Abstract:Based on previous work we extend a primal-dual semi-smooth Newton method for minimizing a general $L^1$-$L^2$-$TV$ functional over the space of functions of bounded variations by adaptivity in a finite element setting. For automatically generating an adaptive grid we introduce indicators based on a-posteriori error estimates. Further we discuss data interpolation methods on unstructured grids in the context of image processing and present a pixel-based interpolation method. For the computation of optical flow we derive an adaptive finite element coarse-to-fine scheme for resolving large displacements. The coarse-to-fine scheme speeds-up the computing time tremendously.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop an adaptive finite element method to minimize the L1 - L2 - TV functional based on Total Variation (TV), thereby improving motion estimation in image processing. Specifically, this method aims to:
1. **Expand existing numerical methods**: Based on previous work, this research extends the primal - dual semi - smooth Newton method to the adaptive finite element environment for minimizing the L1 - L2 - TV functional in general form.
2. **Introduce adaptive mesh generation techniques**: In order to automatically generate an adaptive mesh, the author introduces an indicator based on a - posteriori error estimates. This enables the algorithm to dynamically adjust the fineness of the mesh according to the error estimate, thereby improving computational efficiency and accuracy.
3. **Propose a new interpolation method**: The data interpolation method on unstructured meshes is discussed, and a pixel - based interpolation method is proposed to optimize the interpolation effect in image processing.
4. **Accelerate optical flow computation**: To solve the large - displacement problem in image sequences, an adaptive coarse - to - fine scheme is proposed, which significantly improves the computational speed.
### Mathematical Model and Formulas
This research considers the minimization problem of the L1 - L2 - TV functional in the following form:
\[
\inf_{u \in V} \left\{ \alpha_1 \int_\Omega \varphi_{\gamma_1}(|Tu - g|) \, dx + \frac{\alpha_2}{2} \|Tu - g\|^2_{L^2} + \frac{\beta}{2} \|Su\|^2_{V_S} + \lambda \int_\Omega \varphi_{\gamma_2}(|Du|_F) \, dx \right\} =: E(u)
\]
where:
- \( V := H^1(\Omega)^m \) is a continuously embedded Hilbert space.
- \( S: V \to V_S \) is a bounded linear operator, and \( V_S \) is also a Hilbert space.
- \( T: L^2(\Omega)^m \to L^2(\Omega) \) is a bounded linear operator.
- \( g \in L^2(\Omega) \) is the given data.
- \( | \cdot |_F \) represents the Frobenius norm.
- \( \alpha_1, \alpha_2, \lambda, \beta \geq 0 \) are adjustable weight parameters.
- \( \gamma_1, \gamma_2 \geq 0 \) are Huber regularization parameters, and the corresponding Huber function is defined as:
\[
\varphi_\gamma(x) :=
\begin{cases}
\frac{1}{2\gamma} x^2 & \text{if } |x| \leq \gamma, \\
|x| - \frac{\gamma}{2} & \text{if } |x| > \gamma.
\end{cases}
\]
### Main Contributions
1. **Adaptive finite element framework**: By introducing a - posteriori error estimates, the generation of adaptive meshes is achieved, thereby improving computational efficiency and accuracy.
2. **Efficient optical flow computation**: The proposed adaptive coarse - to - fine scheme significantly reduces the computation time while improving the accuracy in the case of large displacements.
3. **Improved interpolation method**: A new pixel - adaptive interpolation method is proposed, which optimizes the interpolation effect of image data on the finite element mesh.
In summary, this paper solves the problems of efficiency and accuracy in motion estimation in image processing by combining adaptive finite element techniques and improved interpolation methods.