A unified neutral network approach to digital image halftoning

摘要

A technique for digital halftoning of images that is based on a symmetric parallel formulation of the error diffusion halftoning method is developed. Appropriately trained or designed near-neighbor-connected symmetric neural networks are used to generate halftoned images in an efficient parallel way, based on the minimization of a frequency-weighted mean-squared error criterion. Image halftoning by symmetric networks is described using an appropriate image distortion measure. The problem of determining the network interconnection weights is examined. Examples which illustrate the performance of the proposed technique compared to that of conventional or more complex methods are given.