Abstract:The $\ell^1$ and total variation (TV) penalties have been used successfully in many areas, and the combination of the $\ell^1$ and TV penalties can lead to further improved performance. In this work, we investigate the mathematical theory and numerical algorithms for the $\ell^1$-TV model in the context of signal recovery: we derive the sample complexity of the $\ell^1$-TV model for recovering signals with sparsity and gradient sparsity. Also we propose a novel algorithm (PGM-ISTA) for the regularized $\ell^1$-TV problem, and establish its global convergence and parameter selection criteria. Furthermore, we construct a fast learned solver (LPGM-ISTA) by unrolling PGM-ISTA. The results for the experiment on ECG signals show the superior performance of LPGM-ISTA in terms of recovery accuracy and computational efficiency.
What problem does this paper attempt to address?
### What problems does this paper attempt to solve?
This paper aims to solve the key problems in signal recovery, especially within the Compressed Sensing (CS) framework. Specifically, the paper focuses on how to combine ℓ1 - regularization and Total Variation (TV) regularization to recover signals with sparsity and gradient sparsity. The main research contents include:
1. **Mathematical theory**:
- Derived the sample complexity in signal recovery for the ℓ1 - TV model, that is, the minimum number of measurements required for successful signal recovery.
- Analyzed the influence of signal sparsity and gradient sparsity on the recovery performance.
2. **Numerical algorithms**:
- Proposed a new algorithm - Proximal Gradient Mapping ISTA (PGM - ISTA), for solving the optimization problem with ℓ1 - TV regularization, and proved its global convergence and parameter selection criteria.
- Constructed a fast learning solver (Learned PGM - ISTA, LPGM - ISTA) by expanding the PGM - ISTA algorithm to improve computational efficiency.
3. **Experimental verification**:
- Through experiments (such as electrocardiogram signal recovery) demonstrated the superior performance of LPGM - ISTA in recovery accuracy and computational efficiency.
### Formula summary
- The form of the ℓ1 - TV regularization problem is:
\[
(\ell_1\text{-TV}) : \min_{x \in \mathbb{R}^n} \lambda_1 \|x\|_1 + \lambda_2 \|x\|_{\text{TV}}, \quad \text{s.t.} \quad \|Ax - y\|_2 \leq \epsilon,
\]
where $\lambda_1 \geq 0$ and $\lambda_2 \geq 0$ are regularization parameters, $\|x\|_{\text{TV}}=\|Dx\|_1$ represents the total variation semi - norm, and $D$ is a row - cyclic difference matrix.
- The upper bound of sample complexity:
\[
\delta(D(g, x)) \leq \Phi(s_r, s_g),
\]
where
\[
\Phi(s_r, s_g)=n - \frac{6}{\pi}(\lambda_1(n - s_r)+\sqrt{2\lambda_2(n - 1 - s_g)})^2(3n\lambda_1^2 + 4(2n + s_g - 4)\lambda_2^2+12\lambda_1\lambda_2\min\{s_r, s_g\}).
\]
- Recovery guarantee:
\[
m > (\sqrt{\Phi(s_r, s_g)+t})^2 + 1,
\]
if the above condition is satisfied, then with probability at least $1 - e^{-t^2/2}$,
\[
\|x^* - \hat{x}\|_2 \leq \frac{2\epsilon}{\sqrt{m - 1 - \Phi(s_r, s_g)-t}}.
\]
### Summary
This paper, by combining ℓ1 - regularization and total variation regularization, proposed a new mathematical theory and numerical algorithm, and solved the problems of sparsity and gradient sparsity in signal recovery. Through theoretical analysis and experimental verification, it has been proved that the new method has superiority in recovery accuracy and computational efficiency.