Multiresolution, error-convergence halftone algorithm

Abstract

A new halftone algorithm is described. The algorithm is designed for implementation on a parallel architecture in order to provide fast, progressive coding of moderate-resolution images. The design is based on a multiresolution, hierarchical, pyramidal structure. At each pyramid level, the binarized image is compared with the original, gray-tone image over a successively larger window of pixels for calculation of a weighted averaged error. Within each level, selected binarized pixels are tested for possible changes in the binary assignment. The binary assignment is changed if the change results in a lower average error over the entire window. Varying the selection of test pixels can cause the same process to provide clustered-dot patterns and dithering. A comparison of performance with the best implementation of the error-propagation algorithm is presented visually. Quality is compared also in terms of isotropy of the texture and the appropriate blue-noise characteristics in areas of uniform gray tone. The benefits of this algorithm are realized with moderate-resolution display of the order of 512 dots X 512 dots. The processing can be carried out on smaller blocks since the results can be combined without any visible seams or edge effects.