One-dimensional least-squares model-based halftoning

A least-squares model-based approach to digital halftoning is proposed. It exploits both a printer model and a model of visual perception. It produces a halftoned image that minimizes the squared error between the response of the visual model to the original image and the response of the printer and visual models to the halftoned image. For one-dimensional halftoning, in which each column is processed independently, it is shown that least-squares halftoning can be optimally performed with the Viterbi algorithm. Applying this approach to simple visual and printer models yields better halftones than do conventional one-dimensional methods. Although one-dimensional halftoning is seldom used in practice, the results of this method are of interest because their appearance is due only to the models and the fundamental nature of halftoning. Thus they can serve as a guide to the design of two-dimensional approaches for which only approximate leastsquares solutions are possible.