Image Processing Reference
In-Depth Information
6.4.6 C OMPARISON OF D IFFERENT I NVERSION A LGORITHMS
We now brie
y compare four different mapping algorithms when used for building the
inverse map. We modeled a legacy Xerox color digital printer by creating a 13 3 LUT.
The LUT is a mapping from a device-dependent CMY to a device-independent L*a*b*
color space. The grid points in the CMY color space are uniformly sampled. The inverse
map is generated by uniformly sampling the L*a*b* space and
finding the inverse of
each L*a*b* node on the grid using one of the following algorithms: (a) Shepard, (b)
moving matrix, (c) ICI, and (d) CG. Out-of-gamut colors were
rst mapped to the
nearest point in the gamut by a technique that maintains the hue angle.
The quantitative metric used to evaluate the inversion accuracy is the
D E a *
accuracy for interpolated colors between the mapped input L*a*b* and the resulting
output L*a*b*. For 240 in-gamut test colors (Figure 6.15), the
D E a * statistics are
simulated. For this simulation, a 13 3 printer forward LUT size was chosen. The 240
in-gamut colors are provided as input to the inverse LUT and the output of the
inverse LUT is interpolated through the forward LUT. The interpolation in the
forward LUT uses a trilinear algorithm, which is common to all methods. The output
of the interpolation is compared through the input to get the error performance. The
error results are shown in Table 6.11.
FIGURE 6.15 (See color insert following page 428.)
View of 240 test colors used for
comparing inversion algorithms.
TABLE 6.11
Comparison of Four Different Printer Inverse Algorithms
Standard
Deviation Mean D E a * รพ2 s Minimum D E a * Maximum D E a *
Mean D E a *
Algorithm
Shepard
1.644
0.5448
2.7346
0.151
3.807
Moving matrix
1.149
0.809
2.764
0.023
5.923
CG
0.187
0.5243
1.236
0.001
5.113
ICI
0.044
0.1182
0.2812
0.002
1.681
Search WWH ::




Custom Search