Approximate Bayesian denoising for deep image reconstruction in the presence of signal-dependent noise
Résumé
Recently, a variety of unrolled networks have been proposed for image reconstruction. These can be interpreted as parameter-optimized algorithms that incorporate steps that are traditionally encountered during the optimization of hand-crafted objectives . Here, we address the problem of training such networks in the presence of signal-dependent noise, which is more realistic that the common additive Gaussian noise . We focus on algorithms that requires the inversion of large signal-dependent matrices during training, which increases considerably the training time compared signal-independent inversions that can be precomputed before training. In particular, we describe how to approximate the denoising step of the deep expectation-maximization network to reduce the computational cost and memory requirements while limiting the reconstruction error. We present reconstruction results from simulated and experimental data at different noise levels. Our network yields higher reconstruction peak signal-to-noise ratios than other similar approaches and greater robustness in the practical case where the noise level is unknown or is badly estimated.
Origine : Fichiers produits par l'(les) auteur(s)