Exploiting Image Local And Nonlocal Consistency For Mixed Gaussian-Impulse Noise Removal

摘要

Most existing image denoising algorithms can only deal with a single type ofnoise, which violates the fact that the noisy observed images in practice areoften suffered from more than one type of noise during the process ofacquisition and transmission. In this paper, we propose a new variationalalgorithm for mixed Gaussian-impulse noise removal by exploiting image localconsistency and nonlocal consistency simultaneously. Specifically, the localconsistency is measured by a hyper-Laplace prior, enforcing the localsmoothness of images, while the nonlocal consistency is measured bythree-dimensional sparsity of similar blocks, enforcing the nonlocalself-similarity of natural images. Moreover, a Split-Bregman based technique isdeveloped to solve the above optimization problem efficiently. Extensiveexperiments for mixed Gaussian plus impulse noise show that significantperformance improvements over the current state-of-the-art schemes have beenachieved, which substantiates the effectiveness of the proposed algorithm.