Homogeneous Penalizers and Constraints in Convex Image Restoration

Abstract

Recently convex optimization models were successfully applied for solving various problems in image analysis and restoration. In this paper, we are interested in relations between convex constrained optimization problems of the form argmin{Φ(x) subject to Ψ(x) ≤ τ } and their penalized counterparts argmin{Φ(x) + λΨ(x)}. We recall general results on the topic by the help of an epigraphical projection. Then we deal with the special setting Ψ := L · with L ∈ R m,n and Φ := ϕ(H·), where H ∈ R n,n and ϕ : R n → R ∪ {+∞} meet certain requirements which are often fulfilled in image processing models. In this case we prove by incorporating the dual problems that there exists a bijective function such that the solutions of the constrained problem coincide with those of the penalized problem if and only if τ and λ are in the graph of this function. We illustrate the relation between τ and λ for various problems arising in image processing. In particular, we point out the relation to the Pareto frontier for joint sparsity problems. We demonstrate the performance of the constrained model in restoration tasks of images corrupted by Poisson noise with the I-divergence as data fitting term ϕ and in inpainting models with the constrained nuclear norm. Such models can be useful if we have a priori knowledge on the image rather than on the noise level.

Minimization and parameter estimation for seminorm regularization models with I -divergence constraints

“…For examples we refer to [25]. For finite, convex penalizers (which does not include D(·, b)), the existence of a Lagrange multiplier λ ≥ 0 is assured by [55 In the next section, will apply Theorem 2.1 with respect to the functions F := D(b, H·) and…”

Epigraphical projection and proximal tools for solving constrained convex optimization problems

“…the noise statistical properties. Note also that there exist some conceptual Lagrangian equivalences between regularized solutions to inverse problems and constrained ones, although some caution should be taken when the regularization functions are nonsmooth (see [36] where the case of a single regularization parameter is investigated).…”

Biomedical Image Reconstruction: From the Foundations to Deep Neural Networks

“…Generally, which form is chosen is a matter of personal preference. For a much deeper look at this issue, see [26].…”