Reproducing Color Images as Duotones

Overview

In this project, we venture into the world of offset printing. Color reproduction on a printing press traditionally uses four inks: cyan, magenta, yellow and black. An impressive range of colors can be achieved by halftoning these four inks, but many colors lie outside the achievable range. The long-term vision inspiring this project is that inks should be chosen to exactly reproduce all the colors present in an original image.

This project addresses the special case of selecting inks for duotone printing, a relatively inexpensive process in which just two inks are used. Traditional duotone printing almost always uses black as one of the two inks. The resulting reproduction is an "enhanced grayscale" image: a grayscale image with a hint of the chosen accent color. We would like to use duotone printing to achieve full color reproduction. Our system takes an image as input and allows the user to select 0, 1, or 2 inks. The output consists of the remaining inks or inks that will best reproduce the image as a duotone and the appropriate color separations.

Duotone Mapping

The core of our algorithm is the mapping from original image colors to the colors achievable by two inks, the duotone gamut. Viewed in three-dimensional color space, the original image colors are a cloud of points and the duotone gamut is a curved surface with straight edges. The corners of this surface are the colors of the paper, each individual ink, and the combination of both inks.

To map a set of image colors, we perform three transformations:

original image colors

1. luminance transform
(along green axis)

2. ink-spread transform
(along blue axis)

3. normal projection
(along red axis)

Selecting Inks

Once we know how to map an image onto a given gamut, we need to find the best gamut for a particular image. Since a gamut is defined by two inks, we need to choose the best ink pair. We use simulated annealing, a heuristic optimization technique, to select the inks. The objective function we minimize is the pixel-wise difference between the original image and the mapped image, measured in a perceptually-uniform color space.

Results

The images displayed below were printed and then scanned.

CMYK reproduction

duotone with
PANTONE 144 and
PANTONE 546

CMYK reproduction
of Degas' Women

duotone with
PANTONE 144 and
PANTONE 546

duotone with
PANTONE 172 and
PANTONE 354

CMYK reproduction

duotone with
PANTONE 152 and
PANTONE Process Blue

CMYK reproduction

duotone with
PANTONE 152 and
PANTONE Process Blue

CMYK reproduction
of Schiele's Agony

duotone with
PANTONE 329 and
PANTONE Warm Red on
yellow paper

Personnel

Joanna L. Power, Brad S. West, Eric J. Stollnitz, and David H. Salesin.

More Details

Citation:
Joanna L. Power, Brad S. West, Eric J. Stollnitz, and David H. Salesin. Reproducing color images as duotones. In Proceedings of SIGGRAPH 96, pages 237-248. ACM, New York, 1996.

On-line documents:
Article without duotones [Acrobat file, 2.8 Mb]
Article without duotones [compressed PostScript file, 3.0 Mb]

Contact:
jpower@cs.washington.edu