Abstract:We consider the restoration of piecewise constant images where the number of the regions and their values are not fixed in advance, with a good difference of piecewise constant values between neighboring regions, from noisy data obtained at the output of a linear operator (e.g., a blurring kernel or a Radon transform). Thus we also address the generic problem of unsupervised segmentation in the context of linear inverse problems. The segmentation and the restoration tasks are solved jointly by minimizing an objective function (an energy) composed of a quadratic data-fidelity term and a nonsmooth nonconvex regularization term. The pertinence of such an energy is ensured by the analytical properties of its minimizers. However, its practical interest used to be limited by the difficulty of the computational stage which requires a nonsmooth nonconvex minimization. Indeed, the existing methods are unsatisfactory since they (implicitly or explicitly) involve a smooth approximation of the regularization term and often get stuck in shallow local minima. The goal of this paper is to design a method that efficiently handles the nonsmooth nonconvex minimization. More precisely, we propose a continuation method where one tracks the minimizers along a sequence of approximate nonsmooth energies $\{J_\eps\}$, the first of which being strictly convex and the last one the original energy to minimize. Knowing the importance of the nonsmoothness of the regularization term for the segmentation task, each $J_\eps$ is nonsmooth and is expressed as the sum of an $\ell_1$ regularization term and a smooth nonconvex function. Furthermore, the local minimization of each $J_{\eps}$ is reformulated as the minimization of a smooth function subject to a set of linear constraints. The latter problem is solved by the modified primal-dual interior point method, which guarantees the descent direction at each step. Experimental results are presented and show the effectiveness and the efficiency of the proposed method. Comparison with simulated annealing methods further shows the advantage of our method.

On Nonconvex and Nonsmooth Regularization for Contrast-Preserving Variational Signal and Image Restoration

Discrete Total Variation Model with Gradient Fidelity Term for Image Restoration

An Efficient Variational Method for Image Restoration

Guided Regularization and its Application for Image Restoration

A Second Order Non-Smooth Variational Model for Restoring Manifold-Valued Images

Linearly Constrained Non-Lipschitz Optimization for Image Restoration

Infimal Convolution Type Regularization of TGV and Shearlet Transform for Image Restoration

A nonconvex regularization model and the ADMM based algorithm

An Overview of Image Restoration Based on Variational Regularization

Convex and non-convex adaptive TV regularizations for color image restoration

Efficient Reconstruction of Piecewise Constant Images Using Nonsmooth Nonconvex Minimization

Image restoration based on the minimax-concave and the overlapping group sparsity

Learning Weakly Convex Regularizers for Convergent Image-Reconstruction Algorithms

Several Nonconvex Mo dels for Image Deblurring under Poisson Noise

Nonlocal Adaptive Biharmonic Regularizer for Image Restoration.

Generalized nonconvex nonsmooth four-directional total variation with overlapping group sparsity for image restoration

Image restoration via the adaptive T V p regularization

Edge-preserving Reconstruction with Contour-Line Smoothing and Non-Quadratic Data-Fidelity

A Nonconvex Model with Minimax Concave Penalty for Image Restoration

High-order Total Variation-Based Poissonian Image Deconvolution with Spatially Adapted Regularization Parameter