Blind inverse halftoning via hybrid denoising and estimation

Abstract

By using the advantages of denoising and estimation method, a blind inverse halftoning algorithm is proposed .It is called blind since we do not require the knowledge of the halftone kernel. The proposed scheme first designs a denoising preprocessor based on the edge measure (EM) of noisy halftone and an initial estimate is achieved with the preprocessor. Then using edge-preserving Huber Markov random field model, the initial estimate is updated via MAP estimation. A fast matrix operation estimation method is developed to accelerate the process. Distinct features of the proposed approach include efficient edge preserving ability while smoothing the halftone patterns, an excellent PSNR performance with comparable low time complexity and memory buffer.