PARTIAL CONVOLUTION FOR TOTAL VARIATION DEBLURRING AND DENOISING BY NEW LINEARIZED ALTERNATING DIRECTION METHOD OF MULTIPLIERS WITH EXTENSION STEP

摘要

In this paper, we propose a partial convolution model for image deblurring and denoising. We also devise a new linearized alternating direction method of multipliers (ADMM) with an extension step. As the computation of its subproblem is simple enough to have closed-form solutions, its per-iteration cost is low; however, the relaxed parameter condition together with the extra extension step inspired by Ye and Yuan's ADMM enables faster convergence than the original linearized ADMM. Preliminary experimental results show that our algorithm can produce better quality results than some existing efficient algorithms within a similar computation time. The performance advantage of our algorithm is particularly evident at high noise ratios.