Image denoising via solution paths

Many image denoising methods can be characterized as minimizing “loss + penalty,” where the “loss” measures the fidelity of the denoised image to the data, and the “penalty” measures the smoothness of the denoising function. In this paper, we propose two models that use the L1-norm of the pixel updates as the penalty. The L1-norm penalty has the advantage of changing only the noisy pixels, while leaving the non-noisy pixels untouched. We derive efficient algorithms that compute entire solution paths of these L1-norm penalized models, which facilitate the selection of a balance between the “loss” and the “penalty.”