Abstract:The stripe noise existing in remote sensing images badly degrades the visual quality and restricts the precision of data analysis. Therefore, many destriping models have been proposed in recent years. In contrast to these existing models, in this paper, we propose a nonconvex model with a DC function (i.e., the difference of convex functions) structure to remove the strip noise. To solve this model, we make use of the DC structure and apply an inexact proximal majorization-minimization algorithm with each inner subproblem solved by the alternating direction method of multipliers. It deserves mentioning that we design an implementable stopping criterion for the inner subproblem, while the convergence can still be guaranteed. Numerical experiments demonstrate the superiority of the proposed model and algorithm.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the stripe - noise problem in remote - sensing images. Specifically, stripe noise seriously affects the visual quality of remote - sensing images and the accuracy of data analysis. Therefore, many destriping models have been proposed. In this paper, the author proposes a non - convex model, using the DC function (i.e., the difference between convex functions) structure to remove stripe noise. To solve this model, the author utilizes the DC structure and applies an inexact approximate primal - dual minimization (PMM) algorithm, where each internal sub - problem is solved by the alternating direction method of multipliers (ADMM).
### Specific description of the problem
1. **Background and problem description**:
- Remote - sensing images play an important role in fields such as environment, agriculture, biology, and mineralogy.
- Due to the limitations of sensors, remote - sensing images are inevitably affected by noise, which reduces the image quality and hinders subsequent applications (such as classification and detection).
- Stripe noise is a common type of noise in remote - sensing images, which appears as horizontal or vertical stripes in the image and seriously affects the image quality.
2. **Mathematical model**:
- The degradation process of stripe noise can be represented by the following formula:
\[
f(x,y)=u(x,y)+s(x,y)
\]
where \( f(x,y) \) is the degraded image, \( u(x,y) \) is the ideal image without stripes, and \( s(x,y) \) is the stripe component.
- Converting it into matrix - vector form:
\[
f = u + s
\]
where \( f, u, s \) are the discretized vector representations of \( f(x,y), u(x,y), s(x,y) \) respectively.
3. **Objective function**:
- To remove stripe noise, the author proposes an objective function that combines three regularization terms:
\[
\min_s\left\{\lambda_1\|\nabla_y s\|_{1,1}+\lambda_2\|\nabla_x f-\nabla_x s\|_{1,1}+\lambda_3\|s\|_{2,1}\right\}
\]
where:
- \( R_1(s)=\|\nabla_y s\|_{1,1} \): Consider the gradient of the stripe component in the vertical direction;
- \( R_2(s)=\|\nabla_x u\|_1 = \|\nabla_x f-\nabla_x s\|_1 \): Consider the smoothness of the ideal image in the horizontal direction;
- \( R_3(s)=\|s\|_{2,1} \): Consider the special column - like structure of the stripe component.
4. **Improved model**:
- To better remove noise, the author introduces the SCAD function to replace the original \( \ell_1 \) regularization term, thus transforming the model into a non - convex model.
5. **Algorithm design**:
- Use the inexact approximate primal - dual minimization (PMM) algorithm to solve the non - convex model, and each internal sub - problem is solved by the dual alternating direction method of multipliers (dADMM).
- Design an achievable stopping criterion to ensure the convergence of the algorithm.
6. **Experimental results**:
- Numerical experiments show that the proposed model and algorithm are superior in removing stripe noise.
In conclusion, this paper aims to effectively remove stripe noise in remote - sensing images and improve image quality and data analysis accuracy by proposing a new non - convex model and an efficient solution algorithm.