Error-Diffusion Fitting of Direct-Binary-Search Halftones

This paper concerns how to employ the simple error-diffusion (ED) operation to approach the ideal halftones of the complex direct-binary-search (DBS) method. The conventional bound-constrained least squares (BLS) algorithm is used in the iteration of kernel estimation and error estimation, to find the maximum-likelihood ED fitting of DBS halftones. This fitting suggests a generic ED kernel. Experimental results show that using the generic kernel in an ED process can provide very good halftones that, in the smooth regions, exhibit very good patterns comparable to the DBS halftones and retain much more edge details in the high-frequency portions.