Training based error diffusion and halftone quality quantification

Seong Wook Han, Purdue University

Abstract

Error diffusion algorithm is most popular in commercial use because it produces tone rendition comparable to iteratively search-based algorithm without the high computational complexity. We first present an improved training procedure for the AM/FM halftoning algorithm based on tone dependent error diffusion (TDED) that uses different sets of filter weights and thresholds depending on input pixel value. For the FM part, the TDED parameters (weights and thresholds) are optimized using a new cost function with normalization to distribute the cost evenly over all frequencies. The cost function of training the AM part is also modified by penalizing variation in measured tone value across the multiple printer settings for a combination of dot size and dot density. A print-to-scan (P2S) training framework for TDED parameters is presented to account for actual printer behavior, and we show an improvement against halftones generated by the software based TDED assuming an ideal printer. Secondly, we present new approaches to quantify halftone texture quality using directional Gaussian filters and provide three metrics that extract important attributes of halftones: granularity, directionality, and inhomogeneity. Synthetic and scanned halftones are used to evaluate the metrics. We then apply the proposed metrics to design the parameters of tone dependent error diffusion (TDED) based on a software simulation and a P2S training framework. We also design FM screens using direct binary search (DBS) with a cost function based on the proposed halftone metrics as a means to reduce tiling artifacts, especially for small mask sizes.

Degree

Ph.D.

Advisors

Allebach, Purdue University.

Subject Area

Electrical engineering

COinS