Inverse Halftoning Based on the Anisotropic Lpa-ici Deconvolution

A new inverse halftoning algorithm for restoring a continuous tone image from a given error diffusion halftone image is presented. The algorithm is based on a novel anisotropic deconvolution strategy [14, 15]. The linear model of error diffusion halftoning proposed by Kite et al. [19] is exploited. It approximates error diffusion as the sum of the convolution of the original grayscale image with a speciÞc kernel and colored random noise. Under this model the inverse halftoning can be therefore formulated as a special deconvolution problem. The deconvolution is performed following the RI-RWI (regularized inverse-regularizedWiener inverse) scheme [14] and exploiting the recently introduced anisotropic LPA-ICI estimator [15]. This adaptive varying scale estimator, based on the directional local polynomial approximation (LPA) technique and the intersection of conÞdence intervals (ICI) scale selection algorithm, allows near optimal edge adaptation. As a result, the reconstructed continuous tone image presents smooth areas faithful to the unknown original and yet preserves all the details found in the halftone. Conventional inverse-halftoning algorithms often produce estimates that are either oversmooth (loss of details) or still noisy. Simulation experiments conÞrm the state-of-the-art performance of the proposed algorithm, both visually and in mean-squared-error sense.