Wednesday, May 02, 2007

Adjusting color with human language commands

This is a terrific development. Hopefully, the technology will be transferred to image editing programs like Photoshop. I would rather tell Photoshop to make the yellow brighter than to hassle selecting all of the yellow in the image then tweaking the hue, saturation, tint, contrast, etc.

Kevin Rivoli for the Xerox Corporation

"Adjusting color on printers should be simpler," Geoffrey Woolfe says.

To the human eye, that flower in the photo is reddish orange, that sky is light blue, that sun shines brilliant yellow.

But when software tells a printer to reproduce that image, it uses a long, unwieldy set of numbers and letters to describe those colors — and a totally different set of characters to describe shades that are a tad lighter, or a bit darker, or a whole lot brighter. The upshot is that most laymen would have to attend the computer equivalent of Berlitz to learn how to get the shades they want.

But if Xerox has its way, that will not be true much longer. This week the company introduced the software equivalent of a translator that can turn plain color speech into fluent computerese. Type the command, “Make the sun a brighter yellow,” and the printer will read, “Go with color CIELAB[88, -3, 64].”

“You shouldn’t have to be a color expert to make the sky a deeper blue or add a bit of yellow to a sunset,” said Geoffrey Woolfe, a principal scientist in the Xerox Innovation Group, who is based in Webster, N.Y., near Rochester. “So we’re providing that middle layer that turns plain speech into mathematical code.”

Color experts are already impressed. “They’ve taken common words and modifiers and transformed them into mathematical directions,” said Roy S. Berns, a professor of color science at the Rochester Institute of Technology, which has just inaugurated a doctoral program in color science.

There are still problems to be ironed out, of course. For now, the software does not differentiate between colors in different parts of an image — in other words, it can tell the printer to make all the blues darker, but it cannot yet tell it to darken the sky but lighten the ocean and leave the blue dress alone. And it is still working on a lexicon of languages — after all, a “brighter” color to one person is a “crisper” color to another and a “sharper” color to a third.

“To understand what someone means when he says the pinks need to be brighter, the software must know what pink means and what brighter means.” Professor Berns said. “And there are individual, cultural and regional differences in the way people use those words.”

Mr. Woolfe readily acknowledges the problems, and he concedes that the product is a couple of years away from commercialization. He envisions the final version as having the capability to learn. Say particular users keep typing, “no, darker” when they ask for chartreuse. “The machine will respond with a color chart that shows what it views as chartreuse, and then will let the user show what he means by the word,” Mr. Woolfe said. If it turns out the user thinks chartreuse is forest green, the software will act accordingly from then on..."


More on Color Language Research (
Post a Comment