Browse Prior Art Database

Separated Luminance and Chrominance Diffusion

IP.com Disclosure Number: IPCOM000108465D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 2 page(s) / 108K

Publishing Venue

IBM

Related People

Edgar, AD: AUTHOR

Abstract

This article handles chrominance and luminance components separately in the palette mapping of a color image. By weighing luminance more than chrominance in the color distance calculation, the graininess of the mapped image is reduced compared to the prior art.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Separated Luminance and Chrominance Diffusion

       This article handles chrominance and luminance components
separately in the palette mapping of a color image.  By weighing
luminance more than chrominance in the color distance calculation,
the graininess of the mapped image is reduced compared to the prior
art.

      In mapping an image to a limited palette of colors, common art
picks the closest color from the palette to represent a given pixel.
Because of the limited palette, there is normally an unavoidable
error in making this choice.  The opposite of this error is added to
the next pixel so that the error is canceled as the next pixel is
processed.  Of course, the next pixel has its own error, but these
errors continue to be diffused forward and canceled. This process
accurately reproduces any color over a large area, leaving a random
high-frequency grainy noise that is more acceptable than large area
color inaccuracies.

      In this error diffusion process the closest colors must be
selected, and therefore an equation is needed giving the distance of
two colors.  In the common art, this distance was normally the
cartesian distance in red, green, and blue color space.  This put
equal emphasis on the luminance, or brightness difference and the
chrominance, or color distance.

      Assume for a moment that the three-dimensional color space is
actually two-dimensional.  As in the figure, the two dimensions are
luminance and chrominance.  In the figure the tiny o's (1) represent
colors in the original image, and the x's (2) represent the colors in
the limited palette.  Let all the colors be uniformly spaced, and let
there be "N" image colors for each palette color. Therefore, around
each palette color "x", there is a region of area "N" containing "N"
image colors for which that palette color is the closest pick.

      First assume that the color distance formula follows the common
art approach of equal emphasis on luminance and chrominance.  Then,
the region of closest pick around each palette color is approximately
a circle (3), as in the figure, and the average error, or noise
produced by the mapping process, is proportional to the radius of
that circle.

      Now assume that the eye was much less sensitive to chrominance
noise than to luminance.  Now the dist...