See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/5597684

Digital color halftoning with generalized error diffusion and multichannel

green-noise masks

Article  in  IEEE Transactions on Image Processing · February 2000

DOI: 10.1109/83.841537 · Source: PubMed

CITATIONS READS
42 276

3 authors, including:

Gonzalo R. Arce
University of Delaware
366 PUBLICATIONS   7,061 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Coded Aperture X-ray Imaging. View project

Nonlinear filters View project

All content following this page was uploaded by Gonzalo R. Arce on 25 August 2014.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000 923

Digital Color Halftoning with Generalized Error

Diffusion and Multichannel Green-Noise Masks
Daniel L. Lau, Gonzalo R. Arce, Senior Member, IEEE, and Neal C. Gallagher, Fellow, IEEE

Abstract—In this paper, we introduce two novel techniques for The problems of moiré and screen angles are avoided in
digital color halftoning with green-noise—stochastic dither pat- frequency modulated halftoning where continuous tone is
terns generated by homogeneously distributing minority pixel clus- produced by varying the distance between printed dots and
ters. The first technique employs error diffusion with output-de-
pendent feedback where, unlike monochrome image halftoning, an not varying the size. Typically, FM halftones are produced
interference term is added such that the overlapping of pixels of by the process of error diffusion which creates a stochastic
different colors can be regulated for increased color control. The arrangement of dots. Besides avoiding moiré, FM halftoning,
second technique uses a green-noise mask, a dither array designed by isolating minority pixels, maximizes the spatial resolution
to create green-noise halftone patterns, which has been constructed of the printed image relative to the printer [2], but this distri-
to also regulate the overlapping of different colored pixels. As is the
case with monochrome image halftoning, both techniques are tun- bution also maximizes the perimeter-to-area ratio of printed
able, allowing for large clusters in printers with high dot-gain char- dots [3]—making FM halftones more susceptible to printer
acteristics, and small clusters in printers with low dot-gain charac- distortions such as dot-gain, the increase in size of a printed
teristics. dot. Whether a function of the printing process (mechanical
Index Terms—AM, color, dither techniques, FM, green-noise, dot-gain) or of the optical properties of the paper (optical
halftoning. dot-gain), dot-gain causes the printed halftone to appear darker
than the original ratio of white-to-black pixels [4]. In printers
with high dot-gain characteristics, AM halftoning, with its
I. INTRODUCTION
lower spatial resolution and moiré, may be the preferred tech-

D IGITAL halftoning is a technique used by binary display

devices to create, within the human eye, the illusion
of continuous tone. Designed to mimic analog techniques,
nique, as its clustered-dots have the lower perimeter-to-area
ratio.
An alternative to AM and FM halftoning, Levien’s [5] error
dot-clustered ordered dithering or amplitude modulated (AM) diffusion with output-dependent feedback is an AM-FM hybrid
halftoning produces this illusion by varying the size of round which creates the illusion of continuous tone by producing a sto-
printed dots which are arranged along an ordered grid. When chastic patterning of dot clusters which vary in both their size
using AM halftoning, the parameters of particular importance and in their separation distance. The major advantage, of this
are the lines-per-inch (lpi) or the number of rows/columns new technique over prior error diffusion schemes, is that by ad-
of the regular grid1 and the screen angle or the orientation justing a single parameter, the output is tunable—capable of cre-
of the regular grid relative to the horizontal axis. Typically, ating halftones with large clusters in printers with high dot-gain
monochrome screens have an angle of as the human visual characteristics and small clusters in printers with low dot-gain
system is least sensitive to diagonal artifacts [1]. characteristics. Error diffusion with output-dependent feedback,
In color printers, the illusion of continuous shades of color is therefore, can trade halftone visibility for printer robustness.
produced by superimposing the binary halftones of cyan, ma- Studied by Lau et al. [2], Levien’s technique creates patterns
genta, yellow, and black (CMYK) inks. As the dots of an AM described in terms of their spectral content as green-noise—con-
halftone form a regular grid, clustered-dot dithering suffers from taining no low or high frequency spectral components. This
moiré—the secondary interference patterns created by superim- green-noise model is presented in accordance with Ulichney’s
posing two or more regular patterns. In order to minimize the [6] blue-noise model which describes the spectral characteris-
appearance of moiré, the screens of cyan, magenta, yellow, and tics of the ideal error-diffused halftone patterns as having no
black are typically oriented at the angles of , , , and low-frequency content. Furthermore, as Mitsa and Parker [7]
to create a pleasant rosette pattern. used the spectral characteristics of blue-noise to generate the
blue-noise mask, a binary dither array which greatly reduces
the computational complexity associated with FM halftoning,
Manuscript received February 5, 1999; revised September 17, 1999. This re-
search was supported in part by the National Science Foundation under Grant Lau et al. [8], using the spatial and spectral characteristics of
CDA-9703088 and by Lexmark International. The associate editor coordinating green-noise, have introduced the green-noise mask.
the review of this manuscript and approving it for publication was Prof. Jan P. The problem yet to be addressed in the evolution of green-
Allebach.
The authors are with the Department of Electrical and Computer Engineering, noise halftoning is its application to color. FM halftoning has
University of Delaware, Newark, DE 19716 USA (e-mail: lau@ece.udel.edu; been studied in great detail with respect to color printing. The
arce@ece.udel.edu; gallaghe@ece.udel.edu). techniques introduced range from simply halftoning each color
Publisher Item Identifier S 1057-7149(00)03564-8.
independently to more complex model-based techniques which
1The highest quality AM halftones will have 150 lpi or more. transform the CMYK color space to alternate spaces such as the
1057–7149/00$10.00 © 2000 IEEE
924 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

CIE La*b* space [9]. Yao and Parker [10] have even introduced
the blue-noise mask to color halftoning.
This paper introduces, to color printing, green-noise
halftoning. The first technique extends error diffusion with
output-dependent feedback to not only cluster pixels of like
color but also to regulate the clustering of pixels of different
colors. That is, with this new technique, the halftoning of
different colors can be correlated such that the superimposing
of different inks can be increased or decreased. This is in direct
contrast to independently halftoning each channel—offering
far greater control of resulting halftone patterns.
A second technique, to be introduced in this paper, incorpo-
rates a desired correlation between colors to construct a mul-
tichannel green-noise mask—extending the capabilities of the
R r
Fig. 1. Spatial ring ( )= fn r  jn 0 mj < r drg used to calculate
: +
prior work in [8] to include color halftoning. By design, this new Rr
the pair correlation ( ) for a binary dither pattern.
mask maintains all the desirable attributes of the monochrome
mask (isotropic, tunable coarseness) while also regulating the
overlapping of pixels of different colors. point process, the statistical properties of are invariant to ro-
tation, and therefore, for an isotropic process is written
II. HALFTONE STATISTICS as and is referred to as the pair correlation. is ex-
plicitly defined as
Point process statistics have a long history in stochastic geom-
etry [11]–[13] and were recently introduced to digital halftoning
by Lau et al. [2] for the study of periodic dither patterns. In this
(2)
framework, is a stochastic model governing the location of
points in . , a sample of , is the set
where is a point in , and the scalar quantity repre- the ratio of the expected number of minority pixels located in
sents the number of ’s in , a subset of . In terms of a the ring (Fig. 1) under
monochrome digital halftone pattern, a point is defined to be a the condition that is a minority pixel to the unconditional
minority pixel [a white (1) pixel for gray level and expected number of minority pixels located in . is
a black (0) pixel for ]. also the average of all pixels in the set .
So given , a binary dither pattern representing a mono- offers an especially useful tool for characterizing a pe-
chrome image of constant gray level , indicates that riodic dither patterns as illustrated in Fig. 2 where three dither
the pixel is a minority pixel; otherwise, . As a patterns representing gray level are shown with
random quantity, the first order moment or the expected value of their corresponding pair correlations (both the calculated and
is its intensity , which is the unconditional probability the ideal). The first pattern (left) is a white-noise dither pattern,
that is a minority pixel with for and as such has a pair correlation for all as the
and for . For a stationary point value of any single pixel in is independent of all other pixels.
process where the statistical properties of are independent of The name (white-noise) derives from the fact that the resulting
, . power spectrum remains constant for all frequencies [6].
A second metric for characterizing the statistical properties The second pattern (center) is composed of blue-noise which
of is the reduced second moment measure defined represents gray level by distributing the minority pixels within
as as homogeneously as possible—resulting in a dither pattern
where minority pixels are placed, on average, a distance of
(1) apart where

the ratio of the probability that pixel is a minority pixel for

under the condition that is a minority pixel to the un- (3)
for
conditional probability that is a minority pixel.
can be interpreted as a measure of the influence of a minority and is the minimum distance between addressable points on
pixel at on the pixel with indicating the display [6], [7]. Referred to as the principle wavelength of
that is more likely to be a minority pixel given and blue-noise, is illustrated in as series of peaks at integer
indicating that is less likely to be a minority multiples of . The term “blue-noise” denotes that the spectral
pixel. indicates that has no influence on components of a blue-noise dither pattern lie almost exclusively
, and in the case of a dither pattern constructed from un- in the high (blue) frequency range.
correlated (white) noise, for all and . The final pattern (right) is green-noise where gray level is
For a stationary point process, where is represented by homogeneously distributed minority pixel clus-
the distance from to and is the direction. For an isotropic ters. These clusters are separated, center-to-center, by an av-
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 925

Fig. 2. Pair correlations, R(r), for (left) white-noise, (center) blue-noise and (right) green-noise dither patterns representing gray level 15=16.

erage distance of , the principle wavelength of green-noise, monochrome binary dither patterns where
where is the gray level of pattern and is the corresponding point
process.
for In this new framework, the quantity is the reduced
(4) second moment measure between colors such that
for
(6)
and is the average number of minority pixels per cluster.
In , it is the separation of clusters apart that leads to
a series of peaks at integer multiples of ; furthermore, it is the is the ratio of the conditional probability that is a minority
clustering of minority pixels that leads to a nonzero component pixel given that a minority pixel exists at sample of to
for near zero with for . The parameter the unconditional probability that is a minority pixel [8].
is the cluster radius and is related to as Similar to , indicates that the location
of minority pixels in colors and are uncorrelated. The pair
correlation between colors and follows as
(5)

where is the area covered by a circle with radius . Lau (7)

et al. [2] note the is most apparent in when the varia-
tion in cluster size is small as increasing the variation leads to a the ratio of the expected number of minority pixels of color
“whitened” dither pattern where the peaks and valleys of located in the ring
become blurred. The term “green” refers to the resulting pat- under the condition that is a minority pixel to the un-
terns’ predominantly mid-frequency spectral components with conditional expected number of minority pixels with color
the blue-noise model a limiting case ( ). located in .

A. Color Halftoning
III. GENERALIZED ERROR DIFFUSION
In the case of a color halftone, the monochrome model
must be revised as a dither pattern is now composed of colors A. Monochrome
where, for generality, the quantity is an arbitrary integer. For In error diffusion (Fig. 3), the output pixel is determined
RGB and CMYK, where images are composed of the additive by adjusting and thresholding the input pixel such that
colors red, green and blue or the subtractive colors cyan, ma-
genta, yellow, and black, and 4, respectively. So for if
(8)
color images, the halftone pattern is now composed of the else
926 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

Fig. 3. Error diffusion algorithm as introduced by Floyd–Steinberg [14].

Fig. 4. Error diffusion with output-dependent feedback algorithm as

where is the diffused quantization error accumulated introduced by Levien [5].
during previous iterations as

(9)

with and . Using

vector notation, (9) becomes
(10)
where and
. Fig. 5. Generalized error diffusion algorithm.
In [5], Levien adds an output-dependent feedback term
(Fig. 4) to (8) such that
where is the binary output pixel of color . Assuming all
(11) channels are halftoned independently, the binary output pixel
else is determined as
where is the hysteresis or feedback term defined as if
(16)
else
(12)
where and are the error and hysteresis terms, re-
spectively, for the th color. The error term, being a vector, is
with and is an arbitrary constant. Referred to as calculated as
the hysteresis constant, acts as a tuning parameter with larger
leading to coarser output textures [5] as increases ( )
or decreases ( ) the likelihood of a minority pixel if the
previous outputs were also minority pixels. Equation (12) can .. .. .. .. ..
. . . . .
also be written in vector notation as
(13) (17)
where and where are the filter weights regulating the dif-
. The calculation of the parameters and fusion of error in the th channel and is the
remains unchanged in Levien’s approach. So in summary vector com-
of Levien’s error diffusion with output-dependent feedback, the posed exclusively from errors in channel such that
binary output pixel is determined as . The hysteresis term
if , also a vector, is calculated as
(14)
else
where such
that and .. .. .. .. .. .. .. ..
. . . . . . . .
.

B. Color
Now consider the -channel case where an output pixel is not ..
the binary pixel but the -dimensional vector such that .

(18)
.. (15)
. where are the filter weights and is the hysteresis constant
that regulates the diffusion of feedback in the th channel.
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 927

Fig. 6. Pair correlations for CMYK halftone patterns with no diffusion between colors using: (left) Floyd–Steinberg error diffusion weights and no hysteresis,
=
(center) Levien diffusion with small hysteresis constant (h 0:5), and (right) Levien diffusion with medium hysteresis constant (h = 1:0).

Generalized even further, (17) becomes pixel at . Finally, we summarize error diffusion by the gen-
eralized error diffusion equation
(23)
.. .. .. .. .. (19)
. . . . .
C. Simulations
where quantization error can now be diffused between channels Shown in Fig. 12(a) is the resulting CMYK dither patterns
through , the error filter weights which regulate the diffusion created by halftoning a pixel color image of con-
of error from channel to channel ; furthermore, (18) becomes: stant color value using the
Floyd–Steinberg [14] error filter weights with no hysteresis
and no dependencies between colors. Before halftoning,
.. .. .. .. low-level white-noise was added to the first scan line of the
. . . . original color image in order to minimize edge effects and also
to unsynchronize the resulting dither patterns. Since in this
configuration where all colors are halftoned exactly the same
way and with each color of the original image identical to the
.. .. .. .. .. (20) other colors, the resulting pattern of each color will also be
. . . . . identical (synchronized) to the other patterns. So by adding a
single or even a few lines of low-level noise eliminates this
where the previous outputs of channel can impact all other synchronization between colors; furthermore, adding several
channels where for as and regulate the columns of white-noise also minimizes edge effects. In this
diffusion of feedback from channel to . paper, dither patterns created by error diffusion are the result
Before concluding this section, we make one last, but signifi- of using a serpentine (left-to-right and then right-to-left) raster
cant, modification to error diffusion (Fig. 5) by first defining the scan on a continuous-tone image where low-level white-noise
thresholding function as ( , ) has been added for the sole purpose
of unsynchronizing each channel to both the edge rows and
columns. The halftoned images are then cropped to exclude
(21) those same rows and columns.
where is the accumulated input For a statistical analysis of the resulting dither pattern, Fig. 6
pixel. The interference matrix is added to (21) as (left) shows four plots labeled cyan, magenta, yellow, and
black corresponding to the CMYK dither pattern in Fig. 12(a).
(22) Shown in the first plot (labeled cyan) is the pair correlation be-
such that is the influence on the thresholding function of tween colors cyan versus cyan [ ], cyan versus magenta
color by the accumulated input of color . The effect of [ ], cyan versus yellow [ ] and cyan versus black
is to increase ( ) or decrease ( ) the likelihood [ ]. The small diamonds placed along the horizontal
of a minority pixel at based on the likelihood of a minority axis indicate the principle wavelengths and cluster radii for
928 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

Fig. 7. = 1 5) with: (left) a negative interference term

Pair correlations for CMYK halftone patterns using Levien diffusion with high hysteresis constant (h :
(S = 00 2), (center) no interference term (
: S =0 ), and (right) a positive interference term (S = +0 2). :

. As would be expected for a monochrome image, the different colors with lesser leading to lesser overlap. That
pair correlation exhibits blue-noise characteristics as is, given that a cyan pixel is very likely to be printed, minority
the pair correlation shows pixels for magenta, yellow and black are less likely to be printed
1) for near zero; at that same pixel location. This behavior is well illustrated in
2) frequent occurrence of the interpoint distance ; the pair correlations where for . For com-
3) decreasing influence with increasing . parison, Fig. 7 (center) [Fig. 12(e)] shows the case where is
Having no diffusion between colors and zero interference ( the identity matrix (no interference) with a flat pair correlation
, the identity matrix), the pair correlations between channels between minority pixels of different colors. For further compar-
are predominantly flat as minority pixels of color have no in- ison, Fig. 7 (right) [Fig. 12(f)] shows the case where is the
fluence on minority pixels of color . The remaining plots show matrix defined by for and for
similar relationships for colors magenta, yellow, and black. . Here, the effect of is to increase the superposition of
Shown in Fig. 6 (center) are the resulting pair correlations minority pixels such that a minority pixel of color with a high
to Fig. 12(b) where Levien’s error diffusion scheme has been likelihood of being printed making a minority pixel of any color
implemented with a low hysteresis constant , no diffu- more likely.
sion between colors and zero interference. With a low hysteresis
constant, this scheme generates blue-noise patterns very similar IV. MULTI-CHANNEL GREEN-NOISE MASKS
to that generated using the Floyd-Steinberg filter weights. Fig. 6 The green-noise mask is a novel approach to dither array
(right) [Fig. 12(c)] shows Levien’s error diffusion scheme with a screening where a continuous-tone image is converted to a
medium hysteresis constant where the patterns begin to binary halftone image by performing a pixel-wise comparison
exhibit clustering as the average size of a minority pixel cluster between the original and the dither array or mask. Previously,
is 1.95 pixels. In each color, the pair correlation exhibits strong halftoning with green-noise has implied error-diffusion based
green-noise characteristics as each plot shows methods which although are tunable (capable of creating
1) clustering as indicated by for ( halftone patterns with large clusters for printers with high
); dot-gain characteristics and small clusters for printers with
2) frequent occurrence of the intercluster distance low dot-gain characteristics) carry a high computational cost.
; Now through the use of a green-noise mask, halftoning can
3) decreasing influence with increasing . As before with create a stochastic patterning of dots with adjustable coarseness
zero influence and no diffusion between colors, the pair but with the same computational freedom as ordered-dither
correlations between colors remains predominantly flat halftoning schemes—an advantage that, for many printing
for all . devices, overcomes the drawbacks of distortions inherent to
The dither patterns of Fig. 7 illustrate the effects of , the dither array halftoning such as tiling artifacts. Many such
interference matrix, with Fig. 7 (left) [Fig. 12(d)] showing the drawbacks, though, can be minimized and sometimes visually
case where is the matrix defined by for and eliminated using device dependent compensation techniques.
for . In all instances where , has Introduced in [8] for monochrome images, the green-noise
the effect of reducing the superposition of minority pixels of mask is defined by the set, , of
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 929

binary green-noise dither patterns with one pattern, , corre-

sponding to each possible discrete gray level (256 patterns
for 8-bit gray-scale images). This set satisfies the stacking con-
straint that for any two gray-levels and with ,
(if then ). As a consequence, a pixel
of the dither array or mask is defined simply
as the minimum for which . The size parameters
and are arbitrary integers with larger masks constructed
by tiling edge-to-edge the original mask such that the
output pixel, , after halftoning the input pixel, , R
Fig. 8. Pair correlation shaping function, ~ (r ), used to construct green-noise
with principle wavelength  .
is defined as
mod mod (24) probability of every element directly adjacent is set to zero; fur-
where is the thresholding function of (21). thermore, as has a peak at , the blue-noise principle
For color halftoning, the multichannel green-noise mask is wavelength, all elements a distance from each new minority
defined by the set and pixel should be increased to ensure a peak exists in the pair cor-
where is the binary green-noise dither pattern for color relation of the final pattern.
and intensity level (for 24-bit RGB color this corresponds to In practice, how much to increase or decrease a given proba-
256 patterns per channel or total patterns). Like the bility, in BIPPCCA, is defined according to , the pair cor-
monochrome set, this set must also satisfy the stacking con- relation shaping function. is a user-defined function based
straint but only within a given color such that on the desired pair correlation with increasing leading to
if for color and intensity levels and with stronger correlations and decreasing leading to reduced.
. A pixel, , of the multichannel green-noise At , minority pixels are completely inhibited. In de-
mask is therefore defined as: signing , it is important to note that does not have ab-
solute control over the resulting , but with careful tuning,
will approximate the shape of . Shown in Fig. 8 is the
.. (25) shaping function used by Lau et al. [8] to construct green-noise
. dither patterns. This function has peaks at integer multiples of
, the green-noise principle wavelength, and valleys mid-way
where the output after halftoning is defined by between. The parameter is a tuning parameter and is shown in
mod mod (26) [8] to create visually pleasing patterns when . Being
piecewise linear, this pair correlation shaping function is an es-
A. Monochrome BIPPCCA pecially simple approximation of the pair correlation of the ideal
green-noise pattern for a given gray level and cluster size, but
The physical construction of binary dither patterns for the by itself, resulting dither patterns tend to look noisy and nonsta-
monochrome green-noise mask is done through BIPPCCA (the tionary.
BInary Pattern Pair Correlation Construction Algorithm). The Because stationarity is a necessary property for digital
basic premise of BIPPCCA is to take an empty array (containing halftoning [6], the most likely pixel will no longer be the
no minority pixels) and assign, to each element, a probability majority pixel with the highest probability, but instead be the
of that element becoming a minority pixel. BIPPCCA will then majority pixel with the highest product
convert the most likely elements to minority pixels, one at a where is the probability of a given pixel and
time, until the ratio of black to white pixels is , the desired is a function of the density of minority pixels within
gray level. The most likely element is the majority pixel with the surrounding area. Referred to as the concentration matrix,
the highest probability during the current iteration, and in order makes majority pixels more likely to become minority
for the resulting dither pattern to have desired statistical prop- pixels in areas of low minority pixel concentration and less
erties (i.e., a desired pair correlation), BIPPCCA will adjust, at likely in areas of high.
each iteration, the probability of each majority pixel in the array In BIPPCCA, the concentration of minority pixels is mea-
according to the current set of minority pixels. sured as the output after applying a low-pass filter, , using
In BIPPCCA, the initial assignment of probabilities is done in circular convolution. In [8], Lau et al. construct green-noise pat-
an uncorrelated manner, but as each new minority pixel is added, terns using the Gaussian filter, , defined as
the probabilities of all neighboring majority pixels are adjusted
according to the desired pair correlation of the resulting pat- (27)
tern. As is a function of the radial distance between pixels,
a majority pixel’s probability is increased if its radial distance where has a wide-spread impulse response for large
from the newest minority pixel, , corresponds to and where clusters are far apart and a narrow-spread impulse re-
decreased if corresponds to . As an example, con- sponse for small where clusters are close together. How much
sider using BIPPCCA to construct a blue-noise pattern where to increase or decrease a probability according to the minority
the pair correlation is zero for near zero. This feature of pixel concentration is then determined by the user through a
is achieved in BIPPCCA if with each new minority pixel, the mapping of the filtered output to the concentration matrix. Fig. 9
930 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

R R
Fig. 10. Pair correlation shaping functions, ~ (r ) (solid line) and ~ (r )
(dashed line), used to construct color green-noise with principle wavelength 
and decreased overlap of minority pixels of different colors.

As described, the above algorithm is not suited to the design of

green-noise masks as the stacking constraint will not be satisfied
for all patterns. BIPPCCA must, therefore, be constrained to
Fig. 9. Mapping function used to construct the concentration matrix C create a pattern, , which is constructed to have, as a subset,
from the output after filtering  with the low-pass filter H using circular
convolution. all constructed patterns for which . must also be a
subset of all constructed patterns for which .
In order to constrain BIPPCCA, assume first that
shows the mapping of concentration values used in [8] to deter- and that . The first step in then to initialize
mine where represents the output after fil- to , where is a minority pixel, instead of an all
tering the binary dither pattern of the current iteration, , with majority pixel array. Step 3 is then applied for each and every
the low-pass filter . In this mapping, values of minority pixel in to the probability matrix, , of uniformly
are scaled in a linear fashion such that distributed random numbers. BIPPCCA then continues at step 2
and . where, in order to satisfy the constraint that , only those
In summary, the steps for monochrome BIPPCCA are per- majority pixels ( ) for which are considered
formed as follows where is initially an array with no for swapping. BIPPCCA then continues using these modified
minority pixels. steps until a sufficient number of minority pixels in exist in .
1) Create an array, , of uniformly distributed For the case of , the same modifications as above are
random numbers such that is the used except that is initialized to where minority pixels are
probability that will become a minority pixel. equal to 0. Step 2 is then constrained by and not .
2) Construct the concentration matrix using a user-de- to and applying step 3 for each and every minority
fined mapping of , the output after filtering pixel in to the probability matrix, , of uniformly dis-
with the low-pass filter using circular convolu- tributed random numbers. BIPPCCA then continues at step 4.
tion, and then locate the majority pixel in with the The second constraint is satisfied in step 2 of BIPPCCA when
highest modified probability (the majority pixel locating the maximum likely majority pixel by considering
such that only those majority pixels in which correspond to minority
for all and where is pixels in the constructed patterns, , for which .
also a majority pixel). Replace that pixel, , with In applying this constrained BIPPCCA to mask design, note
a minority pixel. that the patterns composing the set can be
3) Given the new minority pixel, , adjust the proba- constructed in any order and that order does have an impact
bility of each and every majority pixel, , such that: on the construction of each pattern as is constrained by the
constructed pattern corresponding to the maximum gray level
(28) that is less than and constrained by the constructed pattern
corresponding to the minimum gray level which is greater than
where is the minimum wrap-around distance from . While no criteria for choosing an optimal ordering, or even
the majority pixel to the new minority pixel an initial gray level, has been offered, generating patterns in
defined for an array as shown in (29) a random order may offer better results than by constructing
at the bottom of the page. patterns according to consecutive gray levels. As an example,
4) If the ratio of the total number of minority pixels to the when constructing green-noise masks [8], use the interleaved
total number of pixels in is equal to , the desired ordering
gray-level, then quit with the desired dither pattern given
by ; otherwise, continue at step 2. where .

(29)
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 931

Fig. 11. CMYK green-noise dither patterns created using the VBIPPCCA algorithm with: (left) decreased overlapping, (center) uncorrelated overlapping, and
(right) increased overlapping.

B. Color BIPPCCA 2) For , if the ratio of the total number of

minority pixels to the total number of pixels in is
The physical construction of binary dither patterns for the less than , then
multichannel green-noise mask is done through MBIPPCCA
a) Construct the concentration matrix using a
(the Multichannel BInary Pair Correlation Construction Algo- user-defined mapping of , the output
rithm). In color, a binary dither pattern representing the color after filtering with the low-pass filter
is defined by the set of monochrome using circular convolution.
images where is the binary b) Locate the majority pixel in with the
dither pattern corresponding to the th color with intensity . highest modified probability (the majority pixel
MBIPPCCA constructs these monochrome images according to such that
the previous algorithm, but unlike BIPPCCA, when a minority for all and
pixel is added to the th color, the probabilities corresponding to where is also a majority
majority pixels of color are adjusted according to , the pixel), and replace that pixel, , with a
desired pair correlation between minority pixels of colors and minority pixel.
. So for a CMYK dither pattern, for each minority cyan pixel c) Given the new minority pixel, , adjust
added, MBIPPCCA will make use of the user-defined shaping the probability of each and every majority pixel,
functions , , , and to adjust the for , such that
probabilities of majority pixels in the cyan, magenta, yellow, and (30)
black colors, respectively.
where is the minimum wrap-around distance
Because stationarity is also a desired property for digital
from the majority pixel to the new
color halftoning, MBIPPCCA will apply just as in the
minority pixel .
monochrome case with each color filtered independently of
3) If for all colors , the ratio of the total number of minority
the others. Returning to the CMYK case, this implies that
pixels to the total number of pixels in is equal to
the maximum likely majority pixel of the cyan color is the
, the desired intensity of color , then quit with the de-
majority pixel of the cyan color with the highest product
sired color dither pattern given by the set
where is the probability ; otherwise, continue at step 2b.
array for cyan pixels and is the concentration
Like BIPPCCA, the above algorithm is not suited to the design
matrix formed by applying a user-defined mapping to the
of multichannel green-noise masks as the stacking constraint
concentration of minority cyan pixels. will not be satisfied for all patterns. MBIPPCCA must, there-
In summary, the steps for MBIPPCCA are performed as fol- fore, satisfy the same constraints as BIPPCCA in order to be
lows where is the initial set of empty used for mask construction. The first of these two constraints,
arrays. , is satisfied by first initializing to
1) Create a set of arrays, , of uni- and applying step 2c for each and every minority pixel in .
formly distributed random numbers such that MBIPPCCA can then continue at step 3. The second constraint
is the probability that will become a is satisfied in step 2b of MBIPPCCA when locating the max-
minority pixel. imum likely majority pixel in color by considering only those
932 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

Fig. 12. Color plate 1. CMYK dither patterns with: (a) created via Floyd-Steinberg error diffusion, (b)–(f) created via generalized error diffusion, and (g)–(i)
created via VBIPPCCA.

majority pixels in that correspond to minority pixels in the function shapes the pair correlation between pixels of
constructed patterns, , for which . Note that with these different colors. This pair of shaping functions is used to re-
constraints, patterns of the multichannel green-noise mask can duce the amount of overlap between pixels of different colors
be constructed in any order, and that the order, to which patterns [Fig. 12(g)]. Using this same pair of shaping functions, but with
of any color are constructed, need not be the same as any other for all , patterns with no correlation between
color . channels [Fig. 12(h)] can be constructed. To create a pattern
where the overlapping of pixels of different colors is increased,
C. Simulations the function is set to have the same shape as
Before constructing masks, Fig. 10 shows a set of pair cor- [Fig. 12(i)].
relation shaping functions where the function shapes These patterns of Fig. 12(g)–(i), generated by MBIPPCCA,
the pair correlation between pixels of the same color and the were constructed to represent a pixel input image of
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 933

Fig. 13. Color plate 2. CMYK green-noise masks constructed from VBIPPCCA such that: (a) minimizes dot overlap, (b) has uncorrelated overlap, (c) maximizes
overlap, and (d) maximizes overlap for colors cyan and yellow.

constant color with an average strate MBIPPCCA’s ability to capture the same spatial relation-
of 5 pixels per cluster ( pixels). The statistical mea- ships between pixels as those created in Fig. 12(d)–(f) via gen-
sures of the spatial relationships between pixels for these three eralized error diffusion. The key is in the shaping functions, and
patterns are shown in Fig. 11. The results, shown here, demon- through these shaping functions, the same relationships between
934 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 9, NO. 5, MAY 2000

Fig. 14. Color plate 3. CMYK halftoned images of the: (a) original using, (b–d) generalized error diffusion, and (e–h) multichannel green-noise masks.

minority pixels can be encouraged in the design of green-noise pixel overlap are employed in masks (a), (b) and (c), respec-
masks. tively. Mask (d) is a special mask designed more to demon-
Mask design can be seen in Fig. 13 (left) where the three strate the range of possibilities for dither array generation. In
design criteria: 1) decreased; 2) uncorrelated; and 3) increased this instance, the colors cyan and yellow are designed to overlap
LAU et al.: DIGITAL COLOR HALFTONING WITH GENERALIZED ERROR DIFFUSION AND MULTICHANNEL GREEN-NOISE MASKS 935

while not overlapping with black or magenta. The colors black [3] M. Rodriguez, “Graphic arts perspective on digital halftoning,” in Proc.
and magenta are uncorrelated with respect to each other. The SPIE, Human Vision, Visual Processing, Digital Display V, vol. 2179,
B. E. Rogowitz and J. P. Allebach, Eds., Feb. 1994, pp. 144–149.
CMYK-color scales shown in Fig. 13 (right) are given to fur- [4] T. N. Pappas and D. L. Neuhoff, “Printer models and error diffusion,”
ther illustrate the clustering behavior of each mask. By design, IEEE Trans. Image Processing, vol. 4, pp. 66–79, Jan. 1995.
each mask has an average cluster size of 2 pixels at extreme [5] R. Levien, “Output dependant feedback in error diffusion halftoning,”
in IS&T’s 8th Int. Congr. Advances Non-Impact Printing Technologies,
gray levels ( , 1) and an average cluster size of 12 pixels at Williamsburg, VA, Oct. 25–30, 1992, pp. 280–282.
. [6] R. A. Ulichney, “Dithering with blue noise,” Proc. IEEE, vol. 76, pp.
56–79, Jan. 1988.
[7] T. Mitsa and K. J. Parker, “Digital halftoning technique using a blue
V. CONCLUSIONS noise mask,” J. Opt. Soc. Amer., vol. 9, pp. 1920–1929, Aug. 1992.
[8] D. L. Lau, G. R. Arce, and N. C. Gallagher, “Digital halftoning via green-
In summarizing this paper, it is important to note that this noise masks,” J. Opt. Soc. Amer., vol. 16, pp. 1575–1586, July 1999.
paper does not present a process of optimal color reproduc- [9] J. S. Liu and F. H. Cheng, “Color halftoning—A nonseparable model,”
in Proc. Int. Conf. Image Processing, 1996, pp. 561–564.
tion but instead offers two new techniques for getting there. Fu-
[10] M. Yao and K. J. Parker, “Application of the blue-noise mask in color
ture work will look at optimizing the parameters of green-noise halftoning,” in Proc. SPIE Visual Communications Image Processing,
for specific output devices as, again, the advantage to using vol. 2727, R. Ansari and M. J. Smith, Eds., Feb. 1996, pp. 876–880.
[11] N. A. C. Cressie, Statistics for Spatial Data, New York: Wiley, 1983.
green-noise is that it is tunable—allowing for various cluster
[12] P. J. Diggle, Statistical Analysis of Spatial Point Patterns. London,
sizes for various dot-gain characteristics. Previously, techniques U.K.: Academic, 1983.
such as error diffusion with output-dependent feedback and the [13] D. Stoyan, W. S. Kendall, and J. Mecke, Stochastic Geometry and Its
green-noise mask could only be optimized or tuned within a Applications, New York: Wiley, 1987.
[14] R. W. Floyd and L. Steinberg, “An adaptive algorithm for spatial gray-
given channel/color. Now both can consider the interactions of scale,” in Proc. Society Information Display, vol. 17, 1976, pp. 75–78.
the component colors.
Noting Fig. 14(a) where the continuous tone CMYK image
flowers is shown with its corresponding halftone reproductions, Daniel L. Lau received the B.S.E.E. degree with
the generalized error diffusion scheme gives its best reproduc- highest distinction from Purdue University, West
tion in Fig. 14(d) where the amount of overlap is increased rel- Lafayette, IN, in May 1995, and the the Ph.D. degree
in electrical engineering from the University of
ative to the uncorrelated overlap of Fig. 14(c) and the decreased Delaware, Newark, in May 1999.
overlap of Fig. 14(b). In these three instances, the configurations He spent a year signal and image processing
of parameters ( , , , and ) are exactly the same as those at the Lawrence Livermore National Laboratory,
Livermore, CA. His research interests include
of Section III-C, Fig. 7. image processing, digital halftoning, multimedia,
In Fig. 14(e)–(g), the same comparison of overlap is made and nonlinear filters. He is especially interested in
using the multichannel green-noise masks of Fig. 13(a)–(c), applying digital image processing techniques to
applications in art conservation. He has written several papers on nonlinear
respectively, where increased overlap gives the best color filters for signal processing, and has consulted with industry on halftoning and
reproduction. Although these patterns appear “grainy” relative digital printing.
to their error diffused counterparts, this shortcoming is not a
function of masks in general but is a shortcoming of the design
criteria used in the construction of these specific masks. That Gonzalo R. Arce (S’82–M’82–SM’93) received the B.S.E.E. degree with
is, these masks are composed of clusters which are too large highest honors from the University of Arkansas, Fayetteville, in 1979 and the
for the quality of printer being used (Tektronix Phaser 440 M.S. and Ph.D. degrees in electrical engineering from Purdue University, West
Lafayette, IN, in 1980 and 1982, respectively.
dye-sublimation)—resulting in halftone patterns with visually Since 1982, he has been with the Department of Electrical and Computer En-
disturbing artifacts such as the annoying cyan clusters in the gineering, University of Delaware, Newark, where he is currently a Professor
predominantly magenta flower pedals. and Associate Chair. He frequently serves as a Consultant to industry and gov-
ernment in the areas of digital printing, image processing, communications, and
A mask such as that (not pictured) used in Fig. 14(h) is tomography. He has 15 years of research experience in the areas of halftoning
much better suited to this printer as the spatial relationship and digital printing. He has published over a dozen journal and conference pa-
between minority pixels is closer to blue-noise; furthermore, pers in the area of halftoning. His other research interests include robust signal
processing and its applications, communication theory, image processing, and
this mask also takes into account the improved color reproduc- secure multimedia communications.
tion achieved by increasing the overlap of minority pixels of
different colors. As this mask makes good use of the printers
ability to print individual pixels, it is a clear example of the
Neal C. Gallagher (S’72–M’75–SM’85–F’87) is the Dean of Engineering at
tunability of green-noise for color halftoning. Colorado State University. Over the years, he has published in the areas of sto-
chastic processes, quantization and source coding, electromagnetic theory and
microwave components, and optics. He has held various editorial and committee
REFERENCES positions in the IEEE, OSA, and SPIE. He is currently the Chair of the Informa-
[1] J. Sullivan, L. Ray, and R. Miller, “Design of minimum visual modu- tion Processing, Data Storage, and Holography Technical Group of the OSA’s
lation halftone patterns,” IEEE Trans. Syst., Man, Cybern., vol. 21, pp. Technical Committee and thereby a member of its executive committee.
33–38, Jan./Feb. 1991. Dr. Gallagher is a Fellow of the IEEE for his work in nonlinear digital signal
[2] D. L. Lau, G. R. Arce, and N. C. Gallagher, “Green-noise digital processing and a Fellow of the Optical Society of America for his work on
halftoning,” Proc. IEEE, vol. 86, pp. 2424–2444, Dec. 1998. diffractive optics.

View publication stats