Optimal segmentation of unbounded functions (Q2563972): Difference between revisions

We denote by \(n\) an integer \(\geq 2\), by \(p\) and \(\lambda\) two real numbers with \(p\geq 1\) and \(\lambda>0\), by \(\Omega\) an open subset of the Euclidean \(n\)-dimensional space \(\mathbb{R}^n\), by \({\mathcal H}^{n-1}\) the \((n-1)\)-dimensional Hausdorff measure in \(\mathbb{R}^n\), and by \(g\) a measurable, real-valued function defined on \(\Omega\). The authors prove that the functional \[ F(C, u)=\lambda\int_{\Omega\backslash C}|u-g|^p dx+{\mathcal H}^{n-1}(C\cap\Omega) \] achieves its minimum on pairs \((C,u)\) with \(C\) closed and \(u\) constant on each connected component of \(\Omega\backslash C\). Moreover, they show that the family of connected components of \(\Omega\backslash K\) is locally finite in \(\Omega\), for any minimizer \((K, w)\) of \(F\). Variational problems of the preceding type are encountered for instance in computer vision theory, where a basic problem is to obtain ``optimal segmentations'' of a given image.