Next

Introduction

The “perceptually similar” criterion is very important one, and is linked to the characteristics of the human visual system. If we take a simpler criterion of minimal mean square difference between the original and the halftoned image, we would generate a thresholded image that occludes most of the nuances of the original images. However, this is what is used when we control a computer remotely with netmeeting, for example. Instead of transmitting the whole data for the background image for example, it transmits only a few bits to characterize it. This is a multilevel thresholding, and is very noticeable to the eye. We shall see that with only one bit for each color, we may do a better job than this ! This appears also when reducing the resolution of the screen.

We exploit the fact that our high spatial frequency response drops quickly above 6 cycles per degree, and even less for chrominance components. So high frequency patterns are perceived as their average over a neighbourhood depending on the distance between two adjacent pixels on the screen, and the viewing distance.

Here are some plots of the image that we are going to use throughout this paper, and their thresholded counterpart.

Gamma correction

One needs to account for the fact that the 8-bit values that are put in the frame buffer to trigger the red, green and blue guns of the CRT are not the RGB tristimulus values of the colors displayed on the monitor. This is because the CRT has a nonlinear response to frame buffer values. Thus, we need to pass the RGB values of the image through this non-linearity to obtain the RGB coordinates of the colors displayed on the monitor. This corresponds to the inverse of the gamma correction. The color images are first preprocessed with this point nonlinearity before they are halftoned. This ensures that the colors in the halftone are closest to the color actually rendered on the monitor.

Here is the relationship I have used :

Color halftoning

Another idea that I use to go from an original grayscale halftoning algorithm to a color one is to use the fact that the eye is much more sensitive to luminance variation than to chromatic variations. Therefore, we want to try to put more noise in the chromatic components to better shape the noise in the luminance. This idea is adapted to the main techniques of color halftoning.

This is the linear transformation to change color space, process data as visual pathways intensities, and go back to RGB values.