Image Processing Reference
In-Depth Information
Gamut mapping
algorithm (e.g., soft
gamut mapping)
T
Mapped
L
*
a
*
b
*
Merit
function
3
L
*,
a
*,
b
*
Optimization
algorithm
FIGURE 7.62
Merit-based gamut mapping with closed-loop optimization.
candidate gamut-mapping function. A merit function is constructed based on a high-
level gamut-mapping strategy with each cluster. At least one merit function is asso-
ciated with each cluster. Gamut-mapping parameters are iteratively varied (or tuned)
to optimize their values. The merit-based gamut mapping uses a closed-loop opti-
mization technique (Figure 7.62) to iteratively tune various mapping parameters
associated with the gamut-mapping algorithm. A number of different candidate
algorithms are iterated for each de
ned cluster within the color space. Once optimal
mapping parameters are obtained for each candidate, a best (that minimizes the merit
function) is picked for that cluster. The optimized gamut-mapping functions thus
obtained for adjacent clusters are then blended together using multidimensional
smoothing algorithms or other methods described in Ref. [134] to generate a compos-
ite smooth function that collectively exploits the local advantages of each cluster. The
entire process is done automatically in a computer program.
For illustration of this technique we can choose the soft gamut-mapping
approach, introduced in Section 7.6.3, with two gamut-mapping parameters,
b
and
a
(or x
0
in the centroid clipping algorithm). Other parameters may be used depending
on the gamut-mapping algorithm. In Figure 7.62, a transformation block
maps
the out-of-gamut L*a*b* values to a 1-D function for each color point. This could
be, for example,
''
T
''
D
E
2000
formula, which is to say that a
D
E
2000
mapping is being
emulated. The
D
E
2000
function calculates distances between the out-of-gamut
L*a*b* values and the mapped L*a*b* values for each node color of the current
cluster. The merit function is selected based on what is required by the device
designer during optimization. A mean-squared error function can be used as the
merit function. In this case, the merit function determines the mean-squared error of
the values calculated using the
D
E
2000
function for each node color belonging to
a particular cluster that needs to be mapped. The output of the merit function is a
single value which represents the numerical merit of the mapping parameters. An
optimization block manipulates the gamut-mapping parameters during successive
iterations to generate the best, or optimum, merit function value. The optimization
algorithm could be a brute-force approach, that selects an exhaustive set of combin-
ations of
within their limits. A brute-force approach in many instances does
not take much time to generate the
b
and
a
final results given present processor speeds. When
the optimum merit function value is reached (or exceeded) the iteration loop ends for
the cluster being processed. Note that the above calculation can be done for more
than one gamut-mapping algorithm for each cluster, and the one with the best merit
function output can be chosen.
Search WWH ::
Custom Search