$\ell_0$ Sparsifying Transform Learning with Efficient Optimal Updates and Convergence Guarantees

TL;DR

This paper introduces an efficient algorithm for learning well-conditioned sparsifying transforms using $ ext{l}_0$ sparsity, with proven convergence and superior speed over previous methods, demonstrated in image denoising tasks.

Contribution

It develops a globally convergent, fast algorithm for $ ext{l}_0$ sparsifying transform learning with exact solutions for each step, improving over existing iterative methods.

Findings

The proposed algorithm converges globally to local minima.

It achieves significant speed-ups compared to conjugate gradient-based methods.

Transform learning outperforms synthesis K-SVD in image denoising applications.

Abstract

Many applications in signal processing benefit from the sparsity of signals in a certain transform domain or dictionary. Synthesis sparsifying dictionaries that are directly adapted to data have been popular in applications such as image denoising, inpainting, and medical image reconstruction. In this work, we focus instead on the sparsifying transform model, and study the learning of well-conditioned square sparsifying transforms. The proposed algorithms alternate between a $\ell_0$ "norm"-based sparse coding step, and a non-convex transform update step. We derive the exact analytical solution for each of these steps. The proposed solution for the transform update step achieves the global minimum in that step, and also provides speedups over iterative solutions involving conjugate gradients. We establish that our alternating algorithms are globally convergent to the set of local minimizers of the non-convex transform learning problems. In practice, the algorithms are insensitive to initialization. We present results illustrating the promising performance and significant speed-ups of transform learning over synthesis K-SVD in image denoising.