Convolutional sparse coding (CSC) is a way to compute translation-invariant sparse representations by solving a global optimization problem. Although CSC has lately attracted increasing interest, it is not clear whether it can outperform classical techniques, such as cycle spinning, in image denoising. Our thorough comparisons show that estimates computed via CSC exhibit lower bias, but suffer from larger variance when the representations are not extremely sparse, which seriously impairs the CSC-based denoising of natural images.

This presentation is part of Minisymposium “MS40 - Recent Advances in Convolutional Sparse Representations”
organized by: Giacomo Boracchi (Politecnico di Milano) , Alessandro Foi (Tampere University of Technology) , Brendt Wohlberg (Los Alamos National Laboratory) .

Authors:

Diego Carrera (Politecnico di Milano)

Giacomo Boracchi (Politecnico di Milano)

Alessandro Foi (Tampere University of Technology)

Brendt Wohlberg (Los Alamos National Laboratory)

Keywords:

convolutional sparse coding, image denoising, image enhancement, image representation, inverse problems