Edge coherence-weighted second-order variational model for image denoising

High-order variational models have the ability to remove the staircase effect generated by the total variation regularizer. They, however, tend to blur object edges. To overcome this drawback, we introduce an edge coherence-weighted second-order (ECSO) model for image denoising. We propose novel regularizers that use the edge coherence quantity to adjust the strength of regularization according to the characteristics of each pixel. We then adapt the split Bregman algorithm to solve the proposed model. All the subproblems are solved efficiently using the fast Fourier transform and the shrinkage operator. Extensive experiments show that the proposed model outperforms state-of-the-art high-order variational models for image denoising.