An anisotropic bimodal energy for the segmentation of thin tubes and its approximation with \(\Gamma\)-convergence (Q2629558)

The author studies the problem of detection of thin structures--tubes in 2D and 3D images. He introduces a variational bimodal model represented by the energy functional \[ E(p,\mathbf{M})=\int_\Omega(p-g)^2\,dx+ \beta \int_{S_p}\langle\mathbf{M}\boldsymbol{\nu}_p,\boldsymbol{\nu}_p\rangle^{1/2}\,d\mathcal{H}^{n-1} +\gamma\|\mathbf{M}\|_{W^{1,r}(\Omega)} \] with the \((n-1)\)-dimensional Hausdorff measure, \(p:\Omega\to \{0,1\}\) and a normal unit vector \(\boldsymbol{\nu}_p\) to \(S_p\). The analysis consists in searching for a pair \((p,\mathbf{M})\) minimizing \(E(p,\mathbf{M})\). The functional is approximated by \[ E_\varepsilon(p,\mathbf{M})=\int_\Omega(p-g)^2\,dx+ \beta \int_\Omega\left(9\varepsilon\langle\mathbf{M}\nabla{p},\nabla{p}\rangle+\frac{p^2(1-p)^2}\varepsilon\right)\,dx +\gamma\|\mathbf{M}\|_{W^{1,r}(\Omega)}, \] where \(p:\Omega\to [0,1]\) is a regular function and \(\varepsilon\to 0^+\). The main result is the \(\Gamma\)-convergence of the family \((E_\varepsilon)\) to \(E\).