I’kion Rr,. Vol. 33. No. IO, pp. 1431 1436, 1993 004MYXY ‘Y3 $6.00 + 0.

00
Prmted in Great Hritaln. All rights reserved Copyright t lYY3 Pergamon Pres\ Ltd

Research Note
Michelson Contrast, RMS Contrast and Energy of Various
Spatial Stimuli at Threshold
HELJ;i KUKKONEN,* JYRKI ROVAMO,* KAISA TIIPPANA,* RISTO N;iS;iNEN*t
Rewiwd 2 July 1992; in reuised,fi?rm 30 November 1992

Contrast sensitivity was measured as a function of median spatial frequency for vertical cosine
gratings, narrow-band noise stimuli, and spots with luminance increment or decrement. Contrast
sensitivity was expressed in terms of Michelson contrast, RMS contrast, and contrast energy in order
to demonstrate the characteristics of various contrast measures. Gratings and noise stimuli had either
constant stimulus area or constant number of cycles. Michelson contrast sensitivity was better for
gratings than for noise stimuli or spots, whereas RMS contrast sensitivity was almost equal for
gratings and noise stimuli but lowest for spots. Contrast energy sensitivity takes into account the
stimulus area and was therefore best for spots.

Michelson contrast RMS contrast Contrast energy Gratings Noise stimuli Spots

INTRODUCTION ments. Local contrast (AL/L,) for light spots with sharp
edges has a constant value within the spot area. There-
The definition of contrast is often selected on the basis
fore, for such spots RMS contrast is equal to the Weber
of the stimulus used. For spots detection threshold is
fraction.
usually expressed in terms of the Weber fraction or light
Another contrast metric, contrast energy, has been
energy, for periodic stimuli, like gratings, in terms of
used in studies of human detection efficiency (e.g. Wat-
Michelson contrast and for aperiodic and complex stim-
son, Barlow & Robson, 1983; Kersten, 1984; Burgess &
uli in terms of RMS contrast or contrast energy. How-
Ghandeharian, 1984a, b; Legge, Kersten & Burgess,
ever, in principle Michelson contrast, RMS contrast and
1987). Contrast energy is equal to the local contrast
contrast energy could be applied to all of these stimuli.
signal squared and integrated over the stimulus area.
Michelson and RMS contrasts are the most com-
Thus, contrast energy takes into account both the effect
monly used contrast metrics. The main difference be-
of luminance distribution and stimulus area.
tween them is that Michelson contrast defines contrast
In order to demonstrate how contrast energy, Michel-
using the peak luminance values of the stimulus whereas
son and RMS contrast metrics describe human visual
RMS contrast is based on the standard deviation of
performance at detection threshold for various stimuli,
luminance levels in the stimulus. Michelson contrast is
we measured contrast thresholds for cosine gratings and
straightforward and simple by definition, but it cannot
noise stimuli as well as increment and decrement
be considered very descriptive for stimuli with complex
thresholds for Gaussian spots. The inverses of thresholds
or aperiodic luminance distributions (e.g. Moulden,
were then plotted as a function of median spatial
Kingdom & Gatley, 1990). A better contrast metric for
frequency, which indicates the median of the energy
this kind of stimuli would, therefore, be RMS contrast
spectrum of the stimulus.
(e.g. Pollehn & Roehrig, 1970; Stromeyer & Julesz, 1972;
Koenderink & van Doorn, 1974).
In addition, although Michelson contrast for gratings METHODS
is equal to modulation amplitude divided by mean
luminance, it is not equal to the Weber fraction for light The apparatus has been described in N&inen.
spots, because background luminance is equal to the Kukkonen and Rovamo (1993) and the procedure in
minimum luminance in increment threshold measure- Rovamo, Fransila and NBssnen (1992). Therefore, only
the main features are described here.
The stimuli were produced under computer control
on a high-resolution monitor driven at 60 Hz frame
*Department of’ Vision Sciences, University of Aston. Aston Triangle.
Birmingham 84 7ET, England.
rate by a VGA graphics board. The colour of the
tDepartment of Physiology, University of Helsinki, Siltavuorenpcnger screen was changed to white by means of a switch in the
20 J, SF-00170 Helsinki, Finland. control panel of the monitor. The non-linear luminance
1432 RESEAR~‘l
response of the display was linearised by digital gamma
correction. The modulation transfer function of the
display was constant within the spatial frequency range
used in the experiments. A monochrome signal of 1024
intensity levels was obtained from a monochrome palette
of 16,384 intensity levels by means of a video summation
device and a periodic dithering signal. This allowed the
measurement of contrast sensitivity for sinusoidal
gratings with at least 10 different grey levels even at a
Michelson contrast as low as 0.00125. The Michelson
contrasts of the gratings at and above 0.3% were
checked using a Minolta Luminance Meter LS- 1IO.
The experiments were performed in a dark room to
avoid the stimulation of peripheral vision. The only light
source was the computer screen. Its horizontal and
vertical dimensions were 26.9 and 20.2 cm, respectively.
For spot stimuli the equiluminous surround was limited
by means of a black cardboard to a circular field 20 cm
in diameter. The average luminance of the screen was
50 cd/cm*. The stimuli were viewed binocularly with
natural pupils of 5-6 mm in diameter. The head of the
subject was stabilised using a chin rest. Subjects were
asked to fixate at the centre of the stimulus. No fixation
point was used.
Two experienced subjects, aged 2.5and 27 yr, served as
observers, KT was a corrected myope (od. - 6.25 Ds/os.
-4.25 Ds) and HK was an uncorrected hyperope
( +0.5 Ds oa.). Their binocular Snellen acuities at 5 m
were 1.7 and 1.3, respectively.
The contrast thresholds were determined by a two-
alternative forced-choice algorithm. Stimulus contrast
was changed in steps of 0.1 log,, units. Each trial
consisted of two 500 msec exposures, separated by
600 msec. The stimuli were rapidly switched on and off
by changing the colour look-up table during the vertical
retrace period of the display. A new trial began 250 msec
after the subjects’ response. The contrast required for the
probability of 0.84 correct was estimated from eight
contrast reversals as described by Wetherill and Levitt
(1965). Each data point is based on a geometric mean of
three threshold estimates.
The stimuli consisted of vertical cosine gratings, noise
stimuli and spots. For sharp-edged gratings and noise
stimuli either the stimulus area (A > in deg* or the relative
image area (Virsu & Rovamo, 1979) was constant.
Relative image area (Afi) is expressed in square cycles
(c’) and calculated by multiplying stimulus area (il) and
median spatial frequency squared (1:). Relative image
area as well as the number of cycles was kept constant
by reducing the diameter of the circular stimulus field
in proportion to increasing median spatial frequency.
Median spatial frequency (f,) refers to the median of the
energy spectrum of the stimulus. Thus, half of the
stimulus contrast energy is above and half is below its
median frequency.
The noise stimuli were produced by band-pass filtering
a sample of two-dimensional static white noise. Noise
was produced by adding a random number from an
even distribution with zero mean to each pixel of
0.42 x 0.42 mm2 in size. The basic stimulus was then
I NO’l-E
band-pass filtered using a circularly symmetric log-
Gaussian transfer function:
MTF(,f) = exp( -~nZIf%f;-IIE~~(2)~‘Jj. (1)
wheref= K' +.f,,
21O5is the radial spatial frequency of
the stimulus, f,is the radial centre frequency of the filter,
and b is its half-bandwidth in octaves.
The radial centre frequencies (.A) and stimulus areas
(A ) were chosen to correspond to the spatial frequencies
and areas of the cosine gratings, Gratings with constant
relative image area of 113 c’ contained 12 cycles per
stimulus diameter at all spatial frequencies. We used
median spatial frequencies from 1 to 22.3 c/deg with
approximately half octave steps plus an additional
spatial frequency at 0.5 c/deg. The filter bandwidth of
noise stimuli was 0.5 octave. The other set of stimuli had
a constant stimulus area of 12.7 deg’. Median spatial
frequencies varied between 0.249 and 15.9 c/deg with
approx. 1 octave steps. The filter bandwidth of noise
stimuli was 1 octave. For grating stimuli the median
spatial frequency is equivalent to the nominal spatial
frequency and for noise stimuli with bandwidth of 0.5-l
octaves, as in our study, it is approximately equal to the
filter centre frequency.
Spot stimuli with luminance increments and decre-
ments were generated using Gaussian weighted stimulus
windows of the form
G(r) = exp( -1n 2[r/r,!,]* I, (2)
where r,:* is the radius, at which the luminance deviation
had decreased to half of its maximum value. The diam-
eters (2~~~~)of the spots were 0.5-0.125cm. Median
spatial frequencies (f,) were varied between
0.634-16.5 c/deg by using viewing distances of 57 and
319cm. Stimulus area (A = nr:,,) varied between
1.96 x 10-l and 3.95 x 10. 4 deg*, respectively. There-
fore, the vatue of the relative image area, calculated as
Af ‘,, remained approximately constant varying between
8.27 x 10m2and 1.07 x 10 ’ c” for the spot stimuli used.
RESULTS
In the experiments of Fig. 1 we measured contrast
detection thresholds as a function of median spatial
frequency for cosine gratings and corresponding noise
stimuli. Contrast sensitivity was expressed in terms
of Michelson contrast, RMS contrasts, and contrast
energy. The noise stimuli are described by equation (1).
Both types of stimuli had either a constant stimulus
area (A) of 12.7 deg’ or constant relative image area
of 113 c*, calculated as Af i (for further details see
Methods).
As Fig. l(A, B) shows, Michelson contrast sensitivity
was better for gratings than for noise stimuli at all spatial
frequencies and grating areas studied.
Michelson contrast sensitivity for gratings with con-
stant stimulus area (12.7 deg?) had a band-pass shape
reaching its maximum at about 2 c/deg. At high spatial
frequencies contrast sensitivity was reduced by the
point spread function of the ocular optics (Campbell &
 RESEARCH NOTE 1433

103 + grating12.7 dog2

-D- grating 113 c 2
2
nolsa 12.7 deg
2
nois. 113 c
lo27

E
u) 101: 10’ :

z
g
._ . A . B
100 I . . . . . . ..I . . . ..‘“I 100 rrl
1 o- l 1 00 1 01 102 10-l 1 00 1 0’ 1 02

v) 101: lOl_

E
. c D
100 I . .‘..‘.‘1 . . ...fi 1001 * . . . . ..I . “...‘.1 . . ““-7
1 o-1 1 00 1 01 102 10-l 1 00 1 0’ 1 02

* 103- 103-

E”
0)
E 102- 102-
W

101 - 101. . . .-....r . “.“.V . . .‘rrrl

1 o-l d3kn0 s&Dtial
1 tre&c~2 o(cldeg)
-1 1 00 10’ 1 02

FIGURE 1. Binocular contrast sensitivity as a function of median spatial frequency for vertical cosine gratings and noise
stimuli. The set of stimuli with constant stimulus area of 12.7 deg2 was obtained by using a circular stimulus window with a
diameter of 16 cm at a viewing distance of 228 cm. In the other stimulus set relative image area was kept constant at I13 c2
by reducing the stimulus diameter in inverse proportion to increasing spatial frequency. The circular stimulus windows had
diameters of 12, 8.57, 6,4.29, and 3 cm and corresponding spatial frequencies of 1, I .4, 2, 2.8 and 4 c/cm, respectively. Viewing
distances were 57.3 and 319 cm, except for the spatial frequency of 0.5 c/deg for which viewing distance was 28.6 cm, stimulus
diameter 12 cm, and spatial frequency 1 c/cm. Contrast sensitivity refers to the inverse of Michelson threshold contrast in (A)
and (B), to the inverse of RMS threshold contrast in (C) and (D), and to the inverse of contrast energy at threshold in (E)
and (F). Michelson contrast equals to (L,,, - L,,,)/(L,,, + L,,,). Contrast energy was calculated by applying the following
equation: E = XZ C2(x,y)p2, where C(x,y) = [L(x,y) - LO]/LO is the local contrast, L(x,y) is the local luminance, p2 is the area
of a pixel in solid degrees, and L, is the luminance averaged across the pixel. RMS contrast was then calculated as J(E/A ).
Median spatial frequency (,f,,) refers to the median of the energy spectrum of the stimulus (see Methods). Subjects were HK
in (A), (C) and (E). and KT in (B), (D) and (F), respectively.

Gubish, 1966; Deeley, Drasdo & Charman, 1991) and In Fig. l(A, B) Michelson contrast sensitivity for
sampling limitations of the retina (Curcio, Sloan, noise stimuli had a low-pass shape irrespective of
Packer, Hendrickson & Kalina, 1987; Rovamo & Virsu, whether stimulus area or relative image area was
1979). At low spatial frequencies the small number of constant. The bandwidth of noise stimuli was 0.5-l
square cycles reduced contrast sensitivity (Virsu & octave. Hence, they were probably detected on the
Rovamo, 1979). Accordingly, contrast sensitivity for basis of their higher spatial frequencies when median
gratings with constant relative image area (113 c’) did spatial frequency was 0.25 or 0.5 c/deg. Thus, the
not decrease at low spatial frequencies. low-pass shape with constant stimulus area agrees with
 RMS sensitivity Michelson sensitivity

t f . . . ..,.: . . . ....$ . . ~~
0

n
 RESEARCH NOTE 1435

area of 12.7 deg’ shown in Fig. I(E, F) were identical or noise stimuli. High contrast energy sensitivity for
in shape to their corresponding RMS contrast sensi- spots evidently results from their small areas.
tivity functions in Fig. l(C, D). There was only a
shift in vertical direction resulting from squaring of
DISCUSSION
RMS contrast sensitivity and division by the stimulus
area. Our experiments with vertical cosine gratings, narrow-
For stimuli with constant relative image area of 113 c2 band two-dimensional noise stimuli and Gaussian spots
the shape of the energy sensitivity function was affected showed that, irrespective of the median spatial fre-
by the decrease of stimulus area with spatial frequency. quency, Michelson contrast sensitivity was almost equal
Therefore, in comparison with the shapes of the RMS for noise stimuli and spots but best for gratings. How-
contrast sensitivity functions in Fig. l(C, D). energy ever, RMS contrast sensitivity was almost equal for
sensitivity was reduced by the large stimulus area at low gratings and noise stimuli but lowest for spots. Contrast
spatial frequencies and increased by the small stimulus energy sensitivity, on the other hand, was similar for
area at high spatial frequencies. gratings and noise stimuli irrespective of stimulus area.
As Fig. l(E, F) shows, contrast energy sensitivities However, at each spatial frequency contrast energy
were better for stimuli with constant stimulus area sensitivity was better the smaller the stimulus area. in
(12.7 deg”) at spatial frequencies <3 c/deg where their agreement, contrast energy sensitivity was best for spots,
relative image area was < 113 c’ whereas energy sensi- because their areas were smallest at all median spatial
tivities were lower for stimuli with constant stimulus area frequencies.
at spatial frequencies > 3 c/deg where their relative The result that RMS contrast sensitivity was equal for
image area was > I 13 c’. Contrast energy sensitivity gratings and noise stimuli is in agreement with Moulden
functions measured with constant 12.7deg’ or 113~’ et al. (1990) and Nasanen, Tiippana and Rovamo (1992).
crossed each other at 3 c/deg, where relative image area Moulden et al. (1990) showed that the adaptive power of
was the same for both stimuli. various random-dot stimuli is determined by the stan-
In the experiments of Fig. 2 we measured increment dard deviation of their luminances, that is RMS con-
and decrement thresholds for Gaussian spots described trast, rather than the peak values, i.e. Michelson
by equation (2). Stimulus area decreased from 0.2 to contrast. In addition, Nasanen ef al. (1992) showed that
0.0004 deg2 when median spatial frequency increased the perceived contrasts of grating stimuli consisting of
from 0.636 to 16.5 c!deg. The relative image area was l-4 orientation components and consequently different
approximately constant at 0.1 c’. luminance distributions are equal when their physical
As Fig. 2 shows contrast sensitivity was lower for RMS contrasts are equal. The findings of Moulden et al.
decrements than increments at all median spatial (1990) and N&&en et al. (1992) together with our
frequencies irrespective of the contrast metrics used. In results suggest that RMS contrast is a more generally
Fig. 2(A, B) Michelson contrast sensitivities for spots applicable measure of contrast at and above threshold
were plotted as a function of median spatial frequency. than Michelson contrast.
Contrast sensitivity functions for increments and decre- Our finding that Michelson contrast sensitivity was
ments were low-pass in shape, resembling the shape of better for gratings than for noise stimuli can be ex-
the contrast sensitivity function for cosine gratings with plained as follows: RMS contrast is proportional to the
constant relative image area in Fig. l(A, B). This is in standard deviation of the stimulus luminance distri-
agreement with the fact that the relative image area for bution whereas Michelson contrast is proportional to
spots was constant. approx. 0.1 c’ at all spatial frequen- the maximum luminance deviations of the stimulus. The
cies. However, in quantitative terms, Michelson contrast proportion of pixels with luminances close to the maxi-
sensitivity was similar for spots and noise stimuli but mum and minimum luminances is greater in gratings
clearly higher for gratings. than in noise stimuli. Therefore to obtain the same
Figure 2(C, D) shows increment and decrement sensi- standard deviation a smaller maximum deviation is
tivities expressed in terms of RMS contrast. For spots needed in gratings than in noise stimuli. In accordance,
RMS contrast sensitivities were lower than Michelson when cosine gratings and noise stimuli were both at
contrast sensitivities, because the minimum luminance of threshold, and their RMS contrast were therefore equal,
spot stimuli is equal to the average luminance and Michelson contrast was smaller for cosine gratings than
therefore Michelson contrast gives smaller threshold for noise stitnuli. Consequently, Michelson contrast
values than RMS contrast. In addition, RMS contrast sensitivity was better for gratings than noise stimuli.
sensitivity was clearly lower for spots than for gratings The result that contrast energy sensitivity at each
and noise stimuli. Virsu and Rovamo (1979) have shown spatial frequency was better the smaller the stimulus area
that contrast sensitivity for cosine gratings increases with can be explained by decrease in efficiency of spatial
relative image area. Therefore, lower sensitivities for integration with increasing relative image area in fovea1
spots were due to small relative image area. vision (Nlsanen, Kukkonen & Rovamo, 1993). Thus.
In Fig. 2(E, F) the data is expressed as contrast energy the limiting factor of contrast energy sensitivity seems to
sensitivity. For spots contrast energy functions had a be the sampling aperture of the human visual system.
band-pass shape with a maximum at 4-8 c/deg. Contrast Therefore, fovea1 stimuli with various contrast energy
energy sensitivities were better for spots than for gratings profiles produce practically identical contrast energy
1436 RESEARCH NOTE
sensitivity functions with increasing relative image area
(Nlsanen et al., 1993) reflecting true changes in efficiency
of spatial integration. Moreover, as the value of contrast
energy is not affected by areas where locai contrast is
equal to zero, it takes into account only those stimulus
areas which actually stimulate the visual system.
RMS contrast is problematic when contrast energy is
unevenly distributed across the stimulus field, because in
such cases the value of RMS contrast depends on the
choice of the area over which it is calculated. When the
local contrast of the stimulus decreases with increasing
distance from the stimulus centre, as is the case, for
example, in Gaussian weighted gratings RMS contrast
decreases with increasing area over which it is calculated.
Now, the increase of RMS contrast sensitivity with
increasing stimulus area could result not only from
spatial integration but from decrease of RMS contrast
with increasing stimulus area. This problem can,
however, be avoided by using contrast energy.
In conclusion, as our results showed, the contrast
metrics chosen had a significant effect on the appearance
of contrast sensitivity functions for different kinds of
stimuli. There may be good reasons for using any of the
contrast measures discussed above. However, contrast
energy does not seem to have the intrinsic limitations
typical to Michelson and RMS metrics. Thus, it can be
considered universally more useful than the other two
metrics.
Burgess, A. E. & Ghandeharian, H. (1984a). Visual signal detection.
I. Ability to use phase information. Journal of the Optical Society 01
America, A, f, 90@-905.
Burgess, A. E. & Ghandeharian, H. (1984b). Visual signal detection.
II. Signal-location identification. Journal of the Optical Society af
America, A, I, 906-910.
Campbell, F. W. & Gubish, R. W. (1966). Optical quality of the human
eye. $oumol of PhysioIogy, London, 186, 558.--578.
Campbell, F. W. & Robson, J. G. (1968). Application of Fourier
analysis to the visibility of gratings. Journal of Physiology, 197,
55 l-566.
Curcio, C. A., Sloan, K. R., Packer, 0.. Hendrickson, A. E. & Kalina,
R. E. (1987). Distribution on cones in human and monkey retina,
Individual variability and radial asymmetry. Scrence. 2_??3h,579 581.
Deeley, R. J., Drasdo, N. & Charman, W. N. ( 1991). .4 simple
parametric model of the human ocular modulation transfer func-
tion. Ophthalmic and Physiologic Optics, II, 91-93.
Kersten, D. (1984). Spatial summation in visual noise. Vision Research,
24, 1977-1990.
Koenderink. J. J. & van Doorn, A. J. (1974). Detectability of two-
dimensional band limited noise. Vision Research, 14. 5 15 518.
Legge. G. E., Kersten, D. & Burgess, A. E. (1987). Contrast discrimi-
nation in noise. Journal af the Optical Socie/y of America 4.
391404.
Moulden, B., Kingdom, F. & Gatley, L. F. (1990). The standard
deviation of luminance as a metric for contrast in random-dot
images. Perception, 19, 79-101.
Nasiinen, R., Kukkonen, H. & Rovamo, J. (1991). Effect of image
structure on contrast detectability. Jnuestigafiue Ophtha~mo~og_~,
and
Visual Science, 32, 1270.
N&&en, R., Kukkonen, H. & Rovamo, J. (1993). Spatial integration
of band-pass filtered patterns in noise. Vision Research, 33,903-911.
NlsHnen, R., Tiippana, K. & Rovamo. J. (1992). Perceived contrast of
two-dimensional compound gratings. Perceptual (Suppf. 2). 21, 92.
Pollehn, H. & Roehrig, H. (1970). Effect of noise on the modulation
transfer function of the visual channel. Journal of’the Optical Sobet))
a/ America, 60, 842-848.
Rovamo, J. & Virsu, V. (1979). An estimation and application of the
human cortical magnification factor. ~xperjmenta~ Brain Research,
3% 495-510.
Rovamo, J., Franssila, R. & Nasiinen, R. (1992). Contrast sensitivity
as a function of spatial frequency, viewing distance and eccentricity
with and without spatial noise. Vision Research, 32, 631-637,
Stromeyer, C. F. & Julesz, B. (1972). Spatial-frequency masking in
vision: Critical bands and spread of masking. Journal af the Optical
Society of America, 62, 1221-1232.
Virsu, V. & Rovamo, J. (1979). Visual resolution, contrast sensitivity,
and the cortical magni~cation factor. ~~~~~rnenr~ Bruin Research,
37. 475494.
Watson, A. B., Barlow, H. B. & Robson, J. G. (1983). What does the
eye see best? Nature, 302, 419422.
Wetherill, G. B. & Levitt, H. (1965). Sequential estimation of points
on a psychometric function. British Journal of ~at~ematiea~ and
Statistical P,sychology, 18, l--l 0.
Ackno~iedgements-We thank The Finnish Culture Foundation,
The Academy of Finland, The Finnish Ministry of Education, The
Association of Finnish Ophthalmic Opticians, The Information Centre
of Optics Business, and The Optics Division of Instrumentarium
Corporation for their support.