Scaling of Horizontal and Vertical Fixational Eye Movements
Jin-Rong Lianga,b ∗ , Shay Moshel a , Ari Z. Zivotofskyc ,
Avi Caspic , Ralf Engbertd , Reinhold Kliegld and Shlomo Havlina
†
arXiv:cond-mat/0410615v1 [cond-mat.stat-mech] 24 Oct 2004
a
Minerva Center and Department of Physics, Bar-Ilan University, Ramat-Gan 52900, Israel
Department of Mathematics, East China Normal University, Shanghai 200062, P.R. of China
c
Gonda Brain Research Center, Bar-Ilan University, Ramat-Gan 52900, Israel
d
Department of Psychology, University of Potsdam, P.O.Box 601553, 14415 Potsdam, Germany
b
visual field. Drifts are slow movements, with a mean amplitude within a range of 1.2 − 9 min arc [2], away from
a fixation point. Each instance of drift is necessarily terminated by a microsaccade (cf. Fig.1). Microsaccades
are rapid small amplitude movements ranging between
1′ and 60′ arc and occur at a typical mean rate of 1 to
2 per second [4]. Microsaccades seem to reposition the
eye on the target. Tremor (or physiological nystagmus)
is a high-frequency (ranging from 50 to 100 Hz [2]) oscillations of the eye typically less than 0.01 deg, i.e., less
than the size of one photoreceptor, and is superimposed
on drift. Tremor serves to continuously shift the image
on the retina, thus calling fresh retinal receptors into operation. If an image is artificially fixed on the retina it
fades and disappears within a few seconds [5]. Tremor
causes every point of the retinal image to move approximately the distance between two adjacent foveal cones
in 0.1 seconds and thus causes the image of an object
to constantly stimulate new cells in the fovea [6]. Drift
and tremor movements are rather irregular and show statistical properties of a random walk [7]. Microsaccades,
however, create more linear movement segments embedded in the eyes’ trajectories during fixational movements.
There is evidence that microsaccades are (i) persistent
and anti-persistent at different time scales [3], (ii) show a
characteristic signature of suppression and overshoot in
response to visual change [4,8], and (iii) orient themselves
according to covert shifts of attention [1].
Although finding the specific function of fixational eye
movements has been a long-standing and controversial
topic in eye movements research [9], our concern is not
the purpose of such movements but rather the dynamical behavior of fixational eye movements and if there
is some difference between horizontal and vertical fixational eye movements. In this article we mainly investigate these questions using the detrended fluctuation analysis [10,11], a technique used to detect possible long-term
correlations in time series. We find that the persistence of
horizontal and vertical fixational eye movements exhibit
pronounced different behavior mostly due to the effect
of the microsaccades. This result is in good agreement
with the neurophysiological fact that horizontal and vertical components of saccades are controlled by different
brain stem nuclei [12]. Our study indicates that after removing the microsaccades the scaling behavior of both
components becomes similar. These findings may further
elucidate the mechanisms underlying effects of microsac-
ABSTRACT
Eye movements during fixation of a stationary target prevent the adaptation of the photoreceptors to
continuous illumination and inhibit fading of the image. These random, involuntary, small, movements
are restricted at long time scales so as to keep the
target at the center of the field of view. Here we
use the Detrended Fluctuation Analysis (DFA) in order to study the properties of fixational eye movements at different time scales. Results show different scaling behavior between horizontal and vertical
movements. When the small ballistics movements, i.e.
micro-saccades, are removed, the scaling exponents in
both directions become similar. Our findings suggest
that micro-saccades enhance the persistence at short
time scales mostly in the horizontal component and
much less in the vertical component. This difference
may be due to the need of continuously moving the
eyes in the horizontal plane, in order to match the
stereoscopic image for different viewing distance.
I. INTRODUCTION
When we view a stationary scene, our eyes perform
extremely small autonomic movements. These fixational
(or miniature) eye movements are produced involuntarily
and are characterized by three different types of movements: (i) high-frequency small amplitude tremor, (ii)
slow drift, and (iii) fast microsaccades [1,2]. Generally,
they serve to counteract retinal adaptation by generating small random displacements of the retinal image in
stationary viewing. Studies of fixational eye movements
have been going on since the 1950s, but the role of the
drift, tremor and microsaccadic movements in the visual
process is not yet fully understood [2,3].
When fixating an object its image falls on the fovea,
the region of highest visual acuity in the center of the
∗
†
jrliang@math.ecnu.edu.cn
havlin@ophir.ph.biu.ac.il
1
resetting microsaccades. In this subject, for example, the
drift in the horizontal movement occurred typically to the
right and the microsaccades to the left (Fig.1).
cades on perception and attention [3,4,8] and their role
in the neurophysiology of vision [13–16]. In addition, in
many pathological states the fixation system can be disrupted by slow drift, nystagmus, or involuntary saccades.
However, because all three of these occur in healthy individuals it may be difficult to determine if there is truly
an abnormality present. Thus, further characterizing of
the fixational system may be useful in clinical evaluation
of such dysfunction.
III. METHODS OF ANALYSIS
To study the dynamical behavior of fixational eye
movements we employ the detrended fluctuation analysis
(DFA) which was developed to quantify long-term powerlaw correlations embedded in a nonstationary time series
[10]. The DFA method has been successfully applied to
research fields such as cardiac dynamics [17,11,18–20], human gait [21], climate temperature fluctuations [22,23]
and neural receptors in biological systems [24]. Here we
apply this method to the velocity series derived from the
position series of fixational eye movements. For a position series xi , i = 1, · · · , N + 1, of a horizontal or vertical movement, we first calculate its velocity series vi by
vi = T0 (xi+1 − xi ), where T0 is the sampling rate; in our
experiments T0 = 500 Hz. We chose to use a two-point
velocity in order to avoid any smoothing and clearly characterize the direction and magnitude of a movement. For
other definitions of velocity see [4].
We first calculate the integrated series as a profile
II. DATA
Data was collected from five normal subjects. Eye
movements for these participants were recorded using an
EyeLink-II system with a sampling rate of 500 Hz and
an instrument spatial resolution < 0.005◦ . The subjects
were required to fixate a small stimulus with a spatial
extent of 0.12◦ or 7.2 arc min (3 × 3 pixels on a computer
display, black square on a white background). Each participant performed about 100 trials with a duration of 3
seconds and total of 474 trials were obtained [3]. The
recording of each trial includes position trajectories of
horizontal and vertical components of left eye and right
eye movements.
Y (k) =
k
X
[vi − hvi], k = 1, · · · , N.
(1)
i=1
position in degree
0.2
tremor
horizontal
Subtraction of the mean hvi of the whole series is not
compulsory since it would be eliminated by the detrending in the third step [25]. Thus Y (k) in Eq.(1) represents
actually the “position”.
We then divide the profile Y (k) of N elements into
Nt = int(N/t) non-overlapping segments of equal length
t, where int(N/t) denotes the maximal integer not larger
than N/t. Since the length N of the series is often not a
multiple of the considered time scale t, a short part at the
end of the profile may remain. In order not to disregard
this part of the series, the same procedure is repeated
starting from the opposite end. Therefore, 2Nt segments
are obtained altogether.
Next, we determine in each segment the best polynomial fit of the profile and calculate the variance of the
profile from these best polynomials
0
drift
-0.2
microsaccade
vertical
0.2
0
-0.2
0
500
1000
2000
1500
time (ms)
2500
3000
FIG. 1. Eye position simultaneous recording of horizontal
and vertical components of left eye movements. The traces
show microsaccades, drift and tremor in eye position. In the
horizontal tracing, up represents right and down represents
left; in the vertical tracing, up represents up and down represents down movements.
t
F 2 (ν, t) ≡
1X
{Y ((ν − 1)t + i) − yν (i)}2
t i=1
(2)
for each segment ν, ν = 1, · · · , Nt , and
Figure 1 shows a typical simultaneous recording of horizontal and vertical miniature eye movements for the left
eye from one subject. The horizontal and vertical movements (upper and lower traces in the figure, respectively)
exhibit an alternating sequence of slow drift and resetting
microsaccades. Usually, the subjects show an individual
preponderance regarding the direction of these drifts and
t
F 2 (ν, t) ≡
1X
{Y (N − (ν − Nt )t + i) − yν (i)}2
t i=1
(3)
for ν = Nt + 1, · · · , 2Nt , where yν is the fitting polynomial in segment ν. If this fitting polynomial is linear,
2
-1
-1
10
(a)
F(t)
F(t)
10
-2
10
-3
10
2
10
-2
10
t
10
3
10
-1
1
2
10
10
t
3
10
-1
10
10
(c)
(d)
without microsaccades
-2
10
-3
10
(b)
-3
1
10
F(t)
This computation is repeated over all possible interval
lengths. Of course, in DFA F (t) depends on the DFA
order n. By construction, F (t) is only defined for t ≥
n + 2. For very large scales, for example, for t > N/4,
F (t) becomes statistically unreliable because the number
of segments Nt for the averaging procedure becomes very
small. We therefore limit our results to [n, N/4].
Typically, F (t) increases with interval length t. We
determine the scaling behavior of the fluctuations by
analyzing log-log plots of F (t) versus t. A power law
F (t) ∝ tα , where α is a scaling exponent, represents the
long-range power-law correlation properties of the signal.
If α = 0.5, the series is uncorrelated (white noise); if
α < 0.5, the series is anti-correlated; if α > 0.5, the series
is correlated or persistent.
crossovers and the exponents at small time scales become
similar (Figs. 2(c) and (d)). This result indicates that
microsaccades strongly influence the horizontal components in fixational eye movements. Note, the close similarity of the fluctuation function F (t) in the different 3
seconds trials, in particular after removing the microsaccades, indicates that the scaling exponent is a stationary
and significant characteristic of the eye movement.
F(t)
then it is the first-order detrended fluctuation analysis
(DFA1). This eliminates the influence of possible linear
trends in the profile on scales larger than the segment
[10]. In general, in the nth order DF A (DFAn), yν is the
best nth-order polynomial fit of the profile in segment ν.
Therefore, linear, quadratic, cubic, or higher order polynomials can be used in the fitting procedure. Since the
detrending of the original time series is done by the subtraction of the polynomial fits from the profile, different
order DFA differ in their capability of eliminating trends
of order n − 1 in the series.
Finally, the fluctuation F (t) over the time windows of
size t is determined as a root-mean-square of the variance
v
u
2Nt
u 1 X
F (t) = t
F 2 (ν, t).
2Nt ν=1
without microsaccades
-2
10
-3
1
10
2
10
t
3
10
10
1
10
2
10
t
FIG. 2. Fluctuation functions obtained by DFA2 for horizontal and vertical eye movements from the right eye of a
typical participant: (a) horizontal; (b) vertical; (c) horizontal
[same data as (a)] after removing microsaccades; (d) vertical
[same data as (b)] after removing microsaccades.
In Fig.3 we show the histograms of the scaling exponents for the short time scales, for all trials with and
without microsaccades from the left eyes of all participants. From this plot we notice that, at the short time
scale, the horizontal and vertical components exhibit persistent behavior (α > 0.5) where the horizontal components are much stronger correlated than the vertical. The
average value of the scaling exponents for all trials is
0.76 for the vertical components and 1.1 for the horizontal components (See Table I(A)). The scaling exponents of horizontal components show a broader distribution than the vertical components. However, after removing microsaccades, the fluctuations of horizontal components and the corresponding scaling exponents have
a pronounced change to a narrow distribution while the
vertical components change very little (see Figs.2 (b) and
(d); and Figs.3 (b) and (d)). When comparing the scaling
exponents for the original horizontal series with the scaling exponents for the horizontal removed microsaccades
series, we find that the scaling exponents decrease from
an average value around 1.1 to 0.74, while for the vertical
components the scaling exponents decrease from an average value around 0.76 to 0.74 (Table I). Horizontal and
IV. ANALYSIS OF FIXATIONAL EYE
MOVEMENTS
We applied DFA1-4 to all velocity records derived from
the horizontal and vertical components. Since the scaling exponents of the fluctuation functions obtained by
DFA1-4 are similar, we show here the DFA2 results as a
representative of the DFA analysis.
As can be seen from Figs.2 (a) and (b), the fluctuation
functions of horizontal components have pronounced differences from the fluctuation function of vertical components. This is expressed by several characteristics, which
can be observed. There is a broader range of exponents
in the horizontal compared to the vertical. The crossover
times, from large exponents (at short time scales) to
smaller exponents (at large time scales) in horizontal,
also show a broader range compared with the vertical.
Moreover, the scaling exponents at short time scales (between 12 millisecond and 40 milliseconds) for horizontal, are typical larger than the corresponding exponents
for vertical. However, if we remove microsaccades [26]
the fluctuations of both components, the corresponding
3
3
10
vertical become similar after removing microsaccades.
250
(a)
200
150
100
Frequency
Frequency
250
50
0
0.5
1
α
150
100
0
1.5
0.5
1
α
1.5
250
(c)
200
150
100
50
Frequency
Frequency
(b)
200
50
250
0
We thus conclude that, microsaccades in the horizontal components are more dominant than in the vertical
direction in fixational eye movements. The microsaccades enhance the persistence mostly in the horizontal
components at the short time scales. At the long time
scales both horizontal and vertical components are antipersistence and less affected by the microsaccades.
To further test if the above results are indeed affected
by microsaccades, we randomly removed parts of the series under study with the same length as the removed
microsaccades and repeated the DFA analysis. We found
that this procedure does not influence the scaling exponents. Thus, the scaling difference between the series
with and without microsaccades is indeed due to microsaccades.
Finally, we tested if the effect of microsaccades can be
seen also in the power spectral density. To this end we
analyzed the power spectra of the horizontal and vertical
velocity series, for the right eye of a typical participant,
for all trials with and without microsaccades. Results
are shown in Figs.4 (a) and (b) where the microsaccades
are included. The power spectral density of horizontal
and vertical components are found to be different (Figs.4
(a) and (b)). After removing the microsaccades the components become similar (Figs.4 (c) and (d)). This finding also indicates that the effect of the microsaccades in
the horizontal component is stronger than in the vertical. Note, that this effect is seen much clearer in the
DFA curves where only a few trials (of 3 sec) are sufficient to distinguish between the horizontal and vertical
eye movements.
(d)
200
150
100
50
0.5
1
1.5
0
0.5
1
1.5
α
α
FIG. 3. Histograms of the scaling exponents α obtained by
DFA2 at the short time scales [12, 40] ms for all the horizontal and vertical trials (with duration of 3s) with and without
microsaccades from the left eyes of all participants: (a) horizontal; (b) vertical; (c) horizontal without microsaccades; (d)
vertical without microsaccades.
We find that at long time scales (between 300 and
600 milliseconds), the horizontal and vertical components
show anti-persistence behaviour (α < 0.5) (see Table I
(A)), with no significant differences between them. After removing the microsaccades the scaling exponents
at the long time scales, remain almost the same as before. The horizontal components become slightly less
anti-persistent than vertical (see Table I(A)).
1.5
1.5
(b)
(a)
B
HL
HR
VL
VR
S(f)
0
0
Long time scale
0.29 ± 0.14
0.31 ± 0.14
0.34 ± 0.13
0.30 ± 0.12
removed
0.26±0.11
0.26±0.10
0.36±0.14
0.35±0.13
0.5
50
100 150 200
Frequency (Hz)
0
0
250
1.5
50 100 150 200
Frequency (Hz)
250
1.5
(d)
(c)
1
1
S(f)
Short time scale
1.13 ± 0.26
1.05 ± 0.25
0.76 ± 0.08
0.76 ± 0.09
Microsaccades
0.74 ± 0.06
0.73 ± 0.05
0.74 ± 0.05
0.74 ± 0.04
0.5
S(f)
component
HL
A
HR
VL
VR
1
S(f)
1
TABLE I. Average values of the scaling exponents obtained
by DFA2 for all fixational eye movements we measured. HL
= horizontal movements of left eyes, HR = horizontal movements of right eyes, VL = vertical movements of left eyes, and
VR = vertical movements of right eyes.
0.5
0
0
0.5
50
100 150 200
Frequency (Hz)
250
0
0
50
100 150 200
Frequency (Hz)
250
FIG. 4. Power spectral density for the velocity series derived from the horizontal and vertical components from the
right eye of one typical participant. (a) horizontal with microsaccades; (b) vertical with microsaccades; (c) horizontal
after removing the microsaccades; (d) vertical after removing
the microsaccades
4
[5] L.A. Riggs, F. Ratliff, J.C. Cornsweet, & T.N. Cornsweet,
Journal of the Optical Society of America, 43, 495 (1953).
[6] B. P. Olveczky, S. A. Baccus & M. Meister, Nature 5,
401 (2003).
[7] R. Engbert, R. Kliegl, Binocular coordination in microsaccades. In: J. Hyönä, R. Radach, H. Deubel (eds.)
The Mind’s Eyes: Cognitive and Applied Aspects of Eye
Movements. Elsevier, Oxford, (pp. 103-117) (2003).
[8] M. Rolfs, R. Engbert & R. Kliegl, Psychological Science
(in press).
[9] E. Kowler & R. M. Steinman, Vision Research, 20
273(1980).
[10] C. -K Peng, S. V. Buldyrev, S. Havlin, M. Simons, H.
E. Stanley and A. L. Goldberger, Phys. Rev. E 49, 1685
(1994). S. V. Buldyrev et al. Phys. Rev. E 51 , 5084
(1995).
[11] A. Bunde, S. Havlin, J.W. Kantelhardt, T. Penzel, J.H.
Peter, K. Voigt, Phys. Rev. Lett. 85, 3736(2000).
[12] D.L. Sparks, Nature Reviews Neuroscience 3, 952-964
(2002).
[13] S. Martinez-Conde, S. L. Macknik, and D. H. Hubel, Nature Neuroscience, 3, 251 (2000).
[14] S. Martinez-Conde, S. L. Macknik, and D. H. Hubel, Preceedings of the National Academy of Sciences U.S.A. 99,
13920 (2002).
[15] W. Bair, & L. P. O’Keefe, Visual Neuroscience 15, 779
(1998).
[16] M. Greschner, M. Bongard, P. Rujan, & J. Ammermuller,
Nature 5, 341 (2002).
[17] C.-K. Peng, S. Havlin, H. E. Stanley and A. L. Goldberger, Chaos 5, 82(1995).
[18] P. Ch. Ivanov, M. G. Rosenblum, C.-K. Peng, J. E. Mietus, S. Havlin, H. E. Stanley and A. L. Goldberger, Nature (London) 383, 323(1996).
[19] P. Ch. Ivanov, A. Bunde, L. A. Nunes Amaral, S. Havlin,
J. Fritsch-Yelle, R.M. Baevsky, H. E. Stanley and A. L.
Goldberger, Europhys. Lett. 48, 594(1999).
[20] Y. Ashkenazy, P. Ch. Ivanov, S. Havlin, C.-K. Peng, A.
L. Goldberger, and H. E. Stanley Phys. Rev. Lett. 86,
1900(2001).
[21] J. M. Hausdorff, C.-K. Peng, Z. Ladin, J. Wei and
A.L.Goldberger, J. Appl. Physiol. 78, 349(1995).
[22] E. Koscielny-Bunde, A. Bunde, S. Havlin, H. E. Roman,
Y.Goldreich, and H. -J. Schellnhuber, Phys. Rev. Lett.
81, 729 (1998).
[23] A. Bunde, J. Eichner, R. Govindan, S. Havlin, E.
Koscielny-Bunde, D. Rybski and D. Vjushin, Nonextensive Entropy-Interdisciplinary Applications, edited by M.
Gell-Mann and C. Tsallis, New York Oxford University
Press, 2003.
[24] S. Bahar, J. W. Kantelhardt, A. Neiman, H. H. A. Rego,
D. F. Russell, L. Wilkens, A. Bunde, and F. Moss, Europhys. Lett. 56, 454(2001).
[25] J. W. Kantelhardt, E. Koscielny-Bunde, H. H. A. Rego,
S. Havlin, and A. Bunde, Physica A 295, 441 (2001).
[26] Microsaccades were detected from the velocity series. We
remove velocities above a threshold of 15 deg/sec. On
average between 6 and 7 microsaccades are removed for
each 3 second trial.
V. DISCUSSION
When the visual world is stabilized on the retina, visual
perception fades as a consequence of neural adaptation.
During normal vision we continuously move our eyes involuntarily even as we try to fixate our gaze on a small
stimulus, preventing retinal stabilization and the associated fading of vision [1]. The nature of the neural activity
correlated with microsaccades at different levels in the visual system has been a long standing controversy in eyemovements research. Steinman [27] showed that a person
may select not to make microsaccades, and still be able to
see the object of interest, whereas Gerrits & Vendrik [28]
and Clowes [29] found that optimal viewing conditions
were only obtained when both microsaccades and drifts
were present. Since microsaccades can be suppressed voluntarily in high acuity observation tasks [30,31], it was
concluded that microsaccades serve no useful purpose and
even that they represent an evolutionary puzzle [4,9].
Our study using DFA suggests that microsaccades play
different roles on different time scales in vertical and horizontal components in the correction of eye movements,
consistent with [3]. Moreover we show that due to microsaccades there is also different scaling behaviour in
horizontal and vertical fixational eye movements. Our
results suggest that microsaccades at short time scales,
enhance the persistence mostly in horizontal movements
and much less in the vertical movements.
Our findings that the persistence in horizontal and vertical fixational eye movements, which are controlled by
different brain stem nuclei, exhibit pronounced different
behavior also show that the role of microsaccades in horizontal movements are more dominant. These findings
may provide better understanding of the recent neurophysiological findings on the effects of microsaccades on
visual information processing [13–16].
We thank Tomer Kalisky for his assistance in applying the DFA method, and Samuel Ron and Lance M.
Optican for constructive suggestions on the manuscript.
This work is supported in part by grants KL955/6
and KL955/9 (Deutsche Forschungsgemeinschaft) and by
NSFC (No. 10271031) and the Shanghai Priority Academic Discipline.
[1] S. Martinez-Conde, S. L. Macknik and D. H. Hubel, Nature Reviews Neuroscience 5, 229 (2004).
[2] F. Møller, M. L. Laursen, J. Tygesen, A. K. Sjølie,
Graefe’s Arch. Clin. Exp. Ophthalmol, 240, 765 (2002).
[3] R. Engbert & R. Kliegl, Psychological Science (2004), in
press.
[4] R. Engbert & R. Kliegl, Vision Research, 43, 1035 (2003).
5
[27] R. M. Steinman, Science 181, 810 (1973).
[28] H. J. M. Gerrits, A. J. H. Vendrik, Vision Research, 10,
143 (1970).
[29] M. B. Clowes, Optica Acta 9, 65 (1962).
[30] B. Bridgeman and J. Palca, Vision Research 20, 813
(1980).
[31] B. J. Winterson and H. Collewijn, Vision Research 16,
1387 (1976).
6