optimization , 2015, Journal of Scientific Computing.

[12]  Luca Calatroni,et al.  On the inverse Potts functional for single-image super-resolution problems , 2020, ArXiv.

[13]  Stefan Harmeling,et al.  Image denoising: Can plain neural networks compete with BM3D? , 2012, 2012 IEEE Conference on Computer Vision and Pattern Recognition.

[14]  L. Rudin,et al.  Nonlinear total variation based noise removal algorithms , 1992 .

[15]  Léon Bottou,et al.  Stochastic Gradient Descent Tricks , 2012, Neural Networks: Tricks of the Trade.

[16]  Djemel Ziou,et al.  Image Quality Metrics: PSNR vs. SSIM , 2010, 2010 20th International Conference on Pattern Recognition.

[17]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[18]  Tobias Kluth,et al.  Regularization by Architecture: A Deep Prior Approach for Inverse Problems , 2019, Journal of Mathematical Imaging and Vision.

[19]  J. S. Moll.  The anisotropic total variation flow , 2005 .