A remark on denoising of greyscale images using energy densities with varying growth rates (Q1701763)

In this paper, the authors study two variational problems, namely, \[ J[w]:=\int_\Omega F(\nabla w)dx+\frac{\lambda}{2}\int_{\Omega-D}| w-f|^2dx\to\min\text{ in }W^{1,1}(\Omega)\eqno{(1)} \] and \[ K[w]:=\int_\Omega F(\nabla w)dx+\int_{\Omega-D}\rho| w-f|^2dx\to\min\text{ in }W^{1,1}(\Omega),\eqno{(2)} \] where \(\Omega\) is a bounded Lipschitz region in \(\mathbb R^2\), the function \(f:\Omega\to\mathbb R\) represents the noisy data for which is assumed that \[ 0\leq f\leq 1\text{ a.e. on }\Omega,\eqno{(3)} \] \(\lambda>0\) is a parameter, \(F:\mathbb R^2\to[0,\infty)\) is a density function of class \(C^2\) satisfying \(F(0)=0\) and \(DF(0)=0,\rho:[0,\infty)\to[0,\infty)\) is a function of class \(C^1\) that is strictly increasing and strictly convex, and \(D\) is a measurable subset of \(\Omega\) such that \[ 0\leq\mathcal L^2(D)<\mathcal L^2(\Omega),\eqno{(4)} \] \(\mathcal L^2\) being the Lebesgue measure in \(\mathbb R^2\). Both problems are generalizations of the well-known variational problem proposed by \textit{L. I. Rudin} et al. [Physica D 60, No. 1--4, 259--268 (1992; Zbl 0780.49028)] as a model for the restoration of a noisy greyscale image \(f\). The idea of the authors is to use densities \(F\) that admit a certain flexibility of the growth rate, which means that the growth rate of \(F(\nabla w)\) can be prescribed in terms of \(|\nabla w|\). In this regard, it is assumed that there exist some constants \(c_5,c_ 6>0\) and exponents \[ p,\mu\in(1,\infty)\eqno{(5)} \] such that \[ c_5(1+|\xi|)^{-\mu}|\eta|^2\leq D^2F(\xi)(\eta,\eta)\leq c_6(1+|\xi|)^{p-2}|\eta|^2,\quad \xi,\eta\in\mathbb R^2.\eqno{(6)} \] The main results of this paper are the following two theorems. Theorem 1.2. Let (3), (5), and (6) hold together with (4). In addition, we assume that \(p,\mu<2\). Then, the variational problem (1) admits a unique solution \(u\). This solution additionally satisfies \(0\leq u\leq 1\) almost everywhere on \(\Omega\) as well as \(u\in W_{\mathrm{loc}}^{1,s}(\Omega)\) for any finite \(s\). Theorem 1.3. With \(\rho\) from above let \(f\), \(F\), and \(D\) satisfy (3), (6), and (4) respectively. Assume that \[ \limsup\limits_{t\to\infty}\frac{\rho(t)}{t^m}<\infty \] for some \(m\geq 1\). Moreover, let \(1<\mu<3/2\) and \(1<p<\mu\). Then, the variational problem (2) has a unique solution \(u\). We have \(0\leq u\leq 1\) almost everywhere on \(\Omega\); moreover, \(|\nabla u|\) is in \(L^s_{\mathrm{loc}}(\Omega)\) for any finite \(s\). If the density \(F\) is balanced in the sense that \[ \big| D^2F(\xi)\big||\xi|\leq c_{11}(F(\xi)+1),\quad \xi\in\mathbb R^2 \] for some constant, then the condition \(1<p<\mu\) can be replaced by the requirement \(p\in(1,2)\). Finally, some examples of densities \(F\) satisfying (6) are given.