Digital image halftoning is the process of converting an image composed of many shades of grey into an image containing only black and white. The goal is to compute the locations of the black and white dots to retain as much subjective quality in the image as possible. Image halftoning is widely used in laser and ink jet printers, monochrome monitors, and newsprint.

This talk will focus on error diffusion algorithms which provide an efficient way of converting grayscale images into high quality halftones. We show that the form error diffusion algorithms for digital halftoning is equivalent to a nonseparable two-dimensional noise-shaping feedback coder, which is a class of delta-sigma modulator. By applying delta-sigma modulation analysis techniques, we explain features in halftones produced by classical error diffusion schemes, and allow the optimization of these schemes for high visual quality.

After describing delta-sigma modulation, we will analyze the behavior of the two key components in a noise-shaping feedback coder-- the quantizer and the two-dimensional filter in the feedback path. The quantization error is not white noise but instead highly correlated with the input image. To account for this correlation, we model the quantizer as a gain plus additive noise. This model accurately predicts the edge sharpening and noise shaping caused by all error diffusion schemes and permits an extension of error diffusion to oversampled imagery. Next, we derive a measure for subjective quality based on the gain value and filter coefficients. The measure gives us a formal, mathematical framework for analyzing and optimizing the subjective quality of image halftoning by error diffusion.

Biography