Bregman implementation of Meyer's $G-$norm for cartoon + textures decomposition

Jerome Gilles,Stanley Osher

2024-10-30

Abstract:In this paper, we design a very simple algorithm based on Split Bregman iterations to numerically solve the cartoon + textures decomposition model of Meyer. This results in a significant gain in speed compared to Chambolle's nonlinear projectors.

Computer Vision and Pattern Recognition,Functional Analysis

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Design an efficient algorithm based on Split Bregman iteration for numerically solving the cartoon + texture image decomposition model proposed by Meyer**. Specifically, the paper aims to significantly improve the computational speed by introducing the Split Bregman iteration method, thereby replacing the existing Chambolle non - linear projection method.

### Problem Background

1. **Cartoon + Texture Image Decomposition Model**
- Meyer proposed a theoretical model, using Total Variation (TV) to model the cartoon part and a specific space \( G \) to represent the oscillation pattern (texture).
- Aujol et al. proposed an effective method to numerically solve this model, using Chambolle's non - linear projection method, but this method requires a large number of iteration times to converge.

2. **Bregman Iteration Method**
- Goldstein et al.'s work in the field of optimization shows that Bregman iteration is a very efficient and fast method, especially suitable for solving TV problems.
- However, as far as the author knows, no literature has proposed using Bregman iteration to solve the \( G \)-norm problem based on Meyer.

### Main Contributions of the Paper

- **Algorithm Design**: The paper describes an algorithm based on Split Bregman iteration for solving the function minimization problem of Meyer - based \( G \)-norm.
- **Speed Improvement**: The experimental results show that the new Bregman - iteration - based method is more efficient in decomposing images, especially significantly superior to Chambolle's non - linear projection method in terms of convergence speed.

### Summary of Mathematical Formulas

- **Meyer's \( G \)-norm Decomposition Model**
\[
F_{\lambda,\mu}(u,v)=J(u)+J^*(v / \mu)+\frac{\lambda}{2}\|f - u - v\|_2^2
\]
where \( J(u) \) is the total variation, \( J^*(v / \mu) \) is the dual form of the \( G \)-norm, \( f \) is the original image, \( u \) is the cartoon part, and \( v \) is the texture part.

- **Split Bregman Iteration Update Formula**
\[
\hat{v}=f-\frac{1}{\lambda}\text{PROF}(\lambda f,\frac{1}{\lambda\mu})
\]
where \( \text{PROF}(\lambda f,\frac{1}{\lambda\mu}) \) represents the ROF model output applied to \( \lambda f \) with a coefficient of \( \frac{1}{\lambda\mu} \).

### Conclusion

The paper proposes a new algorithm based on Split Bregman iteration for solving Meyer's \( G \)-norm cartoon + texture decomposition model. The experimental results show that this method significantly improves the computational speed while maintaining the decomposition effect.