Image display algorithms for high- and low-dynamic-range display devices

Erik Reinhard Abstract — With interest in high-dynamic-range imaging mounting, techniques for displaying such
Timo Kunkel images on conventional display devices are gaining in importance. Conversely, high-dynamic-range
Yoann Marion display hardware is creating the need for display algorithms that prepare images for such displays. In
Jonathan Brouillat this paper, the current state of the art in dynamic-range reduction and expansion is reviewed, and in
particular the theoretical and practical need to structure tone reproduction as a combination of a
Rémi Cozot
forward and a reverse pass is passed.
Kadi Bouatouch
Keywords — High-dynamic-range imaging reduction, expansion, tone mapping, inverse tone
mapping.

range (HDR) image acquisition has already created the

1 Introduction
Real-world environments typically contain a range of illumi-
nation much larger than can be represented by conventional
8-bit images. For instance, sunlight at noon may be as much
as 100 million times brighter than starlight.1,2 The human
visual system is able to detect four or five log units of illumi-
nation simultaneously and can adapt to a range of about ten
orders of magnitude over time.2
On the other hand, conventional 8-bit images with val-
ues between 0 and 255 have a useful dynamic range of
around two orders of magnitude. Such images are repre-
sented typically by 1 byte/pixel for each of the red, green,
and blue channels. The limited dynamic range afforded by
8-bit images is well-matched to the display capabilities of
CRTs. Their range, while being larger than two orders of
magnitude, lies partially in the dark end where human vision
has trouble discerning very small differences under normal
viewing circumstances. Hence, CRTs have a useful dynamic
range of 2 log units of magnitude. Currently, very few dis-
play devices have a dynamic range that significantly exceeds
this range.
The notable exception are LCDs with an LED back
panel where each of the LEDs is separately addressable.3,4
With the pioneering start-up BrightSide being taken over by
Dolby, and both Philips and Samsung demonstrating their
own displays with spatially varying backlighting, hardware
developments are undeniably moving towards higher dynamic
ranges.
It is therefore reasonable to anticipate that the variety
in display capabilities will increase. Some displays will have
a much higher dynamic range than others, whereas differ-
ences in mean luminance will also increase due to a greater
variety in backlighting technology. As a result, the burden on
general-purpose display algorithms will change. High-dynamic-
need for tone-reproduction operators, which reduce the
dynamic range of images prior to display.5,6 The advent of
HDR display devices creates a need for tone-reproduction
operators to expand the dynamic range of images such that
traditional 8-bit images can be displayed appropriately on
HDR displays. In particular, the enormous number of con-
ventional 8-bit images may have to be expanded in range
prior to display on such devices. Algorithms for dynamic-
range expansion are commonly called inverse-tone-repro-
duction operators.
In this paper, we survey the state of the art in tone
reproduction as well as inverse tone reproduction. In addi-
tion, we will summarize desirable features of tone reproduc-
tion and inverse-tone-reproduction algorithms.
2 Dynamic-range reduction
Capturing the full dynamic range of a scene implies that in
many instances the resulting HDR image cannot be directly
displayed on a conventional display, as its range is likely to
exceed the two orders of magnitude range afforded by con-
ventional display devices. Figure 1 (top) shows an image of
a scene with a dynamic range far exceeding the capabilities
of a conventional display. By capturing the full dynamic
range of this scene, followed by tone mapping the image, an
acceptable rendition of this scene may be obtained (Fig. 1,
bottom).
A simple compressive function would be to normalize
an image (see Fig. 2, top). This constitutes a linear scaling
which is sufficient only if the dynamic range of the image is
slightly higher than the dynamic range of the display. For
images with a significantly higher dynamic range, small
intensity differences will be quantized to the same display
value such that visible details are lost. For comparison, the

Extended revised version of a paper presented at the SID Symposium, Seminar & Exhibition (SID ‘07) held in Long Beach, California, May 20–25, 2007.
E. Reinhard is with the University of Bristol, Department of Computer Science, Merchant Venturers Bodg., Woodland Rd., Bristol N/A BS8 1UB,
U.K.; telephone +44-117-954-5140, fax –5208, e-mail: reinhard@cs.ucf.edu. He is also with IRISA, Campus de Beaulieu, France, and the University
of Central Florida, Orlando, FL.
T. Kunkel is with the University of Bristol, U.K.
Y. Marion, J. Brouillat, R. Cozot, and K. Bouatouch are with IRISA, Campus de Beaulieu, Rennes, France.
© Copyright 2007 Society for Information Display 1071-0922/07/1512-0997$1.00

Journal of the SID 15/12, 2007 997

FIGURE 1 — With conventional photography some parts of the scene
may be under- or over-exposed (top). Capture of this scene with nine
exposures, and assemblage of these into one HDR image followed by
tone reproduction, affords the result shown on the bottom.
bottom image in Fig. 2 is tone-mapped non-linearly showing
detail in both the light and dark regions.
In general, linear scaling will not be appropriate for
tone reproduction. The key issue in tone reproduction is
then to compress an image while at the same time preserving
one or more attributes of the image. Different tone-repro-
duction algorithms focus on different attributes such as con-
trast, visible detail, brightness, or appearance. Ideally,
displaying a tone-mapped image on a low-dynamic-range
display device would recreate the same visual response in
the observer as the original scene. Given the limitations of
display devices, this is, in general, not achievable, although
we may approximate this goal as closely as possible.
3 Spatial operators
In the following sections we discuss tone-reproduction
operators which apply compression directly on pixels. Often
global and local operators are distinguished. Tone-repro-
duction operators in the former class change each pixel’s
luminance values according to a compressive function
998 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
FIGURE 2 — Linear scaling of HDR images to fit a given display device
may cause significant detail to be lost (top). For comparison, the bottom
image is tone-mapped, allowing details in both bright and dark regions
to be visible.
which is the same for each pixel.7–13 The term global stems
from the fact that many such functions need to be anchored
to some values that are computed by analyzing the full
image. In practice, most operators use the geometric aver-
age Lv over the N image pixels to steer the compression:
F 1 log d + L (x, y) I .
Lv = exp GH N xÂ,y b v gJK (1)
The small constant δ is introduced to prevent the average
to become zero in the presence of black pixels. The lumi-
nance of each pixel is indicated with Lv, which can be com-
puted from RGB values if the color space is known. If the
color space in which the image is specified is unknown, then
the second best alternative would be to assume that the
image uses sRGB primaries and white point, so that the
luminance of a pixel is given by
Lv (x, y) = 0.2125R(x, y) + 07154
. G(x, y) + 0.0721B(x, y).
The geometric average is normally mapped to a pre-
defined display value. The main challenge faced in the design
of a global operator lies in the choice of compressive func-

FIGURE 3 — A local tone-reproduction operator (left) and a global
tone-reproduction operator (right).14 The local operator shows more
detail, as, for instance, seen in the insets.
tion. Many functions are possible, which are for instance
based on the image’s histogram 12 (Section 4.2) or on data
gathered from psychophysics (Section 4.3).
On the other hand, local operators compress each
pixel according to a specific compression function which is
modulated by information derived from a selection of neigh-
boring pixels, rather than the full image.14–25 The rationale
is that the brightness of a pixel in a light neighborhood is
FIGURE 4 — Halos are artifacts commonly associated with local tone-reproduction operators. Chiu’s operator is used here without
smoothing iterations to demonstrate the effect of division (left).
different than the brightness of a pixel in a dark neighbor-
hood. Design challenges for local operators involve choos-
in g th e co mpressive function, the size of the local
neighborhood for each pixel, and the manner in which local
pixel values are used. In general, local operators are able to
achieve better compression than global operators (Fig. 3),
albeit at a higher computational cost.
Both global and local operators are often inspired by
the human visual system. Most operators employ one of two
distinct compressive functions, which is orthogonal to the
distinction between local and global operators. Display val-
ues Ld(x, y) are most commonly derived from image lumi-
nances Lv(x, y) by the following two functional forms:
Lv (x, y)
Ld (x, y) = , (2)
f (x, y)
Lnv (x, y)
Ld ( x, y) = n . (3)
Lv (x, y) + g n (x, y)
In these equations, f(x, y) and g(x, y) may either be
constant or a function which varies per pixel. In the former
case, we have a global operator, whereas a spatially varying
function results in a local operator. The exponent n is a con-
stant which is either fixed or set differently per image.
Journal of the SID 15/12, 2007 999
 Equation (2) divides each pixel’s luminance by a value
derived from either the full image or a local neighborhood.
As an example, the substitution f(x, y) = Lvmax/255 in Eq. (2)
yields a linear scaling such that values may be directly quan-
tized into a byte and can therefore be displayed. A different
approach would be to substitute f(x, y) = Lvblur(x, y), i.e.,
divide each pixel by a weighted local average, perhaps obtained
by applying a Gaussian filter to the image.15 While this local
operator yields a displayable image, it highlights a classical
problem whereby areas near bright spots are reproduced
too dark. This is often seen as halos, as demonstrated in
Fig. 4.
The cause of halos stems from the fact that Gaussian
filters blur across sharp contrast edges in the same way that
they blur small and low contrast details. If there is a high
contrast gradient in the neighborhood of the pixel under
consideration, this causes the Gaussian blurred pixel to be
significantly different from the pixel itself. By using a very
large filter kernel in a division-based approach such large
contrasts are averaged out, and the occurrence of halos can
be minimized. However, very large filter kernels tend to
compute a local average that is not substantially different
from the global average. In the limit that the size of the filter
kernel tends to infinity, the local average becomes identical
to the global average and therefore limits the compressive
power of the operator to be no better than a global operator.
Thus, the size of the filter kernel in division-based operators
presents a trade-off between the ability to reduce the dynamic
range and the visibility of artifacts.
4 Global tone-reproduction operators
While linear scaling by itself would be sufficient to bring the
dynamic range within the display’s limits, this does not typi-
cally lead to a visually pleasant rendition. In addition to
compressing the range of values, it is therefore necessary to
preserve one or more image attributes. In the design of a
tone-reproduction operator, one is free to choose which
image attribute should be preserved. Some of the more
common examples are discussed in the following.
4.1 Brightness matching
Tumblin and Rushmeier have argued that brightness, a sub-
jective visual sensation, should be preserved.8 Given the
luminance Lv of a pixel, its brightness B can be approxi-
mated using
g
B = 0.3698
FG L IJ
v
, (4)
HL K
A 4.2
where LA is the adapting luminance, γ models the visual
system’s non-linearity, and the numeric constant is due to
fitting the function to measured data. To preserve bright-
ness, the brightness reproduced on the display Bd should be
1000 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
matched to the scene brightnesses Bv. This leads to an ex-
pression relating scene and display luminances Lv and Ld:
g (Lv ) g (Ld )
FL I v
L d = L d, A GH L JK
v, A
, (5)
where Ld,A and Lv,A are the adapting luminances for the
display and the scene, respectively. To account for the fact
that it may be undesirable to always map mid-range scene
values to mid-range display values, this expression is multi-
plied with a correction term.8 The γ( ) function is essentially
the logarithm of its parameter:
R|2.655 if L > 100 cd/m 2 U|
g(L) = S
|T1.855+0.4log eL + 2.3 ◊10 j otherwiseV|W
-5
. (6)
10
The key observation to make is that matching bright-
nesses between scene and display leads to a power function,
with the exponent determined by the adapting luminances
of the scene and the display. This approach therefore consti-
tutes a form of gamma correction. It is known to work well
for medium-dynamic-range scenes, but may cause burn-out
if used to compress HDR images. In the following, we will
argue that allowing burned-out areas can be beneficial for
the overall appearance of the image. However, it is impor-
tant to retain control over the number of pixels that become
over-exposed. This issue is discussed further in Section 8.1.
Furthermore, this operator is one of the few which
essentially comprise a forward and a backward pass. The
forward pass consists of computing brightness values derived
from image luminances. The backward pass then computes
luminance values from these brightness values which can
subsequently be displayed. This is theoretically correct, as
any tone-reproduction operator should formally take lumi-
nance values as input and produce luminance values as out-
put. If the reverse step is omitted, this operator would
produce brightness values as output. If the human visual
system is presented with such values, it will interpret these
brightness values as luminance values, and therefore proc-
ess values in the brain that have been perceived twice.
To avoid this, it is theoretically correct to apply a for-
ward pass to achieve range compression, followed by a
reverse pass to reconstitute luminance values and adjust the
values for the chosen display. It should not matter whether
this display is a conventional monitor or a HDR display. Fur-
thermore, this operator exhibits the desirable property that
tone-mapping an image that was tone-mapped previously,
results in an unaltered image.26 These issues are discussed
further in Section 8.
Histogram adjustment
A simple but effective approach to tone reproduction is to
derive a mapping from input luminances to display lumi-
nances using the histogram of the input image.12,27 Histo-
gram equalization would simply adjust the luminances so
FIGURE 5 — Tone-mapping function created by reshaping the
cumulative histogram of the image shown in Fig. 6.

that the probability that each display value occurs in the

output image is equal. Such a mapping is created by com-
puting the image’s histogram and then integrating this his-
togram to produce a cumulative histogram. This function
can be used directly to map input luminances to display
values.
However, dependent on the shape of the histogram, it
is possible that some contrasts in the image are exaggerated,
rather than attenuated. This produces unnatural results,
which may be overcome by restricting the cumulative histo-
gram to never attain a slope that is too large. The threshold
is determined at each luminance level by a model of human-
contrast sensitivity. The method is then called histogram
adjustment, rather than histogram equalization.12 An exam- FIGURE 6 — Image tone-mapped using histogram adjustment.
ple of display mapping generated by this method is shown
in Fig. 5. This mapping is derived from the image shown in
Fig. 6. in radiometric units. This requirement does not necessarily
This method can be extended to include simulations exist for approaches based on engineering principles.
of veiling luminance, illumination-dependent color sensitiv-
ity, and visual acuity. Although knowledge of human-con-
trast sensitivity is incorporated into the basic operator,12 we 4.3 Sigmoidal compression
may view this operator as a refined form of histogram
Equation (3) has an S-shaped curve on a log-linear plot and
equalization.
is called a sigmoid for that reason. This functional form fits
The importance of this observation is that histogram
data obtained from measuring the electrical response of
equalization is essentially rooted in engineering principles, photo-receptors to flashes of light in various species.28 It has
rather than a simulation of the human visual system. It is also provided a good fit to other electro-physiological and
therefore natural to regard this operator as “unit-less” – it psychophysical measurements of human visual func-
transforms luminance values to different luminance values, tion.29–31
as opposed to Tumblin and Rushmeier’s operator which Sigmoids have several desirable properties. For very
transforms luminance values to brightness values if applied small luminance values, the mapping is approximately lin-
in forward mode only. As a consequence, there is no theo- ear, so that contrast is preserved in dark areas of the image.
retical need to apply this model in reverse. The function has an asymptote at 1, which means that the
This argument can be extended to other tone-repro- output mapping is always bounded between 0 and 1. A fur-
duction operators. In general, when an algorithm simulates ther advantage of this function is that for intermediate val-
aspects of the human visual system, this by itself creates the ues, the function affords an approximately logarithmic
theoretical need to apply both the forward and inverse ver- compression. This can be seen for instance in Fig. 7, where
sions of the algorithm to ensure that the output is measured the middle section of the curve is approximately linear on a

Journal of the SID 15/12, 2007 1001

FIGURE 7 — Over the middle range values, sigmoidal compression is
approximately logarithmic. The choice of a semi-saturation constant
determines how input values are mapped to display values.
log-linear plot. To illustrate this, both Ld = Lv/(Lv + 1) and
Ld = 0.25 log (Lv) + 0.5 are plotted in this figure,a showing
that these functions are very similar over an input range cen-
tred around 1.
In Eq. (3), the function g(x, y) may be computed as a
global constant or as a spatially varying function. Following
common practice in electro-physiology, we call g(x, y) the
semi-saturation constant. Its value determines which values
in the input image are optimally visible after tone mapping,
as shown in Fig. 7. The effect of choosing different semi-
saturation constants is also shown in this figure. The term
semi-saturation constant derives from the fact that when the
input Lv reaches the same value as g(x, y), the output becomes
0.5.
In its simplest form, g(x, y) is set to Lv/k, so that the
geometric average is mapped to user parameter k (which
corresponds to the key of the scene).14 In this case, a good
initial value for k is 0.18, which conforms to the photo-
graphic equivalent of middle gray (although some would
argue that 0.13 rather than 0.18 is neutral gray). For particu-
larly light or dark scenes, this value may be raised or low-
ered. Alternatively, its value may be estimated from the
image itself.32 A variation of this global operator computes
the semi-saturation constant by linearly interpolating between
the geometric average and each pixel’s luminance:33
g(x, y) = aLv (x, y) + (1 - a)Lv .
The interpolation is governed by user parameter a ∈
[0,1], which has the effect of varying the amount of contrast
in the displayable image.
The exponent n in Eq. (3) determines how pronounced
the S-shape of the sigmoid is. Steeper curves map a smaller
useful range of image values to the display range, whereas
shallower curves map a larger range of input values to the
display range (see Fig. 8). Studies in electro-physiology
report values between n = 0.2 and n = 0.9.30
a
The constants 0.25 and 0.5 were determined by equating the values as
well as the derivatives of both functions for x = 1.
1002 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
FIGURE 8 — The exponent n determines the steepness of the curve.
Steeper slopes map a smaller range of input values to the display range.
5 Local tone-reproduction operators
The several different variants of sigmoidal compression
shown above are all global in nature. This has the advantage
that they are fast to compute, and they are very suitable for
medium-to-high-dynamic-range images. Their simplicity
makes these operators suitable for implementation on
graphics hardware as well. For very-high-dynamic-range
images, however, it may be necessary to resort to a local
operator since this may give somewhat better compression.
5.1 Local sigmoidal operators
A straightforward method to extend sigmoidal compression
replaces the global semi-saturation constant by a spatially
varying function, which once more can be computed in sev-
eral different ways. Thus, g(x, y) then becomes a function of
a spatially localized average. Perhaps the simplest way to
accomplish this is to once more use a Gaussian blurred
image. Each pixel in a blurred image represents a locally
averaged value which may be viewed as a suitable choice for
the semi-saturation constant. Although g(x, y) is now no
longer a constant, we continue to refer to it as the semi-satu-
ration constant.
As with division-based operators discussed in the pre-
vious section, we have to consider haloing artifacts. If sig-
moids are used with a spatially varying semi-saturation
(7)
constant, the Gaussian filter kernel is typically chosen to be
very small to minimize artifacts. In practice, filter kernels of
only a few pixels wide are sufficient to suppress significant
artifacts while at the same time producing more local con-
trast in the tone-mapped images. Such small filter kernels
can be conveniently computed in the spatial domain without
losing too much performance. There are, however, several
different approaches to compute a local average, which are
discussed in the following section.
5.2 Local neighborhoods
In local operators, halo artifacts occur when the local aver-
age is computed over a region that contains sharp contrasts
with respect to the pixel under consideration. It is therefore
important that the local average is computed over pixel val-
ues that are not significantly different from the pixel that is
being filtered.
This suggests a strategy whereby an image is filtered
such that no blurring over such edges occurs. A simple but
computationally expensive way is to compute a stack of
Gaussian blurred images with different kernel sizes, i.e., an
image pyramid. For each pixel, we may choose the largest
Gaussian which does not overlap with a significant gradi-
ent. The scale at which this happens can be computed as
follows.
In a relatively uniform neighborhood, the value of a
Gaussian blurred pixel should be the same regardless of the
filter kernel size. Thus, in this case, the difference between
a pixel filtered with two different Gaussians should be
around zero. This difference will only change significantly if
the wider filter kernel overlaps with a neighborhood con-
taining a sharp contrast step, whereas the smaller filter ker-
nel does not. A difference of Gaussians (DoG) signal
LiDoG(x, y) at scale i can be computed as follows:

LDoG
i ( x, y) = Ris (x, y) - R2is (x, y). (8)

Here, Riσ(x, y) is the Gaussian filtered image at scale

i and R2iσ(x, y) is the filtered image at scale 2i. The kernel
size doubles at each scale, with σ being the size of the filter
kernel at the smallest scale.
It is now possible to find the largest neighborhood
around a pixel that does not contain sharp edges by examin-
FIGURE 9 — Scale-selection mechanism. The top image shows the
ing differences of Gaussians at different kernel sizes i:14 tone-mapped result. The image on the bottom encodes the selected scale
for each pixel as a gray value. The darker the pixel, the smaller the scale.
LDoG
i (x, y) A total of eight different scales were used to compute this image.
>t i = 1...8 . (9)
Ris (x, y) + a

Here, the DoG filter is divided by one of the Gaus- Lv

sians to normalize the result, and thus enable comparison
against the constant threshold t which determines if the
neighborhood at scale i is considered to have significant
detail. The constant α is added to avoid division by zero.
For the image shown in Fig. 9 (top), the scale selected
for each pixel is shown in Fig. 9 (bottom). Such a scale selection
mechanism is employed by the photographic tone-repro-
duction operator14 as well as in Ashikhmin’s operator.21
Once for each pixel the appropriate neighborhood is
known, the Gaussian blurred average Lblur = Li for this
neighborhood may be used to steer the semi-saturation con-
stant, such as, for instance, employed by the photographic
tone-reproduction operator:
Lv
Ld = . (10)
1 + Lblur
It is instructive to compare the result of this operator
with its global equivalent, which is defined as
Ld = . (11)
1 + Lv
Images tone-mapped with both forms are shown in
Fig. 10. The CIE94 color difference metric shown in this
figure (bottom) shows that the main differences occur near
(but not precisely at) high-frequency high-contrast edges,
predominantly seen in the clouds. These are the regions
where more detail is produced by the local operator.
An alternative approach includes the use of edge-
preserving smoothing operators, which are designed spe-
cifically for removing small details while keeping sharp
contrasts in tact. Such filters have the advantage that sharp
discontinuities in the filtered result coincide with the
same sharp discontinuities in the input image, and may
therefore help to prevent halos. 26 Several such filters, such
as the bilateral filter, trilateral filter, Susan filter, the LCIS
algorithm, and the mean shift algorithm are suit-
able, 22–25,34,35 although some of them are expensive to
compute. Edge-preserving smoothing operators are dis-
cussed Section 5.4.

Journal of the SID 15/12, 2007 1003


FIGURE 10 — This image was tone-mapped with both global and local
versions of the photographic tone-reproduction operator (top and
middle). The CIE94 color difference is shown at the bottom.
5.3 Sub-band systems
Image pyramids can be used directly for the purpose of tone
reproduction, provided the filter bank is designed care-
fully.36 Here, a signal is decomposed into a set of signals that
can be summed to reconstruct the original signal. Such
algorithms are known as sub-band systems, or alternatively
as wavelet techniques, multi-scale techniques, or image
pyramids.
1004 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
An image consisting of luminance signals Lv(x, y) is
split into a stack of band-pass signals LiDoG(x, y) using (8),
whereby the n scales i increase by factors of 2. The original
signal Lv(x, y) can then be reconstructed by simply sum-
ming the sub-bands:
n
Lv (x, y) = Â LDoG
i (x, y). (12)
i=1
Assuming that Lv is a HDR image, a tone-mapped
image can be created by first applying a non-linearity to the
band-pass signals. In its simplest form, the non-linearity
applied to each LiDoG(x, y) is a sigmoid.26 However, as
argued by Li et al., summing the filtered sub-bands then
leads to distortions in the reconstructed signal.36 To limit
such distortions, either the filter bank may be modified or
the non-linearity may be redesigned.
Although sigmoids are smooth functions, their appli-
cation to a (sub-band) signal with arbitrarily sharp disconti-
nuities will yield signals with potentially high frequencies.
The high-frequency content of sub-band signals will cause
distortions in the reconstruction of the tone-mapped signal
Lv.36 The effective gain G(x, y) applied to each sub-band as
a result of applying a sigmoid can be expressed as a pixel-
wise multiplier:
LDoG DoG
s,i ( x, y) = Li (x, y) G(x, y). (13)
To avoid distortions in the reconstructed image, the
effective gain should have frequencies no higher than the
frequencies present in the sub-band signal. This can be
achieved by blurring the effective gain map G(x, y) before
applying it to the sub-band signal. This approach leads to a
significant reduction in artifacts and is an important tool in
the prevention of halos.
The filter bank itself may also be adjusted to limit dis-
tortions in the reconstructed signal. In particular, to remove
undesired frequencies in each of the sub-bands caused by
applying a non-linear function, a second bank of filters may
be applied before summing the sub-bands to yield the recon-
structed signal. If the first filter bank which splits the signal
into sub-bands is called the analysis filter bank, then the
second bank is called the synthesis filter bank. The non-
linearity described above can then be applied in between
the two filter banks. Each of the synthesis filters should be
tuned to the same frequencies as the corresponding analysis
filters.
An efficient implementation of this approach, which
produces excellent artifact-free results, is described by Li et
al.36 an example image is shown in Fig. 11. The image bene-
fits from clamping the bottom 2% of the pixels and the top
1% of the pixels, which is discussed further in Section 6.2.
Although this method produces excellent artifact-free
images, it has the tendency to over-saturate the image. This
effect was ameliorated in Fig. 11 by desaturating the image
using the technique described in Section 6.1 (with a value of
0.7, which is the default value used for the sub-band approach).
However, even after desaturating the image, its color fidel-
FIGURE 11 — Tone reproduction using a sub-band architecture, compu-
ted here using a Haar filter.36

ity remained a little too saturated. Further research would

be required to determine the exact cause of this effect,
which is shared with gradient domain compression (Section
5.5). Finally, it should be noted that the scene depicted in
Fig. 11 is a particularly challenging image for tone repro- FIGURE 12 — Bilateral filtering removes small details, but preserves
duction. The effects described here would be less pro- sharp gradients (top). The associated detail layer is shown on the bottom.
nounced for many HDR photographs.

Applying the bilateral filter to an image results in a blurred

5.4 Edge-preserving smoothing operators image in which sharp edges remain present (Fig. 12, top).
Such an image is normally called a “base” layer. This layer
An edge-preserving smoothing operator attempts to remove
has a dynamic range similar to the input image. The HDR
details from the image without removing high-contrast
image may be divided pixelwise by the base layer, obtaining
edges. An example is the bilateral filter,37–39 which is a spa-
a “detail” layer LD(x, y) which contains all the high-frequency
tial Gaussian filter multiplied with a second Gaussian oper-
detail, but typically does not have a very-high dynamic range
ating in the intensity domain. With Lv(x, y) the luminance
(Fig. 12, bottom):
at pixel (x, y), the bilateral filter LB(x, y) is defined as
LD (x, y) = Lv (x, y) LB (x, y). (16)
w(x, y, u, v)Lv (u, v)
LB (x, y) = , (14)
w( x, y, u, v) By compressing the base layer before recombining
into a compressed image, a displayable low-dynamic-range
b g b g
w(x, y, u, v) = Â Â Rs1 x - u, y - v Rs 2 Lv (x, y), Lv (u, v) . (15)
u v
image may be created (Fig. 13). Compression of the base
layer may be achieved by linear scaling. Tone reproduction
Here, σ1 is the kernel size used for the Gaussian oper- on the basis of bilateral filtering is executed in the logarith-
ating in the spatial domain and σ2 is kernel size of the inten- mic domain.
sity domain Gaussian filter. The bilateral filter can be used Edge-preserving smoothing operators may be used to
to separate an image into “base” and “detail” layers.24 compute a local adaptation level for each pixel, to be applied

Journal of the SID 15/12, 2007 1005


FIGURE 13 — An image tone-mapped using bilateral filtering. The base
and detail layers shown in Fig. 12 are recombined after compressing the
base layer.
in a spatially varying or local tone reproduction operator. A
local operator based on sigmoidal compression can for
instance be created by substituting Lblur(x, y) = LB(x, y) in
Eq. (10).
Alternatively, the semi-saturation constant g(x, y) in
Eq. (3) may be seen as a local adaptation constant and can
therefore be locally approximated with LB(x, y),40 or with
any of the other filters mentioned above. As shown in Fig. 7,
the choice of semi-saturation constant shifts the curve hori-
FIGURE 14 — The image on the left is tone-mapped using gradient domain compression. The magnitude of the gradients ||∇Lw|| is
mapped to a gray scale in the right image (white is a gradient of 0; black is the maximum gradient in the image).
1006 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
zontally such that its middle portion lies over a desirable
range of values. In a local operator, this shift is determined
by the values of a local neighborhood of pixels and is thus
different for each pixel. This leads to a potentially better
compression mechanism than a constant value could afford.
5.5 Gradient-domain operators
Local adaptation provides a measure of how different a pixel
is from its immediate neighborhood. If a pixel is very differ-
ent from its neighborhood, it typically needs to be attenu-
ated more. Such a difference may also be expressed in terms
of contrast, which could be represented with image gradi-
ents (in log space):
b g
—L = Lv (x + 1, y) - Lv (x, y), Lv (x, y + 1) - Lv (x, y) . (17)
Here, ∇L is a vector-valued gradient field. By attenu-
ating large gradients more than small gradients, a tone
reproduction operator may be constructed.41 Afterwards,
an image can be reconstructed by integrating the gradient
field to form a tone-mapped image. Such integration must
be approximated by numerical techniques, which is achieved
by solving a Poisson equation using the Full Multi-Grid
Method.42 The resulting image then needs to be linearly
scaled to fit the range of the target display device. An exam-
ple of gradient domain compression is shown in Fig. 14.
This form of dynamic-range reduction is especially useful in
applications where detail visibility is of importance.
 FIGURE 15 — Per-channel gamma correction may desaturate the image. The images were desaturated with values of s = 0.2, s = 0.5,
and s = 0.8.

Gelb effect can be handled correctly. The Gelb effect mani-

5.6 Lightness perception
fests itself when the brightest part of a scene is placed next
The theory of lightness perception provides a model for to an object that is even brighter. Whereas the formerly
the perception of surface reflectances Lightness is brightest object was perceived as white, after the change
defined as relative perceived surface reflectance. To cast this object no longer appears white, but light gray.
this into a computational model, the image needs to be
automatically decomposed into frameworks, i.e., regions
of common illumination. As an example, the window 6 Post-processing
shown in the bottom photograph of Fig. 17 would con-
After tone reproduction, it is possible to apply several post-
stitute a framework separate from the remainder of the
processing steps to either improve the appearance of the
interior. The influence of each framework on the total
image, adjust its saturation, or correct for the display
lightness needs to be estimated, and the anchors within
device’s gamma. Here, we discuss two frequently applied
each framework must be computed. 43–45 techniques which have a relatively large impact on the over-
It is desirable to assign a probability to each pixel of all appearance of the tone-mapped results. These are a tech-
belonging to a particular framework. This leaves the possi- nique to desaturate the results, and a technique to clamp a
bility of a pixel having non-zero participation in multiple percentage of the lightest and darkest pixels.
frameworks, which is somewhat different from standard
segmentation algorithms that assign a pixel to at most one
segment. To compute frameworks and probabilities for each
6.1 Color in tone reproduction
pixel, a standard K-means clustering algorithm may be applied.
For each framework, the highest luminance rule may Tone-reproduction operators normally compress luminance
now be applied to find an anchor. This means that within a values, rather than work directly on the red, green, and blue
framework, the pixel with the highest luminance would components of a color image. After these luminance values
determine how all pixels in this framework are likely to be have been compressed into display values Ld(x, y), a color
image may be reconstructed by keeping the ratios between
perceived. However, direct application of this rule may result
color channels the same as before compression (using s =
in the selection of a luminance value of a patch that is per-
1):13
ceived as self-luminous. As the anchor should be the highest
luminance value that is not perceived as self-luminous, se- s
F I (x, y) I
R
lection of the highest luminance value should be preceded I R,d (x, y) = GH L (x, y)JK
v
Ld (x, y),
by filtering the area of the local framework with a large
s
Gaussian filter.
(x, y) = G
F I (x, y)I
G
The anchors for each framework are used to compute
the net lightness of the full image. This then constitutes a
IG,d
H L (x, y)JK
v
Ld (x, y), (18)

s
computational model of lightness perception, which can be
(x, y) = G
F I (x, y) I
B
extended for the purpose of tone reproduction.
One of the strengths of using a computational model
I R,d
H L (x, y)JK
v
Ld (x, y).

of lightness perception for the purpose of tone reproduc- Alternatively, the saturation constant s may be chosen
tion, is that conventionally difficult phenomena such as the to be smaller than 1. Such per-channel gamma correction

Journal of the SID 15/12, 2007 1007

may desaturate the results to an appropriate level, as shown
in Fig. 15.41
The results of tone reproduction may sometimes
appear unnatural, because human color perception is non-
linear with respect to overall luminance level. If we view an
image of a bright outdoors scene on a monitor in a dim
environment, we are adapted to the dim environment rather
than the outdoors lighting. By keeping color ratios constant,
we do not take this effect into account. In addition, other phe-
nomena, which indicate interactions between different color
attributes, such as the Stevens, Hunt, and Bezold–Brücke
effects are not accounted for.b The above approach should
therefore be seen as a limited control to account for a com-
plex phenomenon.
A more-comprehensive solution is to incorporate
ideas from the field of color appearance modeling into tone-
reproduction operators.19,33,46 The iCAM image appear-
ance model is the first color-appearance model operating on
images.20,46,47 It can also be used as a tone-reproduction
operator. It therefore constitutes an important trend towards
the incorporation of color appearance modeling in dynamic-
range reduction and vice-versa.
A rudimentary form of color-appearance modeling
within a tone-reproduction operator is afforded by the sig-
moidal compression scheme outlined in Section 4.3.33
Nonetheless, we believe that further integration of tone
reproduction and color-appearance modeling is desirable
for the purpose of properly accounting for the differences in
adaptation between scene and viewing environments.
6.2 Clamping
A common post-process to tone reproduction is clamping. It
is, for instance, part of the iCAM model, as well as the sub-
band encoding scheme. Clamping is normally applied to
both very dark as well as very light pixels. Normally, a per-
centile of pixels which will be clamped. This gives good con-
trol over the final appearance of the image. By selecting a
percentile of pixels to be clamped, inevitably detail will be
lost in the dark and light areas of the image. However, the
remainder of the luminance values is spread over a larger
range, and this creates better detail visibility for large parts
of the image.
The percentage of pixels clamped varies usually between
1% and 5% dependent on the image. The effect of clamping
is shown in Fig. 16. The image on the left shows the results
without clamping, and therefore all pixels are within the dis-
play range. In the image on the right, the darkest 7% of the
pixels were clamped, as well as the lightest 2% of the pixels.
This has resulted in an image that has reduced visible detail
in the steps of the amphi theater, as well as in the wall and
b
The Stevens effect causes apparent contrast to increase with luminance.
The Hunt effect is an increase in colorfulness with luminance, and the
Bezold–Brücke hue shift is seen as a change in hue with increasing
luminance.
1008 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
the bushes in the background. However, the overall appear-
ance of the image has improved, and the clamped image
conveys the atmosphere of the environment better than the
directly tone-mapped image.
To preserve the appearance of the environment, the
disadvantage of losing detail in the lightest and darkest areas
is frequently outweighed by a better overall impression of
brightness. The photograph shown in Fig. 16 was taken dur-
ing a very bright day, and this is not conveyed well in the
unclamped images.
Finally, the effect of clamping has a relatively large
effect on the results. For typical applications it is an attrac-
tive proposition to add this technique to any tone-reproduc-
tion operator. However, as only a few tone-reproduction
operators incorporate this feature as standard, it also clouds
the ability to assess the quality of tone-reproduction opera-
tors. The difference between operators appears to be of
similar magnitude as the effect of clamping.
7 Mappings for HDR displays
The light emitted by an HDR display is, by definition,
spread over a much larger range than we see on most cur-
rent display devices. As a result, many physical scenes could
be captured in HDR and then displayed directly. As such,
the need for tone reproduction will be removed for some
images. However, this is not the case for all environments.
If the image has a range much higher than the HDR display
can handle, then non-linear compression schemes will con-
tinue to be a necessary pre-display step.
As opposed to conventional displays, HDR display
devices emit enough light to be sufficiently different from
the average room-lighting conditions. An important yet
poorly understood issue is that the human visual system
(HVS) will adapt in part to the device and in part to the
room environment. Such partial (or mixed) adaptation is
notoriously difficult to model and is certainly not a feature
of current tone-reproduction algorithms. A good tone-repro-
duction algorithm for HDR display devices would probably
have to account for the partial adaptation of the viewer. This
would include all forms of adaptation, including, for instance,
chromatic adaptation.
However, other than this fundamental issue, the dis-
tinction between HDR and LDR displays is arbitrary. As
such, we would argue that a good tone-reproduction algo-
rithm needs to be adaptable to any kind of display. Similarly,
in the context of dynamic-range management, we see little
difference between tone-reproduction operators and inverse-
tone-reproduction operators, even though the former is
used for reducing the dynamic range of an image to match
a display with a lower range, and the latter is used for expand-
ing the range of an image to match the dynamic range of a
display with a higher range. A good tone-reproduction opera-
tor would be able to both compress the dynamic range of an
image as well as expand it.
 FIGURE 16 — Example of clamping. Both images were tone-mapped using photographic tone reproduction. The left image is not
clamped, whereas 7% of the darkest and 2% of the lightest pixels were clamped in the right image.
Nonetheless, there are other issues related to dynamic-
range expansion, which will have to be taken into account.
These do not relate to the expansion of values per se, but are
related to artifacts in the source material that may be ampli-
fied to become more visible. For instance, by expanding the
luminance range of an image, the lossy compression applied
to JPEG images may become visible. Second, the non-linear
encoding of pixel values may, after expansion, lead to visible
quantization artifacts. Finally, under- and over-exposed areas
may require separate processing to salvage a lack of data in
these regions.48
Some solutions for these problems have been pro-
posed. For instance, Banterle et al. invert the photographic
operator for dynamic-range expansion.49 As this effectively
results in an inverse sigmoid, this makes the implicit but
erroneous assumption that the input image is given in units
which correspond to either photo-receptor output or some
perceptual quantity. Blocky artifacts, for instance, those
arising form JPEG encoding, are avoided by determining
pixels belonging to light sources in the input image and apply-
ing a different interpolation scheme for those pixels.
Rempel et al. have found that linear up-scaling can
typically be performed up until a contrast of 5000:1 before
the image takes on an unnatural appearance.50 Their solu-
tion is therefore to rely predominantly on linear up-scaling,
which is consistent with the finding that, at least for rela-
tively short exposure times, humans prefer to view linear
up-scaled image over non-linearly scaled images.51 The
appearance of artifacts is minimized by application of noise
filtering and quantization reduction through the use of a
bilateral filter. Pixel encodings of 235 or higher in video formats
are normally assumed to indicate light sources or highlights.
Those pixels can be enhanced separately. Alternatively,
highlights could be detected with a dedicated algorithm
before being scaled separately from the remainder of the
image.52
In summary, the main problems associated with dis-
playing conventional images on HDR display devices revolve
around avoiding the visibility of artifacts in the capture and
encoding of legacy content. So far, simple functions have
proved adequate for dynamic-range expansion, although we
would not draw the conclusion that this would be the case
for all images and all display conditions.
8 Tone reproduction and inverse tone
reproduction
Many tone-reproduction operators are modeled after some
aspects of human vision. The computed display values
Journal of the SID 15/12, 2007 1009
therefore essentially represent perceived quantities, for
instance, brightness if the tone-reproduction operator is
based on a model of brightness perception. If we assume
that the model is an accurate representation of some aspect
of the HVS, then displaying the image and observing it will
cause the HVS to interpret these perceived values as lumi-
nance values.
The HVS thus applies a second perceptual transform
on top of the one applied by the algorithm. This is formally
incorrect. A good tone-reproduction operator should follow
the same common practice as employed in color-appearance
modeling and apply both a forward and a reverse trans-
form.53 The forward transform can be any algorithm
thought to be effective at compressing luminance values.
The reverse transform will then apply the algorithm in
reverse, but with display parameters inserted. This approach
compresses luminance values into perceived values, while
the reverse algorithm will convert the perceived values back
into luminance values.
Tumblin and Rushmeier’s approach correctly takes
this approach, as does the Multi-Scale Observer model,19 all
color appearance models and gradient-domain operators,
and the aforementioned sub-band system. However, several
perceptually based operators are applied only in forward
mode, including the photographic operatorc and the sigmoi-
dal operator inspired by photo-receptor physiology.33 While
these operators are known to produce visually plausible
results, we note that they are effectively not producing dis-
play luminances but brightnesses or other equivalent per-
ceptual attributes.
FIGURE 17 — Forward and reverse model with n = 0.7 and k = 0.3 for
an HDR image with a relatively modest dynamic range of 2.8 log units
(top) and an image with a much higher dynamic range (bottom); n = 0.7,
8.1 Sigmoids revisited k = 0.08, 8.3 log units dynamic range.
Here, we discuss the implications of adding a reverse step
to sigmoidal compression. Recall that Eq. (3) can be rewrit-
m. By making these substitutions, we have replaced all the
ten as
image-related user parameters (n and g) with their display-
Lny (x, y) related equivalents (m and Ld,mean). The resulting inverse
V(x, y) = , (19) equation, computing display values Ld from previously com-
Lnv (x, y) + g n (x, y) puted perceived values V, is then
where V is a perceived value (for instance, a voltage if this
equation is thought of as a simple model of photoreceptor
F V(x, y)Lmd,mean I 1 m
Ld (x, y) = G
H 1 - V(x, y) JK
. (20)
physiology). The function g(x, y) continues to return either
a globally or locally computed adaptation value, which is
based on the image values. For a conventional display, we would set Ld,mean to
To convert these perceived values back to luminance 128. The exponent m is also a display-related parameter and
values, this equation then needs to be inverted, whereby determines how display values are spread around the mean
g(x, y) is replaced with a display adaptation value (see also display luminance. For low-dynamic-range display devices,
Section 4.1). For instance, we could try to replace g(x, y) this value can be set to 1, thereby simplifying the above
with the mean display luminance Ld,mean. The other user equation to
parameter in this model is the exponent n that for the reverse V(x, y)Ld,mean
model we will be replaced with a display-related exponent Ld (x, y) = . (21)
1 - V(x, y)

c The computation of display values is now driven entirely

Although this operator can be explained as based on photographic prin-
ciples, it’s underlying model is a perceptual model of brightness percep- by the mean luminance of the image (through the computa-
tion.54 tion of g), the mean display luminance Ld,mean, as well as the

1010 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
 FIGURE 19 — Photographic operator with key k = 0.3 (top) and k = 0.08
FIGURE 18 — Forward model only with n = 0.7 and k = 0.3 (top) and (bottom). By using the photographic operator, we have effectively
n = 0.7 and k = 0.08 (bottom). changed the exponent to n = 1.

exponent n which specifies how large a range of values

around the mean luminance will be visualized. As a result,
light or dark the image should be reproduced (see Section
the inverse transform may create display values that are out-
4.3).14 The exponent n can be thought of as a measure of
side the display range. These will have to be clamped.
how much contrast there is in the image.
Given that for most display devices the peak lumi-
Two results using visually determined optimal parameter
nance Ld,max, as well as the black level Ld,min are known, it
settings are shown in Fig. 17. The display settings are those
is attractive to use these natural boundaries to clamp the
for an average display with an assumed black level of 1
display values against. This is perhaps a more natural choice
cd/m2 and a peak luminance of 300 cd/m2. As a conse-
than clamping against a percentile, as discussed in Section 6.2.
quence, Ld,mean was set to 150.5. The left image is repro-
Given that we will gamma correct the image after-
duced in a satisfactory manner. However, the amount of
wards, we may assume that the display range of Ld is linear.
burn-out that has occurred in the window of the right image
As such, we can now compute the mean display luminance
is too much. However, it is difficult to find a good trade-off
as the average of the display’s black level and peak lumi-
between having less burn-out and introducing other arti-
nance:
facts with this method.
dL i
+ Ld,max 2. (22) For comparison, we show the forward-only transform
d,min
with otherwise identical parameter settings in Fig. 18. Note
Because there is no good theoretical ground for choos- that the top image looks more flat now, largely because the
ing any specific value for the exponent m, we will set this exponent n is now no longer optimal. The window in the
parameter to 1 for now. As a result, all display-related parame- bottom image now appears more correct, as the brown glass
ters are fixed, leaving only the image-dependent parameter panels are now clearly visible. We also show the output of
n as well as the choice of semi-saturation constant g( ). For the photographic operator for the same pair of images in
our example, we will follow Reinhard et al., and set g( ) = Fig. 19. The exponent n is effectively set to 1, but the key
Lv/k, where the user parameter k determines how overall value is the same as in the previous figures. Although these

Journal of the SID 15/12, 2007 1011

images are computed with a forward transform only, their
visual appearance remains closer to the real environment
than the images in Fig. 17.
Finally, it is desirable that a tone-reproduction opera-
tor does not alter an image that is already within the display
range.26 In the model proposed here, this is implicitly
achieved, as for n = m and g = Ld,mean, the reverse transform
is the true inverse of the forward transform. This is borne
out in the CIE94 color-difference metric, which is uni-
formly 0 for all pixels after running the algorithm twice.
8.2 Combined forward/reverse sigmoids
Although applying both a forward and a reverse transform is
formally the correct approach to tone reproduction, there is
thus a problem for images with a very-high dynamic range.
For such images it is difficult, if not impossible, to find
parameter settings that lead to an acceptable compression.
To see why this is, we can plug the forward transform of Eq.
(3) into the inverse of Eq. (20):
1m
F L n I
GG n
v L n
m
d,mean
JJ Lnv m Ld,mean
L (L K)
Ld =G v v
n JJ = nm
= cLnv m , (23)
GG 1 - L v
JJ dL ki
v a range of different tasks and displayed material.55–60
GH L + dL ki
n
v v
n
K When dealing with different displays, each having
where
Ld,mean
c=
nm
d Lw k i modeling, we have argued that both a forward and a reverse
is a constant. Of course, this is essentially the same result as
was obtained by matching image and display brightnesses in
Tumblin and Rushmeier’s brightness-matching operator
(see Section 4.1). Hence, applying a sigmoid in forward and
reverse mode amounts to applying a power function.
In our experience, this approach works very well in
cases where a medium amount of compression is required.
For instance, a medium-dynamic-range image can be effec-
tively tone-mapped for display on a low-dynamic-range dis-
play device. Alternatively, it should be possible to tone-map
most HDR images for display on HDR display devices using
this technique. However, for high compression ratios, a dif-
ferent approach would be required.
A direct consequence is that we predict that color-ap-
pearance models such as CIECAM02 cannot be extended to
transform data over large ranges. It is well-known that CIE-
CAM02 was never intended for transforming between sig-
nificantly different display conditions. However, this can be
attributed to the fact that the psychophysical data on which
this model is based was gathered over a limited dynamic
range. The above findings suggest that, in addition, extension
of CIECAM02 to accommodate large compression ratios
would require a different functional form.
1012 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
Whether the inclusion of spatial processing, such as a
spatially varying semi-saturation constant, yields more satis-
factory results remains to be seen. As can be understood
from Eq. (23), replacing g( ) = Lv/k with a spatially varying
function means that each pixel is divided by a spatially
determined denominator. Such an approach was pioneered
in Chiu et al.’s early work,15 and has been shown to be prone
to haloing artifacts (see Secs. 3 and Fig. 4). To minimize the
occurrence of halos in such a scheme, the size of the aver-
aging kernel used to compute g( ) must be chosen to be very
large; typically a substantial fraction of the entire image. But
in the limit that the filter kernel becomes the entire image,
this means that each pixel is divided by the same value,
resulting in a spatially invariant operator.
9 Discussion
Tone reproduction for low-dynamic-range display devices is
nowadays a reasonably well understood problem. The majority
of images can be compressed well enough for applications
in photography and entertainment, and any other applica-
tions that do not critically depend on accuracy. Recent vali-
dation studies show that some algorithms perform well over
their own dynamic range, it becomes more important to
consider tone-reproduction operators that can be parame-
terized for both different types of images and different types
(24) of display. Following common practice in color-appearance
transform are necessary.
However, we have identified a disconnect between
theory and practice. In particular, a reverse operation should
follow a forward transform to convert perceived values back to
values amenable for interpretation as luminances. If we
apply this to the class of sigmoidal functions, of which color
appearance models form a part, then we effectively reduce
the compressive function to a form of gamma correction. It
is more difficult to produce visually plausible results this
way, as less control can be exercised over the trade-off
between burn-out, contrast, and visual appearance.
To solve this problem, we would either have to find an
alternative reasoning whereby the inverse model does not
have to be applied or instead develop new tone-reproduc-
tion operators which both include a forward and reverse
model and produce visually plausible and controllable results.
The backward step would have the additional benefit of
being adaptable to any type of display, including HDR display
devices. We anticipate that this implies that such operators
would obviate the need for dedicated inverse-tone-repro-
duction operators, although image processing to counter the
effects of quantization, spatial compression, as well as pos-
sible exposure artifacts will remain necessary.

References
1 L Spillmann and J S Werner, eds., Visual Perception: The Neurological
Foundations (Academic Press, San Diego, 1990).
2 J A Ferwerda, “Elements of early vision for computer graphics,” IEEE
Computer Graphics Applications 21(5), 22–33 (2001).
3 H Seetzen, L A Whitehead, and G Ward, “A high dynamic range display
using low and high resolution modulators,” SID Symposium Digest
Tech Papers 34, 1450–1453 (2003).
4 H Seetzen, W Heidrich, W Stuerzlinger, G Ward, L Whitehead, M
Trentacoste, A Ghosh, and A Vorozcovs, “High dynamic range display
systems,” ACM Trans Graphics 23(3), 760–768 (2004).
5 E Reinhard, G Ward, S Pattanaik, and P Debevec, High Dynamic
Range Imaging: Acquisition, Display and Image-Based Lighting (Mor-
gan Kaufmann Publishers, San Francisco, 2005).
6 E Reinhard, “Overview of dynamic range reduction,” SID Symposium
Digest Tech Papers 38, 1049–1052 (2007).
7 N J Miller, P Y Ngai, and D D Miller, “The application of computer
graphics in lighting design,” J IES, 14, 6–26 (1984).
8 J Tumblin and H Rushmeier, “Tone reproduction for computer gener-
ated images,” IEEE Computer Graphics Applications 13(6), 42–48
(1993).
9 G Ward, “A contrast-based scalefactor for luminance display,” Paul
Heckbert, ed., Graphics Gems IV (Academic Press, Boston, 1994), pp.
415–421.
10 J A Ferwerda, S Pattanaik, P Shirley, and D P Greenberg, “A model of
visual adaptation for realistic image synthesis,” SIGGRAPH Conf Proc,
249–258 (1996).
11 F Drago, K Myszkowski, T Annen, and N Chiba, “Adaptive logarithmic
mapping for displaying high contrast scenes,” Computer Graphics
Forum 22(3) (2003).
12 G Ward, H Rushmeier, and C Piatko, “A visibility matching tone
reproduction operator for high dynamic range scenes,” IEEE Trans
Visualization Computer Graphics 3(4) (1997).
13 C Schlick, “Quantization techniques for the visualization of high dynamic
range pictures,” Peter Shirley, Georgios Sakas, and Stefan Müller, eds.,
Photorealistic Rendering Techniques (Springer-Verlag, Berlin, Ger-
many, 1994), pp. 7–20.
14 E Reinhard, M Stark, P Shirley, and J Ferwerda, “Photographic tone
reproduction for digital images,” ACM Trans Graphics 21(3), 267–276
(2002).
15 K Chiu, M Herf, P Shirley, S Swamy, C Wang, and K Zimmerman,
“Spatially nonuniform scaling functions for high contrast images,” Proc
Graphics Interface ‘93, 245–253 (1993).
16 D J Jobson, Z Rahman, and G A Woodell, “Retinex image processing:
Improved fidelity to direct visual observation,” Proc IS&T 4th Color
Imaging Conference: Color Science, Systems, and Applications 4,
124–125 (1995).
17 Z Rahman, D J Jobson, and G A Woodell, “A multiscale retinex for
color rendition and dynamic range compression,” SPIE Proc: Applica-
tions of Digital Image Processing XIX, 2847 (1996).
18 Z Rahman, G A Woodell, and D J Jobson, “A comparison of the
multiscale retinex with other image enhancement techniques,” IS&T’s
50th Annual Conference: A Celebration of All Imaging 50, 426–431
(1997).
19 S N Pattanaik, J A Ferwerda, M D Fairchild, and D P Greenberg, “A
multiscale model of adaptation and spatial vision for realistic image
display,” SIGGRAPH ‘98 Conf Proc, 287–298 (1998).
20 M D Fairchild and G M Johnson, “Meet iCAM: An image color
appearance model,” IS&T/SID 10th Color Imaging Conference, 33–38
(2002).
21 M Ashikhmin, “A tone mapping algorithm for high contrast images,”
Proc 13th Eurographics Workshop on Rendering, 145–155 (2002).
22 S N Pattanaik and H Yee, “Adaptive gain control for high dynamic
range image display,” Proc Spring Conf in Computer Graphics, 24–27
(2002).
23 A V Oppenheim, R Schafer, and T Stockham, “Nonlinear filtering of
multiplied and convolved signals,” Proc IEEE 56(8), 1264–1291
(1968).
24 F Durand and J Dorsey, “Fast bilateral filtering for the display of
high-dynamic-range images,” ACM Trans Graphics 21(3), 257–266
(2002).
25 P Choudhury and J Tumblin, “The trilateral filter for high contrast
images and meshes,” Proc Eurographics Symposium on Rendering,
186–196 (2003).
26 J M DiCarlo and B A Wandell, “Rendering high dynamic range im-
ages,” Proc SPIE Electronic Imaging 2000 Conference 3965, 392–401
(2000).
27 J Duan, G Qiu, and M Chen, “Comprehensive fast tone mapping for
high dynamic range image visualization,” Proc Pacific Graphics (2005).
28 K I Naka and W A H Rushton, “S-potentials from luminosity units in
the retina of fish (cyprinidae),” J Physiology (London) 185(3), 587–599
(1966).
29 J Kleinschmidt and J E Dowling, “Intracellular recordings from gecko
photoreceptors during light and dark adaptation,” J General Physiology
66(5), 617–648 (1975).
30 D C Hood, M A Finkelstein, and E Buckingham, “Psychophysical tests
of models of the response function,” Vision Research 19(4), 401–406
(1979).
31 D C Hood and M A Finkelstein, “Comparison of changes in sensitivity
and sensation: Implications for the response-intensity function of the
human photopic system,” J Experimental Psychol: Human Perceptual
Performance 5(3), 391–405 (1979).
32 E Reinhard, “Parameter estimation for photographic tone reproduc-
tion,” J Graphics Tools 7(1), 45–51 (2003).
33 E Reinhard and K Devlin, “Dynamic range reduction inspired by
photoreceptor physiology,” IEEE Trans Visualization Computer
Graphics 11(1), 13–24 (2005).
34 J Tumblin and G Turk, “LCIS: A boundary hierarchy for detail-preserv-
ing contrast reduction,” Alyn Rockwood, ed., SIGGRAPH 1999, Com-
puter Graphics Proceedings, 83–90 (1999).
35 D Comaniciu and P Meer, “Mean shift: a robust approach toward
feature space analysis,” IEEE Trans Pattern Analysis Machine Intelli-
gence 24(5), 603–619 (2002).
36 Y Li, L Sharan, and E H Adelson, “Compressing and companding high
dynamic range images with subband architectures,” ACM Trans
Graphics 24(3), 836–844 (2005).
37 C Tomasi and R Manduchi, “Bilateral filtering for gray and color
images,” Proc IEEE International Conference Computer Vision,
836–846 (1998).
38 S Paris and F Durand, “A fast approximation of the bilateral filter using
a signal processing approach,” European Conference on Computer
Vision (2006).
39 B Weiss, “Fast median and bilateral filtering,” ACM Trans Graphics
25(3) (2006).
40 P Ledda, L-P Santos, and A Chalmers, “A local model of eye adaptation
for high dynamic range images,” Proc ACM Afrigraph, 151–160 (2004).
41 R Fattal, D Lischinski, and M Werman, “Gradient domain high dy-
namic range compression,” ACM Trans Graphics 21(3), 249–256
(2002).
42 W H Press, S A Teukolsky, W T Vetterling, and B P Flannery, Numerical
Recipes in C: The Art of Scientific Computing, 2nd edn. (Cambridge
University Press, 1992).
43 G Krawczyk, R Mantiuk, K Myszkowski, and H-P Seidel, “Lightness
perception inspired tone mapping,” Proc First ACM Symposium on
Applied Perception in Graphics and Visualization, 172 (2004).
44 G Krawczyk, K Myszkowski, and H-P Seidel, “Lightness perception in
tone reproduction for high dynamic range images,” Computer Graph-
ics Forum 24(3), 635–645 (2005).
45 G Krawczyk, K Myszkowski, and H-P Seidel, “Computational model of
lightness perception in high dynamic range imaging,” Proc IS&T/SPIE
Human Vision and Electronic Imaging 6057 (2006).
46 M D Fairchild and G M Johnson, “The iCAM framework for image
appearance, image differences, and image quality,” J Electronic Imag-
ing (2004).
47 N Moroney and I Tastl, “A comparison of retinex and iCAM for scene
rendering,” J Electronic Imaging 13(1) (2004).
48 L Wang, L-Y Wei, K Zhou, B Guo, and H-Y Shum, “High dynamic
range image hallucination,” Eurographics Symposium on Rendering
(2007).
49 F Banterle, P Ledda, K Debattista, and A Chalmers, “Inverse tone
mapping,” GRAPHITE ‘06: Proc 4th Intl Conf on Computer Graphics
and Interactive Techniques in Australasia and Southeast Asia, 349–356
(2006).
50 A Rempel, M Trentacoste, H Seetzen, D Young, W Heidrich, L White-
head, and G Ward, “On-the-fly reverse tone mapping of legacy video
and photographics,” ACM Trans Graphics (Proc SIGGRAPH 2007)
26(3) (2007).
Journal of the SID 15/12, 2007 1013
51 A O Akyuz, E Reinhard, R Fleming, B Riecke, and H Bulthoff, “Do
HDR displays support LDR content? A psychophysical evaluation,”
ACM Trans Graphics 26(3) (2007).
52 L Meylan, S Daly, and S Susstrunk, “Tone mapping for high dynamic
range displays,” Proc IS\&T/SPIE Electronic Imaging: Human Vision
and Electronic Imaging. XII 6492 (2007).
53 M D Fairchild, Color Appearance Models (Addison-Wesley, Reading,
MA, 1998).
54 F J J Blommaert and J-B Martens, “An object-oriented model for
brightness perception,” Spatial Vision 5(1), 15–41 (1990).
55 F Drago, W L Martens, K Myszkowski, and H-P Seidel, “Perceptual
evaluation of tone mapping operators with regard to similarity and
preference,” Technical Report MPI-I-2002-4-002 (Max Plank Institut
fur Informatik, 2002).
56 J Kuang, H Yamaguchi, G M Johnson, and M D Fairchild, “Testing
HDR image rendering algorithms,” Proc IS&T Color Imaging Confer-
ence, 315–320 (2004).
57 P Ledda, A Chalmers, T Troscianko, and H Seetzen, “Evaluation of
tone mapping operators using a high dynamic range display,” ACM
Trans Graphics 24(3), 640–648 (2005).
58 A Yoshida, V Blanz, K Myszkowski, and H-P Seidel, “Perceptual evalu-
ation of tone mapping operators with real-world scenes,” Proc SPIE
Human Vision and Electronic Imaging X 5666, 192–203 (2005).
59 A Yoshida, R Mantiuk, K Myszkowski, and H-P Seidel, “Analysis of
reproducing real-world appearance on displays of varying dynamic
range,” Computer Graphics Forum 25(3) (2006).
60 M Ashikhmin and J Goral, “A reality check for tone mapping opera-
tors,” ACM Trans Appl Perception 4(1) (2007).
Erik Reinhard received his Ph.D. in computer science from the Univer-
sity of Bristol in 2000. Following a post-doctoral position at the Univer-
sity of Utah (2000–2002) and assistant professor at the University of
Central Florida (2002–2005), he returned to Bristol as a lecturer in Janu-
ary 2006. He founded the prestigious ACM Transactions on Applied Per-
ception and has been Editor-in-Chief since its inception in 2003. He is
lead author of the only book on High Dynamic Range Imaging. He was
program chair for Eurographics PGV 2002, ACM APGV 2006, and the
Eurographics Short Papers programme 2008, and is an invited speaker
for the Society for Information Display’s main conference (2007). Erik
has published more than 12 journal papers and more than 25 confer-
ence papers, including three at SIGGRAPH.
1014 Reinhard et al. / Image display algorithms for high- and low-dynamic-range display devices
Timo Kunkel is a Ph.D. student at the University of Bristol, U.K. He
received his M.Sc. degree from Albert Ludwigs University in Freiburg,
Germany. His current research interests include color, color imaging,
high-dynamic-range imaging, and computational photography.
Yoann Marion received his M.S. degree in computer science from the
University of Rennes, France, in 2007. He did his research project at
IRISA, which was concerned with the use of color-appearance models
in rendering algorithms.
Jonathan Brouillat received his M.S. degree in computer science from
the University of Rennes, Rennes, France, in 2006. He is a Ph.D. student
in computer science at INRIA, Rennes, France. His main research inter-
ests are computer graphics, interactive and realistic rendering and
global illumination.
Rémi Cozot obtained his B.S. degree in engineering from the Ecole
Nationale Supérieure de Mécanique et Aérotechnique of Poitiers
(ENSMA), France, in embedded software in 1991, M.S. degree of sci-
ence from the University of Poitiers, France, in 1992, and his Ph.D. in
computer science in 1996 from the University of Rennes 1. He is cur-
rently an assistant professor at the University of Rennes and a permanent
member of the Computer Graphics Research Team Bunraku at IRISA/
INIRIA in Rennes, France. His research interests include computers
graphics, digital imaging, rendering and perception, and color appear-
ance models.
Kadi Bouatouch is an electronics and automatic systems engineer (ENSEM
1974). He was awarded a Ph.D. in 1977 and a higher doctorate on
computer science in the field of computer graphics in 1989. His is work-
ing on global illumination, lighting simulation for complex environ-
ments, GPU-based rendering and computer vision. He is currently
Professor at the university of Rennes, France, and researcher at IRISA
(Institut de Recherche en Informatique et Systemes Aleatoires). He is
member of Eurographics, ACM, and IEEE. He was member of the pro-
gram committee of several conferences and workshops and referee for
several Computer Graphics journals, including The Visual Computer,
IEEE Computer Graphics and Applications, IEEE Transactions on Visuali-
zation and Computer Graphics, and IEEE Transactions on Image Proc-
essing. He also acted as a referee for many conferences and workshops.