A Mini-Batch Quasi-Newton Proximal Method for Constrained Total-Variation Nonlinear Image Reconstruction

Tao Hong,Thanh-an Pham,Irad Yavneh,Michael Unser

2024-08-17

Abstract:Over the years, computational imaging with accurate nonlinear physical models has drawn considerable interest due to its ability to achieve high-quality reconstructions. However, such nonlinear models are computationally demanding. A popular choice for solving the corresponding inverse problems is accelerated stochastic proximal methods (ASPMs), with the caveat that each iteration is expensive. To overcome this issue, we propose a mini-batch quasi-Newton proximal method (BQNPM) tailored to image-reconstruction problems with total-variation regularization. It involves an efficient approach that computes a weighted proximal mapping at a cost similar to that of the proximal mapping in ASPMs. However, BQNPM requires fewer iterations than ASPMs to converge. We assess the performance of BQNPM on three-dimensional inverse-scattering problems with linear and nonlinear physical models. Our results on simulated and real data show the effectiveness and efficiency of BQNPM,

Optimization and Control,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the image reconstruction problem under the non - linear physical model. Specifically, the author focuses on how to reduce the computational cost while maintaining high reconstruction quality. Although the traditional Accelerated Stochastic Proximal Methods (ASPMs) are effective, the cost of each iteration is high, especially when dealing with large - scale data sets. To solve this problem, the author proposes a new method - the Batch - Quasi - Newton Proximal Method (BQNPM), which is specifically used for the image reconstruction problem with total - variation regularization.

### Main contributions of the paper:

1. **Proposed a new optimization method**:
- **BQNPM**: This method reduces the cost of each iteration by using the Weighted Proximal Mapping (WPM) and requires fewer iterations than ASPMs, thus improving the overall efficiency.

2. **Solved the computational bottleneck**:
- By introducing the method of diagonal matrix and low - rank update, the Hessian matrix is effectively estimated, reducing the memory usage and computational complexity.

3. **Wide range of applications**:
- The author verified the effectiveness of BQNPM on the three - dimensional inverse scattering problem, including image reconstruction under linear and non - linear physical models. The experimental results show that BQNPM performs better than the existing methods on both simulated and real data.

### Specific problem description:

In image reconstruction, the goal is to recover a high - quality image from noisy measurements. This usually involves solving a composite minimization problem in the following form:

\[
\min_{x \in C} \left( \frac{1}{L} \sum_{\rho = 1}^L \frac{1}{2} \| H_\rho(x) - y_\rho \|_2^2+\lambda \text{TV}(x) \right)
\]

where:
- \(x\in\mathbb{R}^N\) is the vectorized image.
- \(C\subset\mathbb{R}^N\) is a convex set.
- \(H_\rho:\mathbb{R}^N\rightarrow\mathbb{C}^M\) is the physical mapping from the image to the measurement.
- \(y_\rho\in\mathbb{C}^M\) is the \(\rho\) - th measurement.
- \(\text{TV}(x)\) is the total - variation regularization term, which is used to impose prior knowledge.
- \(\lambda > 0\) is the parameter for balancing the data fidelity term and the regularization term.

### Core challenges of the problem:

1. **Non - linear physical model**: The exact physical model is usually non - linear, which makes the computational cost of each iteration very high.
2. **Large - scale data sets**: When the data set is large, the traditional ASPMs need to calculate the complete gradient, which is computationally infeasible.
3. **Efficient convergence**: A method that can converge quickly within a small number of iterations while maintaining high reconstruction quality is required.

### Solution:

- **BQNPM**: By using the batch - quasi - Newton method combined with the weighted proximal mapping, it effectively reduces the cost of each iteration and achieves high reconstruction quality within a small number of iterations.

### Experimental verification:

- The author verified the effectiveness of BQNPM in Optical Diffraction Tomography (ODT). Especially under the non - linear model of strongly scattering samples, BQNPM shows a faster convergence speed and higher reconstruction quality.

In conclusion, this paper proposes a new optimization method BQNPM, aiming to solve the image reconstruction problem under the non - linear physical model. By reducing the cost of each iteration and the number of required iterations, it improves the overall computational efficiency and reconstruction quality.