525-532i.

qxd 10/18/02 10:04 AM Page 525

Color Space Conversion

for the Laser Film
Recorder using 3-D LUT
By Ado Ishii

W
A new method of color space conversion ith the conventional film recording process
based on a one-dimensional (R, G, and B
for the laser film recorder is proposed.
independent) LUT, a reproduced image on
The input image file (10 bits or more, RGB) print films generally has problems in colorimetry unless
is converted by a three-dimensional color correction is performed prior to printing. Errors
lookup table (3-D LUT), then fed to the film occur not only in hue, but more seriously in lightness
recorder as data from 10-bit RGB. Using and saturation. This paper first describes the two color
this method, the image on print films is gamuts for CRT monitors and print films. The conven-
colorimetrically improved and matched to tional solution requires an awkward trial-and-error
process for color correction, where the image data is
the image on a CRT monitor. This paper
corrected little by little, and multiple revisions of the
reviews a color transformation by conven- print are repeatedly returned from the labs to the cre-
tional film recording that uses only one- ators for evaluation and approval. It is a time-consum-
dimensional LUT. A 3-D LUT is built based ing process; worse, the print quality is poor.
on a gamut mapping algorithm to compen- Meanwhile, the rendering performance of Computer-
sate for the difference of the color gamut Generated Image (CGI) has greatly advanced in the
between CRT monitors and print films, last decade. Creators of CGIs and feature animation
employing a new soft-clipping method. inevitably prefer using subtler expressions in colors,
ray tracings, shading effects, etc. Consequently, final-
Three gamut compression (clipping) meth-
izing images on a CRT monitor has become indispens-
ods are also evaluated. The calculation on able, leading to a strong demand for establishing a
the 3-D LUT is based on CIE LAB color color conversion method for the film recorder. A suc-
coordinates evaluated by several cessful study will greatly contribute to the expansion of
observers. A guideline for quality control artists’ creativity as well as the capability of film.
of color management in film laboratories A new concept in color science is introduced here,
is also discussed. This new system has and several experiments conducted to arrive at the
conversion method are explored. The data flow in film
been successfully applied to the produc-
production is shown in Fig. 1, where 1-D LUT primarily
tion of feature animation footage. retains the tone reproduction, and a 3-D LUT maps the
input onto a suitable output gamut. The process is a
three-dimensional mapping where hue, lightness, and
saturation are maintained. The system LUT embedded
in the film recorder also plays a part in the density

SMPTE Motion Imaging, November 2002 • www.smpte.org 525

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 526

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

Figure 1. Workflow in the post-production.

reproduction, namely, conversion of a 10-bit log signal

into the target density. This LUT is not shown in Fig. 1,
but is regarded as a portion of the film recorder.
The use of the color difference ⌬E*ab is not conve-
nient for evaluating reproduced images, because mini-
mizing ⌬E*ab means clipping out-of-gamut elements in
a gamut boundary, resulting in loss of details. This
paper uses ⌬E*ab to monitor conversion patterns,
however, image is evaluated mainly on a psychophys-
ical basis. In short, the 3-D LUT is created using the
gamut mapping algorithm and visually evaluated by
several observers.

Color Space on Print Film and CRT

Monitor Figure 2. Density reproduction to 10-bit log code point.
A colorimetric comparison was done: Theoretically,
approximately equivalent to the color temperature of
the color reproduction system of CRT monitors is an
5400oK. Note that the tone reproduction is supported
RGB additive method, in film it is a subtractive method
by the primary 1-D LUT shown in Fig. 1; the embedded
by CMY dyes. Pragmatically, the color reproduction of
1-D LUT maintains the density reproduction. The white
film is more complex, due to secondary absorption and
point given by 8-bit code point of R 255, G 255, and B
the inter-image effect.1,2
255 takes a positive density of approximately G 0.20
In CGI productions, the most common and conve-
(status A). The white point is used as the reference
nient way of rendering images is to use 0.45 gamma
white, i.e., standardized Y=100.
for image files and 2.2 gamma for CRT display. A
Comparing colorimetry, CIE LAB (1976) values are
Barco CID421 monitor was set up with 2.2 gamma and
calculated using the following equations. D55 white
a color temperature of 5500oK. A total of 512 colors
point is applied into Xn, Yn, and Zn.
were measured in CIE XYZ with the Minolta CS-100
colorimeter.
One thousand colors in 10-bit log were recorded and
measured in CIE XYZ with X-rite DTP41/T on the print
(1)
film. The intermediate stock used was Eastman Kodak
5244 and the print stock used was Eastman Kodak
Vision 2383. The film recorder was set up with a sys-
tem LUT, designed to use the relationship between 10-

再
bit codes and status M densities (Fig. 2). Here, a gray (2)
(p)1/3 p > 0.008856
patch with neutral 10-bit code point of R 480, G 480, B f(p) =
7.787p + 16/116 otherwise
480 is interpreted as the status M density of R 1.00, G
1.46, B 1.59. The print for the gray patch is believed to The color difference is defined as Euclidean distance
be the status A density of R 0.91, G 0.89, B 0.87, as

526 SMPTE Motion Imaging, November 2002 • www.smpte.org

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 527

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

⌬E*ab = { (⌬L*)2 + (⌬a*)2 + (⌬b*)2 }1/2 (3) is popular. Therefore, it is important to know how heav-
ily ⌬E*ab depends on the densitometric variance in
Also, the color difference in polar coordinates is daily laboratory operations. For that purpose, the fol-
defined as lowing experiment was carried out:
⌬E* = { (⌬L*)2 + (⌬C*ab)2 + (⌬⌯*)2 }1/2 (4) First, several film strips with multisteps of gray
patches were recorded by slightly shifting the density
where chroma C*ab is defined as
using the system LUT of the film recorder, which is
C*ab = { (a*)2 + (b*)2 }1/2 (5) basically designed with the characteristic shown in
Fig. 2. Adding 10-bit code point 30 to the R value of
The L-C plots, the lightness on the Y-axis, and the recording increases the density (status M) by approxi-
chroma on the X-axis for the three primary colors and mately 0.06. The proportion of variance for positive
three secondary colors are shown in Fig. 3, where the density (status A) to that for negative density (status
gamut boundary for the CRT and the print film are M) versus the 10-bit log code is shown in Fig. 5,
plotted. Note that the white point with 8-bit code point where the LUT was fixed for G and shifted ±30 code
of R 255, G 255, and B 255, which gives L* = 100, is point in 10-bit for R and B. It shows that the gamma in
mapped to the 10-bit code point of ~710, while the mid-density is comparatively high, and others are
laser film recorder has a greater density range with lower for the “toe” and “shoulder” regions. The sam-
linearity. The gamut boundary was plotted here with a pled positive was measured in CIE XYZ, and ⌬E*ab
maximum code point of 810. It is shown that the print was calculated between the output of the proper sys-
film has a narrower gamut in red, green, blue, and tem LUT and the shifted LUT. The proportion of
magenta. Moreover, several points in the same hue ⌬E*ab to the variance of positive density versus the
(green) that were measured for the source and repro- 10-bit log code is shown in Fig. 6. Multiplying the two
duced on the film are shown in Fig. 4. The points are plotted curves (Figs. 5 and 6) shows the trend of
converted only with 1-D LUT, which preserves the ⌬E*ab versus the 10-bit log code (Fig. 7). It is obvious
lightness, L*. It reveals that in the green hue, the that ⌬E*ab is more sensitive in the mid-gray region,
reproduced colors are darker and less saturated. Be and a variation of the density 0.01 (status M) at 420
aware that the hue shift is ignored in Fig. 4. In post- (10-bit log) becomes a ⌬E*ab of 1.84.
production, the color shift has been known as a The result reveals that the variance of the negative
nature of film recording (Fig. 4). Thus, in the color density for the mid-gray, around the LAD gray, is more
matching process of CRT to film, two factors, the sensitive, so the density for the region should be moni-
gamut difference and the mapping algorithm, become tored more carefully. For other density regions, espe-
most important. cially brighter gray and white, the positive density
becomes a more sensitive parameter. To manage the
Quality Control in Film Laboratories value of ⌬E*ab within a variance range of 3, the vari-
In the film recording process, several film frames of ance of the negative density should be controlled with-
density checker are used to control quality. Checker in a range of D 0.02 (status M) and the variance of the
frames monitor the daily variance caused by the positive density within D 0.03 to 0.04 (status A) for
recorder, minor emulsion changes, and the chemical brighter gray (above 550 code point of 10-bit).
process. Separately, in the development process,
chemical analysis is performed daily and the densito- Evaluation on Tone Reproduction
metric variance is usually checked twice a day. As mentioned earlier, in addition to the system LUT
Eastman Kodak Co. offers the EK Imagecare Program embedded in the recorder, another 1-D LUT is created
(Imagica Corp. has been approved for it), which pro- to maintain the tone reproduction. Several images pro-
vides guidelines for controlling chemical process and vided by Studio Ghibli were chosen for evaluation.
densitometric variations. When the LUT was set to provide an overall gamma of
In the post-production process, namely negative and 1.0, which corresponds to the display gamma of 2.2,
positive developing and printing, densitometric analysis the constrast of the reproduced image was poor. This

SMPTE Motion Imaging, November 2002 • www.smpte.org 527

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 528

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

Figure 3. Gamut in specific hues.

528 SMPTE Motion Imaging, November 2002 • www.smpte.org

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 529

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

lack of contrast and milkiness in the dark area were

noticed by observers. Accordingly, a psychophysical
test was performed changing the overall gamma
between 1.00 and 1.45, by 0.1/2.2 step. Observers first
saw the reference images on a CRT monitor in a dark
room, and about 10 min later the film projection start-
ed. The projector was adjusted with white luminance of
54.8 cd/m2 for open gate, and a color temperature of
5600 oK. It was concluded that an overall gamma
between 1.08 and 1.17 gives the best contrast and is,
therefore, cinematic.

Building 3-D LUT

The two color gamuts for CRT and film are different
(Fig. 3). It is not possible to reproduce the out-of-
Figure 4. Color transformation from CRT to film. gamut colors on the film, therefore, it is important to
map these colors to the available gamut color in a bet-
ter way. More experiments were performed to create a
three-dimensional lookup table for better mapping.
Here the 3-D LUT is intended to bridge the two color
gamuts in the CIE LAB color space. It must convert 10
or more bits of log RGB data into 10-bit log RGB data.
Two questions about the design of the 3-D LUT were
considered: the first regarded which direction the
CRTs gamut should be mapped; the second con-
cerned clipping.
In the last two decades, researchers proposed vari-
ous gamut-mapping algorithms for the application of
CRT monitors and hardcopy printers.3-5 Typical algo-
rithms are shown in Fig. 8. All out-of-gamut colors are
mapped to the nearest color on the boundary of the
Figure 5. Dependency of status A on status M density.

Figure 7. Dependency of ⌬E*ab on status M density (vari-

Figure 6. Dependency of ⌬E*ab on status A density (D0.01). ance of D0.01).

SMPTE Motion Imaging, November 2002 • www.smpte.org 529

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 530

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

Figure 8. Gamut mapping direction.

destination gamut (Fig. 8(a)). This minimizes ⌬E*ab,

although any and all details in the out-of-gamut colors
are lost. Others map colors towards the points along
the achromatic (lightness) axis, examples being the
point L*=50 in Fig. 8(b) and the point that has the
same lightness as the source gamut cusp (CUSP algo-
rithm) in Fig. 8(c).
Another concern is gamut compression. An experi-
ment was done to evaluate the quality of three types of
gamut compression: i.e., clipping, linear compression,
and soft clipping (Fig. 9). For interpolation of soft clip-
ping, a quadratic curve was applied. As a gamut map-
ping direction, the cusp algorithm (Fig. 8(c)) was
employed, where the original hue was maintained for
reproduction. The same images chosen for evaluation
of tone reproduction were used, following the same Figure 9. Gamut compression.

procedure. Observers saw the ref-

Table 1—30 Colors in Reference Color Checker erence images first on a CRT moni-
L C H L C H
tor in a dark room, and about 10
min later the film projection was
shown.
1 58.88 104.30 41.57 19 76.86 81.19 69.19
2 85.72 113.46 140.13 20 90.67 92.16 120.67 The 3-D LUT designed for this
3 30.57 125.89 301.43 21 86.57 71.78 169.44 paper has 16 x 16 x 16 grids. Each
4 88.55 59.36 197.55 22 62.88 55.91 254.35
5 63.91 103.68 329.57 23 47.57 109.16 313.81
grid has 10-bit code values for RGB
6 97.49 88.46 100.19 24 60.02 81.75 6.75 that are regressively calculated.
7 77.00 47.46 42.33 25 82.72 26.07 66.24 However, as concluded in the eval-
8 90.98 56.17 139.70 26 60.43 36.68 27.58
9 63.72 62.43 300.57 27 60.23 47.14 151.90 uation of tone reproduction, light-
10 92.95 30.09 199.13 28 53.69 39.74 240.57 ness is not preserved because the
11 80.42 51.81 328.14 29 33.10 39.04 67.24
12 97.94 45.53 102.38 30 60.33 55.25 92.70
overall gamma is somehow modi-
13 29.16 52.02 41.24 fied. Actually, lightness is modified
14 42.50 55.35 140.16 to reflect the overall gamma of 1.08
15 14.04 60.55 298.34
16 43.18 30.01 197.61 to 1.17. Practically, for the regres-
17 31.03 50.55 329.40 sion, the chroma, hue, and light-
18 47.72 43.54 97.59
ness are first mapped to the desti-

530 SMPTE Motion Imaging, November 2002 • www.smpte.org

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 531

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

nation gamut, then only the light-

ness (L*) is recalculated.

Result of Comparison
Five observers judged that clip-
ping causes serious loss of detail
and poor image quality. Also, the
image mapped by linear compres-
sion lacks saturation and bright-
ness. While these methods are not
acceptable, the use of soft clipping
drastically improves saturation and
brightness and less image detail is
lost. All the images processed by
Figure 10 (a). ⌬E*ab evaluation for 30 colors
soft clipping were of good quality
with appearances similar to the
original image on the CRT monitor.
For reference evaluation, 30 col-
ors were chosen and measured.
Table 1 shows the original coordi-
nates in the CIE LAB. These colors
were transformed by the combined
use of a conventional 1-D LUT,
which provided an overall gamma
of 1.17, and a 3-D LUT for soft clip-
ping of gamut mapping. The result
of the measurement in the CIE
LAB is shown in Fig. 10, also, the
transformation of the hue is shown
in Table 2. Figure 10 illustrates that
Figure 10 (b). Evaluation in lightness for 30 colors. the conversion by the 3-D LUT
improves ⌬E*ab for all 30 colors,
but, as mentioned above, L* is not
intended to be preserved according
to the observers' decision. Note
that the original 30 colors include
out-of-gamut colors, however, as
shown in Table 3, the Macbeth
ColorChecker does not include
these colors. In other words, colors
simulated with the Macbeth Color-
Checker are not sufficient to
describe the color transformation
by mapping. Therefore, we de-
signed our own CGI color checker
with 30 colors. Due to out-of-gamut
Figure 10 (c). Evaluation in chroma for 30 colors. colors, rather than ⌬E*ab evaluation,

SMPTE Motion Imaging, November 2002 • www.smpte.org 531

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.
525-532i.qxd 10/18/02 10:04 AM Page 532

COLOR SPACE CONVERSION FOR THE LASER FILM RECORDER USING 3-D LUT

be maintained strictly to achieve

Table 2—Hue Transformation for 30 Colors (degrees) accurate color transformation.
CRT 1-D LUT 3-D LUT CRT 1-D LUT 3-D LUT Acknowledgments
1 41.57 38.81 40.97 19 69.19 68.94 68.84 I am indebted to Shingo Toyotani
2 140.13 148.82 148.82 20 120.67 104.50 120.31
for his continued encouraging
3 301.43 308.86 306.12 21 169.44 200.32 176.11
4 197.55 222.46 194.03 22 254.35 257.97 260.74 advice in technical aspects. Thanks
5 329.57 331.04 330.12 23 313.81 318.36 314.67 also go to Kohei Endo and Hiroaki
6 100.19 87.89 100.15 24 6.75 356.37 5.96
7 42.33 36.69 40.58 25 66.24 52.04 60.21 Hirabayashi for their support in the
8 139.70 143.30 142.02 26 27.58 14.72 25.77 experiments.
9 300.57 298.16 298.23 27 151.90 176.04 161.04
10 199.13 207.63 197.95 28 240.57 245.26 238.16 References
11 328.14 326.80 327.79 29 67.24 71.36 67.74
12 102.38 94.50 100.96 30 92.70 84.82 96.30 1. R.W.G. Hunt, The Reproduction of
13 41.24 39.87 38.06 Colour, Fifth Ed., Ch. 9, Fountain
14 140.16 150.29 140.96 Press, 1995.
15 298.34 299.95 303.70 2. N. Ohta, "Interrelation Between
16 197.61 227.26 199.03 Spectral Sensitivities and Interimage
17 329.40 322.19 323.23 Effects," J. Imaging Science, 35, 94-
18 97.59 89.68 103.30 103, 1991.
3. J. Morovic and M. R. Luo, "Gamut
Mapping Algorithm Based on
Table 3—Colors at 3rd Row of Macbeth Color Checker Psychophysical Experiment," Proc.
of the 5th IS&T/SID Color Imaging
L a b C H Conf., pp. 44-49, 1997.
4. N. Katoh, M. Itoh, and S. Ohno,
Blue 29.66 29.15 -62.98 69.40 294.84 "Three-dimensional Gamut Mapping
using Various Color Difference
Green 55.48 -39.03 25.47 46.60 146.87
Formulae and Color Spaces," J.
Red 41.22 53.25 19.32 56.65 19.94 Electron Imaging, 8:4, 365-379,
Yellow 81.35 -0.69 71.03 71.04 90.56 1999.
Magenta 51.61 51.61 -27.86 58.65 331.63 5. P. G. Herzog and H. Büring,
"Optimizing Gamut Mapping:
Cyan 51.61 -19.55 -39.70 44.25 243.78
Lightness and Hue Adjustments,"
Proc. of the 7th IS&T/SID Color
Imaging Conf., pp. 160-166, 1999.

it requires psychophysical observation as a more effec-

tive and reliable means. On the other hand, improved
color coordinates, e.g., CIECAM97s, and an improved
THE AUTHOR
color difference formula, CIE94, are proposed.
Continued studies are necessary to establish a system- Ado Ishii is a senior research and development
atic method of color evaluation. engineer in the engineering solutions division at
Imagica Corp., Tokyo, Japan, where he develops
Conclusion color management technologies for input and out-
As a result of the experiments, a 3-D conversion LUT put devices for use in post-production facilities. He
was successfully developed and the production quality received a B.E. degree in electrical and informa-
of CRT-to-film conversion was improved. It has been tion engineering from Yokohama National
established that the cusp algorithm is effective as a University, Japan, in 1987. In the same year, he
gamut mapping direction. Also, as a gamut compression joined Imagica Corp. and has since worked on
method, soft clipping demonstrably holds the details of hardware design of image processors for HD
images and improves lightness, saturation, and hue. video and high-resolution film scanner.
Additionally, it has been shown that chemical and opti-
cal processes on mid-grays in post-production should

532 SMPTE Motion Imaging, November 2002 • www.smpte.org

Authorized licensed use limited to: JAY CHENG. Downloaded on July 11,2023 at 09:26:08 UTC from IEEE Xplore. Restrictions apply.