Signal Processing - September 2017 - 169

Color correction
Dynamic range mismatches between
the HDR data and display devices, as
previously shown, are typically handled
by tone mappers, focusing on one di-
mension of the color gamut, along the
luminance direction. This generates
two major drawbacks. First, appearance
effects are often ignored, leading to
images which may appear poorly or too
saturated, as shown in Figure 5(b) and (c)
[8]. Second, such a tone mapper may not
guarantee that all the sample values of
the tone mapped image are within the
available target, as shown in Figure 5(d).
Even though the output luminance may
be reproducible by the display, the chro-
minance may fall out of the available
gamut, resulting in clipping of extreme
colors. This clipping may again intro-
duce hue shifts and image defects [9].

Global

(a)

Local

(b)

(c)

FIGURE 4. (a) Global versus (b) local. The global approach results in lost of details in high contrast
regions. (c) The HDR frame is tone mapped for display purposes.
HDR Input Image
100
90
80
70
(a)
Reduced Saturation

Lightness

its surrounding into account; with that,
local operators can take advantage of
known effects of the human visual sys-
tem such as local eye adaptation to the
luminance. While the former is simple
and efficient, it may fail to reproduce
details in high contrast image regions
[see Figure 4(a)]. Although the latter can
reproduce details in such regions better
[see Figure  4(b)], it often comes at the
cost of increased complexity and com-
putational time; it may also introduce
artifacts around edges.
Despite this classification, we may
also categorize the tone mappers based
on their intent. Three main categories of
tone mappers can be identified, based on
the visual system, for scene reproduction
and for best subjective quality. The first
aims at integrating mechanisms, which
simulate various aspects of the human
visual system, into the tone mapper. This
includes glare, luminance and chromatic
adaptations, night vision, etc. The second
category attempts to reproduce the best
match in color gamut and dynamic range
available for the display on which the
image will be visualized. This is achieved
through the preservation of the original
scene's appearance. The last category
produces images with the most preferred
subjective quality. Typical examples are
operators with parameters that can be
adjusted to achieve a specific artistic goal.

60
50
40
30
20
10
0

(b)

0

Increased Saturation

20

40

60
Chroma

80

100

Gamut Boundaries
L Compression
C Compression
Gamut Clipping
(c)

(d)

FIGURE 5. Tone mapper drawbacks. (a) The HDR input image and changes in appearance due to either
(b) a reduction or (c) an excessive saturation (images courtesy of Francesco Banterle). (d) Pixels may
)
be within the destination gamut only for the lightness channel L ; however, their chroma channel may
still be out of gamut. Here the HDR input image has been tone mapped for display purposes (image
courtesy of Tania Pouli).

To improve the saturation of the
tone mapped image, a simple solution
is to introduce an adjustable parameter
that allows the overall saturation of the
tone mapped image [10] to be controlled.
In the following, let p be a parameter in
[0, 1], then
IEEE SIGNAL PROCESSING MAGAZINE

|

September 2017

|

p
It = c Io m Lt .
Lo

(3)

Here I o is the input HDR image, I t
is the final output tone mapped image
(both in RGB values), L o is the lumi-
nance of the original HDR image, and
169
Table of Contents for the Digital Edition of Signal Processing - September 2017
