Convolutional Sparse Representation with Gradient and Elastic Net Penalty

Abstract

Convolutional Sparse Coding (CSC) models have recently gained considerable attention in signal and image processing. They are very effective in image denoising. For this kind of model, the regularization term is important for improving the performance of image denoising. This paper proposes a convolutional sparse representation model with a combination of gradient and elastic net penalty terms for image Gaussian noise denoising. The Combination has both advantages of gradient regularization terms and elastic net regularization terms and can achieve a better denoising effect. Experiments show that this combined penalty is better than applying gradient penalty and net elastic penalty separately in Gaussian noise denoising application.