Abstract:The paper addresses the generalization of the half-quadratic minimization method for the restoration of images having values in a complete Riemannian manifold. We recall the half-quadratic minimization method using the notation of the c-transform and adapt the algorithm to our special variational setting. We prove the convergence of the method for Hadamard spaces. Extensive numerical examples for images with values on spheres, in the rotation group SO(3) and in the manifold of positive definite matrices demonstrate the excellent performance of the algorithm. In particular, the method with SO(3)-valued data shows promising results for the restoration of images obtained from Electron Backscattered Diffraction which are of interest in material science.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to generalize the half - quadratic minimization method to the inpainting problem of Riemannian manifold - valued images**. Specifically, this paper focuses on the inpainting problem of images taking values on complete and connected Riemannian manifolds. The authors review the half - quadratic minimization method and adapt their algorithm to a special variational setting using the notation of c - transform. They prove the convergence of this method in Hadamard spaces and demonstrate the excellent performance of this algorithm in processing images on spheres, the rotation group SO(3), and the manifold of positive definite matrices through numerical experiments. In particular, for the inpainting of SO(3) - valued data, this method shows promising results in the inpainting of electron back - scatter diffraction images in materials science. ### Key problems and methods 1. **Problem background**: - The traditional half - quadratic minimization method is mainly used for denoising or inpainting real - valued images. - For Riemannian manifold - valued images (such as spheres, rotation groups, positive definite matrices, etc.), a new method is required for inpainting. 2. **Method introduction**: - **Energy functionals**: The paper defines two energy functionals \(J_1(u)\) and \(J_2(u)\), corresponding to anisotropic and isotropic inpainting models respectively. These functionals include a data fidelity term and a regularization term. \[ J_1(u):=\frac{1}{2}\sum_{i\in V}d^{2}(u_i,f_i)+\lambda\sum_{i\in G}\sum_{j\in N(i)^+}\phi(d(u_i,u_j)) \] \[ J_2(u):=\frac{1}{2}\sum_{i\in V}d^{2}(u_i,f_i)+\lambda\sum_{i\in G}\phi\left(\left(\sum_{j\in N(i)^+}d^{2}(u_i,u_j)\right)^{1/2}\right) \] where \(d(\cdot,\cdot)\) is the geodesic distance on the Riemannian manifold, \(\phi:\mathbb{R}_{\geq0}\to\mathbb{R}_{\geq0}\) is a regularization function, and \(\lambda > 0\) is a regularization parameter. 3. **Half - quadratic minimization**: - By introducing an auxiliary variable \(v\), the original problem is transformed into an augmented problem, making the new problem quadratic with respect to the image and separable with respect to the auxiliary variable. - An alternating minimization process is used to solve the augmented problem, allowing for efficient computation at each step. 4. **Convergence**: - The convergence of this method in Hadamard spaces is proved, especially for Hadamard manifolds such as the manifold of positive definite matrices. 5. **Numerical experiments**: - The inpainting effects of this method on different manifolds are verified through a large number of numerical experiments, and it performs particularly well in the inpainting of SO(3) - valued images. ### Summary The main contribution of this paper is the successful generalization of the half - quadratic minimization method to the inpainting problem of Riemannian manifold - valued images, and the proof of its convergence and effectiveness on specific manifolds. This method provides a powerful tool for processing complex manifold - valued images and has potential application value in fields such as materials science.

Restoration of Manifold-Valued Images by Half-Quadratic Minimization

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

Image distance measurement based on manifold learning and manifold high-order approximation

Variational Stereo Matching with Left Right Consistency Constraint

Quotient Set-based Nonlinear Manifold for Image Restoration

Mumford-Shah and Potts Regularization for Manifold-Valued Data with Applications to DTI and Q-Ball Imaging

High-quality image restoration from partial random samples in spatial domain.

Denoising of Sphere- and SO(3)-Valued Data by Relaxed Tikhonov Regularization

Monotone vector fields and the proximal point algorithm on Hadamard manifolds

Fast roughness minimizing image restoration under mixed Poisson-Gaussian noise

A Theoretically Guaranteed Quaternion Weighted Schatten p-norm Minimization Method for Color Image Restoration

Image Restoration by Combined Order Regularization with Optimal Spatial Adaptation

Poissonian Image Restoration Via the -Based Minimization

Convergence of the Time Discrete Metamorphosis Model on Hadamard Manifolds

Quaternion weighted Schatten p -norm minimization for color image restoration with convergence guarantee

Variational Method on Riemann Surfaces Using Conformal Parameterization and Its Applications to Image Processing

Scalable low dimensional manifold model in the reconstruction of noisy and incomplete hyperspectral images

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

Manifold Quadratic Penalty Alternating Minimization for Sparse Principal Component Analysis

Probabilistic Learning Vector Quantization on Manifold of Symmetric Positive Definite Matrices.