See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/5502504

Shift-Tolerance Property of an Optical Double-

Random Phase-Encoding Encryption System

Article in Applied Optics October 2000

DOI: 10.1364/AO.39.004788 Source: PubMed

CITATIONS READS

61 91

4 authors, including:

Ching-Cherng Sun Arthur Chiou

National Central University National Yang Ming University
253 PUBLICATIONS 2,018 CITATIONS 194 PUBLICATIONS 1,663 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ching-Cherng Sun on 20 December 2016.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Shift-tolerance property of an optical
double-random phase-encoding encryption system

Bor Wang, Ching-Cherng Sun, Wei-Chia Su, and Arthur E. T. Chiou

We investigate the shift-tolerance property of the decrypting phase mask in an optical double-random
phase-encoding encryption system. A simple method for improving the shift tolerance of the phase mask
is proposed. We demonstrate how the robustness to data loss of the encrypted image extends the shift
tolerance of the decrypting phase mask. The signal-to-noise ratio is calculated. Both a computer
simulation and an experiment are presented. 2000 Optical Society of America
OCIS codes: 070.4560, 070.6110.

1. Introduction property. In Section 4 we investigate the SNR prop-

Optical information processing systems have shown
great potential for security applications.110 Re-
cently, an optical encryption technique by use of
double-random phase encoding was proposed.3 This
method uses two independent random phase masks:
one on the input plane and one on the Fourier plane.
The encryption technique has good tolerance to data
loss of the encrypted image, to binarization of the
encrypted image, and to different types of noise.7,8
These fault-tolerance properties are valuable for op-
tical security systems. However, the optical double-
random phase-encoding encryption system with a
VanderLugt 4-f system requires extremely precise
alignment,11 especially in the decryption process, ow-
ing to the random phase characteristic. When all
the encrypted data are used for decryption, the image
cannot be decrypted if the decrypting phase mask
deviates from the matching position by more than one
pixel size. In this paper we propose a simple method
to improve the shift tolerance of the decrypting phase
mask and investigate the signal-to-noise ratio SNR
of the decrypted image. We briefly review the
double-random phase-encoding technique in Section
2. In Section 3 we investigate the shift-tolerance
B. Wang, C.-C. Sun ccsun@ios.ncu.edu.tw, and W.-C. Su are
with the Institute of Optical Sciences, National Central University,
Chung-Li, 320 Taiwan. A. E. T. Chiou is with the Institute of
Electrical Engineering, National Dong Hwa University, Hua-Lien,
Taiwan.
Received 8 November 1999; revised manuscript received 5 June
2000.
0003-693500264788-06$15.000
2000 Optical Society of America
erty when the decrypting mask is shifted. A com-
puter simulation and an experiment are presented in
Section 5.
2. Double-Random Phase-Encoding Technique
For simplicity, we use one-dimensional notation. Let
f x denote the image to be encrypted and qx the
encrypted image. Let x and u be two indepen-
dent white sequences, uniformly distributed in 0, 1.
The double-random phase encryption of the image f x
is obtained by the following two operations. First, the
image f x is multiplied by phase-mask function
expi2x. Second, the product f xexpi2x
is convoluted by function hx, which is the impulse
response of the transfer function expi2u. Thus
the encrypted image can be expressed as
qx f xexpi2x h x, (1)
where denotes the convolution operation.
The optical setup of the double-random phase-
encryption system is shown in Fig. 1a. The image
f x contacted with a random phase mask
expi2x is placed on input plane P1 and is illu-
minated by a coherent plane wave. On Fourier
plane P2 the Fourier transform of the product
f xexpi2x is multiplied by the second random
phase code expi2u. Through the inverse Fou-
rier transform the encrypted image q x is obtained
on output plane P3. Here we define the Fourier
transform in inverted coordinates as the inverse Fou-
rier transform. Similarly, the decryption scheme is
shown in Fig. 1b. The encrypted image qx is
placed at input plane Pd1 and is illuminated by a
coherent plane wave. The decrypting phase mask, a

4788 APPLIED OPTICS Vol. 39, No. 26 10 September 2000


Fig. 1. Schematic diagram of the optical double-random phase-
decrypted image can be expressed as10
SNR ND 22, (3)
where N is the pixel number of the mask. Equation
3 shows that the shift tolerance is D when the entire
encrypted image is used for decryption.
Now we investigate how the robustness to loss of the
encrypted data will improve the shift tolerance of the
decrypting phase mask. According to Eq. 1 the
double-random phase-encrypted image can be ex-
pressed as
M1
qx f nexpi2nh x n,
n0
(4)
where M is the sampling number of the image f x.
When only part of the encrypted image is used for
decryption, we can express the decryption process as
follows. The Fourier transform of the incomplete
image can be expressed as

encoding encryption system: a encryption setup, b decryption
M1
setup. x
qxrect f nexpi2n

n0
complex conjugation of expi2u, is placed on x
Fourier plane Pd2. On the output plane Pd3 the hx n rect
random phase-encoded image f xexpi2x is ob-
tained. When the original image is a real function, M1

the phase function expi2x can be removed by f nexpi2nexpi2nu

n0
an intensity-sensitive device.
expi2u sincu

3. Shift Tolerance of the Decrypting Phase Mask M1
As presented in Section 2, if the decrypting phase
mask on Fourier plane Pd2 in Fig. 1b is shifted a

n0
f nexpi2n

distance from the matching position, the Fourier N1

transform of the decrypted data can be expressed as expi2n

0

Qu, Duexpi2u u f ,
where Du denotes the Fourier transform of
(2) expi2sincu ,
f xexpi2x, is the wavelength, and f is the (5)
focal length of the Fourier lens. Since u is a ran- where we normalize the size of the encrypted image
dom function, expi2u u f will be q x to 1 and let 1 and where denotes the
randomly distributed when is larger than one pixel Fourier transform operation. If we remove the term
size of the mask. Thus the inverse Fourier trans- for u f from the summation 0 N1
in the
form of Qu, on plane Pd3 is also a random func- brackets {}, then Eq. 5 can be expressed as

qxrect
x


M1

n0
f nexpi2nsinc

f nexpi2n
f
N1

exp i2n u
f
exp i2 u
f
expi2nexpi2sincu

uf

M1
f nexpi2nsinc exp i2n u exp i2 u
f f f .
n0
Ru, n
(6)

tion, and the encrypted image cannot be decrypted. Since is a random function, the term Ru, n in
If the shift distance is less than one pixel size and Eq. 5 is a random function. When we multiply Eq.
the pixel size of the mask is D. The SNR of the 5 by the shifted decrypting phase function

10 September 2000 Vol. 39, No. 26 APPLIED OPTICS 4789

expi2u f and take the inverse Fourier-
transform, we can obtain the decrypted image ex-
pressed as
M1
fdx sinc f nexpi2n
f
n0
i2
exp n x n rx, n
f is used for decryption and the ideal condition is that
sinc f xexpi2x
f However, from Eq. 8, we know that the intensity of
exp i2
f
x
M1
n0
rx, n,
where rx, n is the inverse Fourier transform of Ru,
nexpi2u f and is still a random func-
tion. Thus the intensity of the decrypted image can
be expressed as
M1
fdx2 2 sinc2
f
f x2
n0
rx, n,
where r x, n is a uniform distributed white noise
generated from the nth sampling point of the original
image. From Eq. 8 it is easy to find that the image
can still be decrypted when the decrypting mask is
shifted by more than one pixel. Since Eq. 8 is the
normalized result, the practical extension of the im-
proved shift tolerance is fL, where L is the width of
the encrypted data used for decryption. Further-
more, if the decrypting mask is placed at the match
position 0, Eq. 8 can be rewritten as
M1
fdx2 2 f x2 rx, n.
n0
Equation 9 presents the property of the tolerance
to data loss of the encrypted image. Comparing Eqs.
8 and 9, we know that the improvement of the shift
tolerance is caused by the spectrum extension, owing
to the spatial restriction of the encrypted image.
4. Signal-to-Noise Ratio
In this section we investigate the SNR property of the
decrypted image when the decrypting mask is shifted
from the matching position. When all the encrypted
data are used and the shift distance of the decrypting
mask is less than one pixel size, the SNR is indepen-
dent of the parameters of the optical encryption sys-
tem. From Eq. 3 we know that the SNR depends
on the shift distance and on the pixel size of the
decrypting mask. Thus use of a phase mask with
lower spatial frequency will produce a larger SNR
value when there is no external noise is added.
However, as proposed in Ref. 7, the noise on the
decrypted image is always a white noise no matter
what kind of external noise is added. That is be-
cause the noise is uniformly distributed on the output
plane caused by the random phase mask. Thus the
distribution range of the noise is proportional to the
spatial frequency of the decrypting mask, and we
4790 APPLIED OPTICS Vol. 39, No. 26 10 September 2000
have to use a mask with higher spatial frequency to
obtain a larger SNR value if the external noise is
added. As a result, the choice of the spatial fre-
quency of the mask depends on the total energy of the
external noise.
Now we consider the condition in which only a part
of the encrypted data is used and assume that f x is
a binary image. Since only of the encrypted image
the encrypted image can be decrypted perfectly, the
average intensity of the decrypted image will be .
(7)
the decrypted image is 2 sinc2f ; the remain-
ing energy 2 sinc2f M becomes the
noise. Thus the average intensity of the noise is
2 sinc2f MN. As a result the SNR
value can be expressed as
N2 sinc2f
SNR . (10)
M 2 sinc2f
(8)
Equation 10 describes the ultimate condition in
which the pixel size of the mask is infinitesimal. As
discussed above, we need a decrypting mask with
higher spatial frequency to reduce the decrypted
noise when an external noise is added. This means
that the pixel size of the mask must be small.
Therefore Eq. 10 can be used to estimate the SNR
value when a high-spatial-frequency mask is used.
5. Computer Simulation and Experiment
In this section we numerically investigate the exten-
sion of the shift tolerance. The character E, as
shown in Fig. 2a, is a binary image of 256 256
(9) pixels. Figure 2b shows a double-random phase-
encoded character E. Figures 2c, 2d, and 2e cor-
responding to 12, 34, and 78, respectively, of the
encrypted image are blocked. When the decrypting
phase mask is placed at the matching position in the
decryption process, the decrypted images of Figs.
2b, 2c, 2d, and 2e are shown in Figs. 3a, 3b,
3c, and 3d, respectively. These figures present
the robustness to the loss of the encrypted data.
Figures 4a 4d show the decrypted results when
the decrypting phase mask is shifted for one pixel
from the matching position in the transverse direc-
tion. The character E cannot be decrypted when all
the encrypted data are used but can be decrypted
when a portion of the encrypted data is used. From
Fig. 5, when the phase key is shifted for two pixels
from the matching position, the character E can be
decrypted only when 14 and 18 the encrypted data
are used. Figure 6 shows that, when the phase key
is shifted for three pixels, the character E can be
decrypted only when 18 the encrypted data is used.
These results demonstrate that the robustness to the
loss of the encrypted data extends the shift tolerance
of the decrypting phase mask. We can use a low-
pass filter to enhance the decrypted image.7 Fur-
thermore, from the simulation, it is worth noting that
the capable pixel shift tolerance is the inverse of the
 Fig. 3. a, b, c, and d are decrypted images of Figs. 2b, 2c,
2d, and 2e, respectively, while the decrypting key is placed at
the matching position.

The input signal the point source from the lower

spatial filter in Fig. 7 was Fourier transformed onto
the ground glass. Then the random-phase-encoded
wave was Fourier transformed again onto the crystal,
and the random-phase-encoded signal was recorded

Fig. 2. a Original binary image of character E, b double-

random phase-encrypted image of character E, c half the en-
crypted image is blocked, d 34 the encrypted image is blocked, e
78 the encrypted image is blocked.

ratio number that the encrypted image used. The

relation can be explained clearly with Eq. 8.
We examine the extension of the shift tolerance
experimentally. Figure 7 shows a schematic dia-
gram of the experimental setup. Since a 4-f system
is a shift-invariant linear system, from Eq. 4 we
know that the encrypted image is the linear combi-
nation of every encrypted point of the input image.
Correspondingly, the decryption process is also a
shift-invariant linear process. From Eq. 7 we
know that the decrypted image is a linear combina-
tion of every point of the input image through the
decryption process. So investigating the property of
the systems impulse response is useful for clarifying
the shift tolerance of the system. Thus we used a
point source as the input signal, so only one random
phase code on the Fourier plane is required. The
focal length of the Fourier lens in the experiment is
150 mm. We used ground glass as the encrypting Fig. 4. a, b, c, and d are decrypted images of Figs. 2b, 2c,
mask, and a LiNbO3:Fe crystal with dimensions 1 2d, and 2e, respectively, while the decrypting key is shifted for
cm 1 cm 1 cm was used as the recording medium. one pixel transversely.

10 September 2000 Vol. 39, No. 26 APPLIED OPTICS 4791


Fig. 5. a, b, c, and d are decrypted images of Figs. 2b, 2c,
2d, and 2e, respectively, while the decrypting key is shifted for
two pixel transversely.
in the crystal with a hologram. This encoding pro-
cess is identical to that shown in Fig. 1a. At first,
shutter S1 was opened and shutter S2 was closed.
The plane wave incident on the crystal is the refer-
ence beam. When the index gratings were recorded
Fig. 6. a, b, c, and d are decrypted images of Figs. 2b, 2c,
2d, and 2e, respectively, while the decrypting key is shifted for
three pixel transversely.
4792 APPLIED OPTICS Vol. 39, No. 26 10 September 2000
Fig. 7. Experimental setup: M, mirror; BS, beam splitter; SF,
spatial filter; PH, pinhole; D, power detector; CL, collimating lens;
GG, ground glass; FL, Fourier lens; S, shutter.
in the crystal, we closed shutter S1 and opened shut-
ter S2 to read the gratings in the crystal. Then the
readout plane wave is the conjugate of the reference
beam. When the diffracted light went back through
the ground glass, the original signal was decrypted.
This process is identical to that shown in Fig. 1b.
The ground glass was placed on a motorized micro-
translation stage, with a resolution of 0.1 m. The
decrypted signal was detected with a power detector.
A pinhole was placed in front of the detector to filter
out the noise. Figure 8 shows the normalized inten-
sity of the decrypted signal with respect to the dis-
placement of the ground glass. The circles denote
the measured data when all the encrypted data re-
corded in the crystal are used. The squares denote
the measured data when half the crystal surface is
blocked. The curves denote the theoretical values
predicated by Eq. 8. The measured tolerance is
larger than the theoretical value for 2 m because
of the finite size of the ground glass, and we can treat
the 2-m finite size as the effective pixel size of the
ground glass. The effective pixel size is measured as
follows. As shown in Fig. 9, the LiNbO3 crystal is
placed just behind the ground glass to record the
Fig. 8. Experimental results. Circles, measured data when all
the encrypted data recorded in the crystal were used for decryp-
tion. Squares, measured data when half the crystal surface is
blocked. Solid and dashed curves, normalized intensity of point
source predicted by Eq. 8 when all and half the encrypted data,
respectively, are used.

Fig. 9. Experimental setup for measuring the effective pixel size
of the ground glass. The LiNbO3 crystal is placed closely behind
the ground glass. Experimental setup: M, mirror; BS, beam
splitter; SF, spatial filter; PH, pinhole; D, power detector; CL,
collimating lens; GG, ground glass; S, shutter.
index gratings. Figure 10 shows the normalized dif-
fraction intensity measured in the experiment with
respect to the displacement of the ground glass.
These results show that the effective pixel size is 2
m. They coincide with the experimental results
shown in Fig. 8. Therefore, when we consider the
effect of the effective pixel size, Eq. 8 can predict the
Fig. 10. Normalized diffraction intensity with respect to the dis-
placement of the ground glass when the LiNbO3 crystal is placed
closely behind the ground glass to record the index gratings.
View publication stats
extension of the shift tolerance well when less en-
crypted data are used.
6. Conclusion
The shift-tolerance property of the decrypting mask
in a double-random phase-encryption system has
been studied. The SNR value has also been inves-
tigated. A phase mask with higher spatial fre-
quency is required for reducing the noise on the
decrypted image when an external noise is added.
At the same time we can use the robustness to data
loss of the encrypted image to extend the shift toler-
ance during the decryption process. Theory and
computer simulations have been presented and fit
well with the experimental observation.
The authors express their appreciation to Bahram
Javidi for his helpful discussion. This study was
supported by National Science Council of the Repub-
lic of China contract NSC88-2215-E-008-013.
References
1. B. Javidi and J. L. Horner, Optical pattern recognition for
validation and security verification, Opt. Eng. 33, 17521756
1994.
2. J. Rodolfo, H. Rabenbach, and J.-P. Huignard, Performance of
a photorefractive joint transform correlator for fingerprint
identification, Opt. Eng. 34, 1166 1171 1995.
3. P. Refregier and B. Javidi, Optical image encryption based on
input plane and Fourier plane random encoding, Opt. Lett.
20, 767769 1995.
4. L. G. Neto and Y. Sheng, Optical implementation of image
encryption using random phase encoding, Opt. Eng. 35, 2459
2463 1996.
5. S. Lai, Security holograms using an encoded reference wave,
Opt. Eng. 35, 2470 2472 1996.
6. B. Javidi, G. Zhang, and J. Li, Experimental demonstration of
the random phase encoding technique for image encryption
and security verification, Opt. Eng. 35, 2506 2512 1996.
7. B. Javidi, A. Sergent, G. Zhang, and L. Guibert, Fault toler-
ance properties of a double random phase encoding encryption
technique, Opt. Eng. 36, 992998 1997.
8. B. Javidi, A. Sergent, and E. Ahouzi, Performance of double
random phase encoding encryption technique using binarized
encrypted images, Opt. Eng. 37, 565569 1998.
9. B. Javidi and E. Ahouzi, Optical security system with Fourier
plane encoding, Appl. Opt. 37, 6247 6255 1998.
10. O. Matoba and B. Javidi, Encrypted optical storage with an-
gular multiplexing, Appl. Opt. 38, 7288 7293 1999.
11. G. Unnikrishnan, J. Joseph, and K. Singh, Optical encryption
system that uses phase conjugation in a photorefractive crys-
tal, Appl. Opt. 37, 8181 8186 1998.
10 September 2000 Vol. 39, No. 26 APPLIED OPTICS 4793