Nonlocal self-similarity within natural images has become an increasingly popular prior in deep-learning models.
Ranked #5 on Grayscale Image Denoising on BSD68 sigma50
Gabor-like filters have been observed in the early layers of CNN classifiers and even throughout low-level image processing networks.
In this work, we propose an unrolled convolutional dictionary learning network (CDLNet) and demonstrate its competitive denoising and joint denoising and demosaicing (JDD) performance both in low and high parameter count regimes.
In addition, we leverage the model's interpretable construction to propose an augmentation of the network's thresholds that enables state-of-the-art blind denoising performance and near-perfect generalization on noise-levels unseen during training.
Ranked #9 on Grayscale Image Denoising on BSD68 sigma25
Sparse representation via a learned dictionary is a powerful prior for natural images.
Robust Principal Component Analysis (RPCA) performs low-rank and sparse decomposition and accomplishes such a task when the background is stationary and the foreground is dynamic and relatively small.