# American Institute of Mathematical Sciences

February  2017, 11(1): 1-23. doi: 10.3934/ipi.2017001

## Augmented Lagrangian method for an Euler's elastica based segmentation model that promotes convex contours

 1 Norwegian Defence Research Establishment, P.O. Box 25, NO-2027 Kjeller, Norway 2 Department of Mathematics, University of Bergen, P.O. Box 7803, NO-5020 Bergen, Norway 3 Department of Mathematics, University of Alabama, Box 870350, Tuscaloosa, AL 35487, USA

* Corresponding author: Wei Zhu

Received  September 2015 Revised  April 2016 Published  January 2017

Fund Project: The first author is supported by the Norwegian Research Council eVita project 214889.

In this paper, we propose an image segmentation model where an $L^1$ variant of the Euler's elastica energy is used as boundary regularization. An interesting feature of this model lies in its preference for convex segmentation contours. However, due to the high order and non-differentiability of Euler's elastica energy, it is nontrivial to minimize the associated functional. As in recent work on the ordinary $L^2$-Euler's elastica model in imaging, we propose using an augmented Lagrangian method to tackle the minimization problem. Specifically, we design a novel augmented Lagrangian functional that deals with the mean curvature term differently than in previous works. The new treatment reduces the number of Lagrange multipliers employed, and more importantly, it helps represent the curvature more effectively and faithfully. Numerical experiments validate the efficiency of the proposed augmented Lagrangian method and also demonstrate new features of this particular segmentation model, such as shape driven and data driven properties.

Citation: Egil Bae, Xue-Cheng Tai, Wei Zhu. Augmented Lagrangian method for an Euler's elastica based segmentation model that promotes convex contours. Inverse Problems & Imaging, 2017, 11 (1) : 1-23. doi: 10.3934/ipi.2017001
##### References:

show all references

##### References:
Two vectors $\nu$, $\mathbf{a}$ and a possible unit vector $\mathbf{b}$ that maximizes the function $\psi(\mathbf{b})$.
The original image (the first row) and the segmentation results (the second row) with different curvature parameter $b$. Two features can be observed from these results: 1) data driven property: as the parameter $b$ increases, objects of relatively small size will be omitted in the final segmentation; 2) shape driven property: with the same parameter $b$, among those objects with equal areas, the one without a convex shape is taken out from the final segmentation (see the staircase-like shape). In this experiment, the other parameters are: $a=10^{-4},r_{1}=50,r_{2}=20,r_{3}=5$
The original image and the segmentation result. The result demonstrates that the invisible part of the hat can be restored by the proposed segmentation model with a relatively large curvature parameter $b$. In this experiment, the parameters are $a=10^{-3},b=80, r_{1}=60,r_{2}=40,r_{3}=10$
The original image and the segmentation result. The result shows that only the ''cap" part of the mushroom is captured while the ''stem" is omitted in order to form a convex segmentation contour by the proposed segmentation model with a relatively large curvature parameter $b$. In this experiment, the parameters are $a=10^{-4},b=18, r_{1}=20,r_{2}=10,r_{3}=5$
The effect of the curvature parameter $b$ on the final segmentation results. One may observe that the smaller the parameter $b$ is chosen, the more salient the indentation of the hat would be. In these two experiments, all the other parameters are the same as those in Fig. 3
The effect of the curvature parameter $b$ on the final segmentation results. One can see that once the parameter $b$ is chosen smaller, more part of the stem will be allowed in the final segmentation. In these two experiments, all the other parameters are the same as those in Fig. 4
The plots of relative residuals (Eq. 43), relative errors in Lagrange multipliers (Eq. 44), and relative error in $u^{k}$ (Eq. 45) for the two examples ''hat" (Left column) and ''mushroom" (Right column). These plots demonstrate the convergence of the iteration process
Augmented Lagrangian method for the $ECV$-$L^{1}$ segmentation model
 1. Initialization: $\phi^{0}$, $q^{0}$, $\mathbf{p}^{0}$, $\mathbf{n}^{0}$, and $\mathbf{\lambda}_{1}^{0}$, $\lambda_{2}^{0}$, $\mathbf{\lambda}_{3}^{0}$. For $k \geq 1$, do the following steps (Step $2\sim 4$): 2. Compute an approximate minimizer $(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k})$ of the augmented Lagrangian functional with the fixed Lagrangian multiplier $\mathbf{\lambda}_{1}^{k-1}$, $\lambda_{2}^{k-1}$, $\mathbf{\lambda}_{3}^{k-1}$: $(32)\;\;\;\;\;\;\;\;\;\;(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k}) \approx \mbox{argmin } \mathcal{L}(\phi,q,\mathbf{p},\mathbf{n}, \mathbf{\lambda}_{1}^{k-1}, \lambda_{2}^{k-1}, \mathbf{\lambda}_{3}^{k-1}).$ 3. Update the Lagrangian multipliers $\mathbf{\lambda}_{1}^{k} = \mathbf{\lambda}_{1}^{k-1}+r_{1}(\mathbf{p}^{k}-\nabla \phi^{k})$ $\lambda_{2}^{k} = \lambda_{2}^{k-1}+r_{2}(q^{k}-\nabla\cdot \mathbf{n}^{k})$ $\mathbf{\lambda}_{3}^{k} = \mathbf{\lambda}_{3}^{k-1}+r_{3}(|\mathbf{p}|\mathbf{n}^{k}-\mathbf{p}^{k}),$ 4. Measure the relative residuals and stop the iteration if they are smaller than a threshold $\epsilon_{r}$.
 1. Initialization: $\phi^{0}$, $q^{0}$, $\mathbf{p}^{0}$, $\mathbf{n}^{0}$, and $\mathbf{\lambda}_{1}^{0}$, $\lambda_{2}^{0}$, $\mathbf{\lambda}_{3}^{0}$. For $k \geq 1$, do the following steps (Step $2\sim 4$): 2. Compute an approximate minimizer $(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k})$ of the augmented Lagrangian functional with the fixed Lagrangian multiplier $\mathbf{\lambda}_{1}^{k-1}$, $\lambda_{2}^{k-1}$, $\mathbf{\lambda}_{3}^{k-1}$: $(32)\;\;\;\;\;\;\;\;\;\;(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k}) \approx \mbox{argmin } \mathcal{L}(\phi,q,\mathbf{p},\mathbf{n}, \mathbf{\lambda}_{1}^{k-1}, \lambda_{2}^{k-1}, \mathbf{\lambda}_{3}^{k-1}).$ 3. Update the Lagrangian multipliers $\mathbf{\lambda}_{1}^{k} = \mathbf{\lambda}_{1}^{k-1}+r_{1}(\mathbf{p}^{k}-\nabla \phi^{k})$ $\lambda_{2}^{k} = \lambda_{2}^{k-1}+r_{2}(q^{k}-\nabla\cdot \mathbf{n}^{k})$ $\mathbf{\lambda}_{3}^{k} = \mathbf{\lambda}_{3}^{k-1}+r_{3}(|\mathbf{p}|\mathbf{n}^{k}-\mathbf{p}^{k}),$ 4. Measure the relative residuals and stop the iteration if they are smaller than a threshold $\epsilon_{r}$.
Alternating minimization method for solving the subproblems.
 1. Initialization: $\widetilde{\phi}^{0}=\phi^{k-1}$, $\widetilde{q}^{0}=q^{k-1}$, $\widetilde{\mathbf{p}}^{0}=\mathbf{p}^{k-1}$, and $\widetilde{\mathbf{n}}^{0}=\mathbf{n}^{k-1}$. 2. For fixed Lagrangian multiplier $\mathbf{\lambda}_{1}=\mathbf{\lambda}_{1}^{k-1}$, $\lambda_{2}=\lambda_{2}^{k-1}$, and $\mathbf{\lambda}_{3}=\mathbf{\lambda}_{3}^{k-1}$, solve the following subproblems : $(33)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\phi}^{1} = \mbox{argmin } \mathcal{L}(\phi,\widetilde{q}^{0},\widetilde{\mathbf{p}}^{0},\widetilde{\mathbf{n}}^{0}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(34)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{q}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},q,\widetilde{\mathbf{p}}^{0},\widetilde{\mathbf{n}}^{0}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(35)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\mathbf{p}}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},\widetilde{q}^{1},\mathbf{p},\widetilde{\mathbf{n}}^{0}, , \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(36)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\mathbf{n}}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},\widetilde{q}^{1},\widetilde{\mathbf{p}}^{1},\mathbf{n}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ 3. $(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k})=(\widetilde{\phi}^{1},\widetilde{q}^{1},\widetilde{\mathbf{p}}^{1}, \widetilde{\mathbf{n}}^{1})$.
 1. Initialization: $\widetilde{\phi}^{0}=\phi^{k-1}$, $\widetilde{q}^{0}=q^{k-1}$, $\widetilde{\mathbf{p}}^{0}=\mathbf{p}^{k-1}$, and $\widetilde{\mathbf{n}}^{0}=\mathbf{n}^{k-1}$. 2. For fixed Lagrangian multiplier $\mathbf{\lambda}_{1}=\mathbf{\lambda}_{1}^{k-1}$, $\lambda_{2}=\lambda_{2}^{k-1}$, and $\mathbf{\lambda}_{3}=\mathbf{\lambda}_{3}^{k-1}$, solve the following subproblems : $(33)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\phi}^{1} = \mbox{argmin } \mathcal{L}(\phi,\widetilde{q}^{0},\widetilde{\mathbf{p}}^{0},\widetilde{\mathbf{n}}^{0}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(34)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{q}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},q,\widetilde{\mathbf{p}}^{0},\widetilde{\mathbf{n}}^{0}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(35)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\mathbf{p}}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},\widetilde{q}^{1},\mathbf{p},\widetilde{\mathbf{n}}^{0}, , \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ $(36)\;\;\;\;\;\;\;\;\;\;\;\;\widetilde{\mathbf{n}}^{1} = \mbox{argmin } \mathcal{L}(\widetilde{\phi}^{1},\widetilde{q}^{1},\widetilde{\mathbf{p}}^{1},\mathbf{n}, \mathbf{\lambda}_{1}, \lambda_{2}, \mathbf{\lambda}_{3})$ 3. $(\phi^{k},q^{k},\mathbf{p}^{k},\mathbf{n}^{k})=(\widetilde{\phi}^{1},\widetilde{q}^{1},\widetilde{\mathbf{p}}^{1}, \widetilde{\mathbf{n}}^{1})$.
The presentation of the sizes, the numbers of iterations, and the computational time for the two real images.
 Image Size of image Number of iterations Time (sec) Fig. 3 $321\times 481$ 1600 472.3 Fig. 4 $481\times 321$ 1600 480.8
 Image Size of image Number of iterations Time (sec) Fig. 3 $321\times 481$ 1600 472.3 Fig. 4 $481\times 321$ 1600 480.8
 [1] Lingfeng Li, Shousheng Luo, Xue-Cheng Tai, Jiang Yang. A new variational approach based on level-set function for convex hull problem with outliers. Inverse Problems & Imaging, , () : -. doi: 10.3934/ipi.2020070 [2] Marion Darbas, Jérémy Heleine, Stephanie Lohrengel. Numerical resolution by the quasi-reversibility method of a data completion problem for Maxwell's equations. Inverse Problems & Imaging, 2020, 14 (6) : 1107-1133. doi: 10.3934/ipi.2020056 [3] Knut Hüper, Irina Markina, Fátima Silva Leite. A Lagrangian approach to extremal curves on Stiefel manifolds. Journal of Geometric Mechanics, 2020  doi: 10.3934/jgm.2020031 [4] Abdelghafour Atlas, Mostafa Bendahmane, Fahd Karami, Driss Meskine, Omar Oubbih. A nonlinear fractional reaction-diffusion system applied to image denoising and decomposition. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020321 [5] Alberto Bressan, Wen Shen. A posteriori error estimates for self-similar solutions to the Euler equations. Discrete & Continuous Dynamical Systems - A, 2021, 41 (1) : 113-130. doi: 10.3934/dcds.2020168 [6] Dan Zhu, Rosemary A. Renaut, Hongwei Li, Tianyou Liu. Fast non-convex low-rank matrix decomposition for separation of potential field data using minimal memory. Inverse Problems & Imaging, , () : -. doi: 10.3934/ipi.2020076 [7] Li-Bin Liu, Ying Liang, Jian Zhang, Xiaobing Bao. A robust adaptive grid method for singularly perturbed Burger-Huxley equations. Electronic Research Archive, 2020, 28 (4) : 1439-1457. doi: 10.3934/era.2020076 [8] Zexuan Liu, Zhiyuan Sun, Jerry Zhijian Yang. A numerical study of superconvergence of the discontinuous Galerkin method by patch reconstruction. Electronic Research Archive, 2020, 28 (4) : 1487-1501. doi: 10.3934/era.2020078 [9] Yuxia Guo, Shaolong Peng. A direct method of moving planes for fully nonlinear nonlocal operators and applications. Discrete & Continuous Dynamical Systems - S, 2020  doi: 10.3934/dcdss.2020462 [10] Noah Stevenson, Ian Tice. A truncated real interpolation method and characterizations of screened Sobolev spaces. Communications on Pure & Applied Analysis, 2020, 19 (12) : 5509-5566. doi: 10.3934/cpaa.2020250 [11] Lei Liu, Li Wu. Multiplicity of closed characteristics on $P$-symmetric compact convex hypersurfaces in $\mathbb{R}^{2n}$. Discrete & Continuous Dynamical Systems - A, 2020  doi: 10.3934/dcds.2020378 [12] Laurence Cherfils, Stefania Gatti, Alain Miranville, Rémy Guillevin. Analysis of a model for tumor growth and lactate exchanges in a glioma. Discrete & Continuous Dynamical Systems - S, 2020  doi: 10.3934/dcdss.2020457 [13] Leilei Wei, Yinnian He. A fully discrete local discontinuous Galerkin method with the generalized numerical flux to solve the tempered fractional reaction-diffusion equation. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020319 [14] Gang Bao, Mingming Zhang, Bin Hu, Peijun Li. An adaptive finite element DtN method for the three-dimensional acoustic scattering problem. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020351 [15] Weiwei Liu, Jinliang Wang, Yuming Chen. Threshold dynamics of a delayed nonlocal reaction-diffusion cholera model. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020316 [16] Siyang Cai, Yongmei Cai, Xuerong Mao. A stochastic differential equation SIS epidemic model with regime switching. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020317 [17] Yining Cao, Chuck Jia, Roger Temam, Joseph Tribbia. Mathematical analysis of a cloud resolving model including the ice microphysics. Discrete & Continuous Dynamical Systems - A, 2021, 41 (1) : 131-167. doi: 10.3934/dcds.2020219 [18] Zhouchao Wei, Wei Zhang, Irene Moroz, Nikolay V. Kuznetsov. Codimension one and two bifurcations in Cattaneo-Christov heat flux model. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020344 [19] Shuang Chen, Jinqiao Duan, Ji Li. Effective reduction of a three-dimensional circadian oscillator model. Discrete & Continuous Dynamical Systems - B, 2020  doi: 10.3934/dcdsb.2020349 [20] Barbora Benešová, Miroslav Frost, Lukáš Kadeřávek, Tomáš Roubíček, Petr Sedlák. An experimentally-fitted thermodynamical constitutive model for polycrystalline shape memory alloys. Discrete & Continuous Dynamical Systems - S, 2020  doi: 10.3934/dcdss.2020459

2019 Impact Factor: 1.373