Bilevel training schemes in imaging for total-variation-type functionals with convex integrands

In the context of image processing, given a k-th order, homogeneous and linear differential operator with constant coefficients, we study a class of variational problems whose regularizing terms depend on the operator. Precisely, the regularizers are integrals of spatially inhomogeneous integrands with convex dependence on the differential operator applied to the image function. The setting is made rigorous by means of the theory of Radon measures and of suitable function spaces modeled on BV. We prove the lower semicontinuity of the functionals at stake and existence of minimizers for the corresponding variational problems. Then, we embed the latter into a bilevel scheme in order to automatically compute the space-dependent regularization parameters, thus allowing for good flexibility and preservation of details in the reconstructed image. We establish existence of optima for the scheme and we finally substantiate its feasibility by numerical examples in image denoising. The cases that we treat are Huber versions of the first and second order total variation with both the Huber and the regularization parameter being spatially dependent. Notably the spatially dependent version of second order total variation produces high quality reconstructions when compared to regularizations of similar type, and the introduction of the spatially dependent Huber parameter leads to a further enhancement of the image details.

[1]  L. Vese.  A Study in the BV Space of a Denoising—Deblurring Variational Problem , 2001 .

[2]  Michael Hintermüller,et al.  An Infeasible Primal-Dual Algorithm for Total Bounded Variation-Based Inf-Convolution-Type Image Restoration , 2006, SIAM J. Sci. Comput..

[3]  Kristian Bredies,et al.  Total Generalized Variation in Diffusion Tensor Imaging , 2013, SIAM J. Imaging Sci..

[4]  Carola-Bibiane Schönlieb,et al.  A Combined First and Second Order Variational Approach for Image Reconstruction , 2012, Journal of Mathematical Imaging and Vision.

[5]  D. Breit,et al.  On the trace operator for functions of bounded 𝔸-variation , 2017, 1707.06804.

[6]  Karl Kunisch,et al.  Total Generalized Variation , 2010, SIAM J. Imaging Sci..

[7]  Michael Hintermüller,et al.  Optimal Selection of the Regularization Function in a Weighted Total Variation Model. Part I: Modelling and Theory , 2017, Journal of Mathematical Imaging and Vision.

[8]  Kristian Bredies,et al.  Inverse problems with second-order Total Generalized Variation constraints , 2020, ArXiv.

[9]  I. Fonseca,et al.  On lower semicontinuity and relaxation , 2001, Proceedings of the Royal Society of Edinburgh: Section A Mathematics.

[10]  Kennan T. Smith.  Formulas to represent functions by their derivatives , 1970 .

[11]  Konstantinos Papafitsoros,et al.  Novel higher order regularisation methods for image reconstruction , 2015 .

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

[13]  Carola-Bibiane Schönlieb,et al.  Bilevel approaches for learning of variational imaging models , 2015, ArXiv.

[14]  C. Schönlieb,et al.  Image denoising: Learning the noise model via nonsmooth PDE-constrained optimization , 2013 .

[15]  K. Bredies,et al.  Regularization of linear inverse problems with total generalized variation , 2014 .

[16]  Christos Sakaridisa,et al.  Processing, Analyzing and Learning of Images, Shapes, and Forms: Part 2 , 2019, Handbook of Numerical Analysis.

[17]  Karl Kunisch,et al.  A Bilevel Optimization Approach for Parameter Learning in Variational Models , 2013, SIAM J. Imaging Sci..

[18]  M. Bergounioux,et al.  A Second-Order Model for Image Denoising , 2010 .

[19]  Antonin Chambolle,et al.  A First-Order Primal-Dual Algorithm for Convex Problems with Applications to Imaging , 2011, Journal of Mathematical Imaging and Vision.

[20]  Carola-Bibiane Schönlieb,et al.  The structure of optimal parameters for image restoration problems , 2015, ArXiv.

[21]  P. Lions,et al.  Image recovery via total variation minimization and related problems , 1997 .

[22]  Martin Burger,et al.  Infimal convolution regularisation functionals of BV and $\mathrm{L}^{p}$ spaces. Part I: The finite $p$ case , 2015, 1504.01956.

[23]  Michael Hintermüller,et al.  Optimal Selection of the Regularization Function in a Weighted Total Variation Model. Part II: Algorithm, Its Analysis and Numerical Tests , 2016, Journal of Mathematical Imaging and Vision.

[24]  John F. Canny,et al.  A Computational Approach to Edge Detection , 1986, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[25]  Prashant A. Athavale,et al.  Weighted TV minimization and applications to vortex density models , 2015, 1509.03713.

[26]  Kristian Bredies,et al.  Recovering Piecewise Smooth Multichannel Images by Minimization of Convex Functionals with Total Generalized Variation Penalty , 2011, Efficient Algorithms for Global Optimization Methods in Computer Vision.

[27]  Carola-Bibiane Schönlieb,et al.  Bilevel Parameter Learning for Higher-Order Total Variation Regularisation Models , 2015, Journal of Mathematical Imaging and Vision.

[28]  L. Ambrosio,et al.  Functions of Bounded Variation and Free Discontinuity Problems , 2000 .

[29]  F. Gmeineder,et al.  Embeddings for A-weakly differentiable functions on domains , 2017, Journal of Functional Analysis.

[30]  Irene Fonseca,et al.  Adaptive Image Processing: First Order PDE Constraint Regularizers and a Bilevel Training Scheme , 2019, Journal of Nonlinear Science.