Exact minimizer for the couple (\(L^{\infty}, BV\)) and the one-dimensional analogue of the Rudin-Osher-Fatemi model (Q2431553)

This paper investigates a one-dimensional analogue of the Rudin-Osher-Fatemi (ROF) model. The starting point is the \(E\)-functional \[ E(t,f;L^\infty,BV):=\inf_{\|g\|_{L^\infty} \leq t} \|g-f\|_{BV}. \] Under the assumptions that \(g:[a,b] \to \mathbb{R}\) is continuous with oscillation bounded from below by \(2t\), the authors construct an algorithm leading to an exact minimizer for the \(E\)-functional. As a corollary, they obtain that the \(K\)-functional defined by \[ K(N,f;BV,L^\infty):=\inf_{g \in L^\infty} (\|f-g\|_{BV} + N\|g\|_{L^\infty}) \] is comparable with \[ \sup_{a \leq x_0 \leq \cdots \leq x_N \leq b} \sum_{i=1}^N |f(x_i) - f(x_{i+1})|. \] As an application to the one-dimensional ROF model, this algorithm also provides a \(BV\) approximation of initial data of a noisy signal, realizing Morozo's discrepancy principle: the error estimate is established in any Banach space in which the real interpolation space \((L^\infty,BV)_{\theta,1}\) is continuously embedded.