Content uploaded by Frédéric Montfort
Author content
All content in this area was uploaded by Frédéric Montfort
Content may be subject to copyright.
Numerical parametric lens for shifting,
magnification, and complete aberration
compensation in digital holographic microscopy
Tristan Colomb
Ecole Polytechnique Fédérale de Lausanne, Institute of Imaging and Applied Optics, CH-1015 Lausanne,
Switzerland
Frédéric Montfort
Lyncée Tec SA, PSE-A, CH-1015 Lausanne, Switzerland
Jonas Kühn
Ecole Polytechnique Fédérale de Lausanne, Institute of Imaging and Applied Optics, CH-1015 Lausanne,
Switzerland
Nicolas Aspert and Etienne Cuche
Lyncée Tec SA, PSE-A, CH-1015 Lausanne, Switzerland
Anca Marian and Florian Charrière
Ecole Polytechnique Fédérale de Lausanne, Institute of Imaging and Applied Optics, CH-1015 Lausanne,
Switzerland
Sébastien Bourquin
Lyncée Tec SA, PSE-A, CH-1015 Lausanne, Switzerland
Pierre Marquet
Centre de Neurosciences Psychiatriques, Département de Psychiatrie DP-CHUV, Site de Cery,
1008 Prilly-Lausanne, Switzerland
Christian Depeursinge
Ecole Polytechnique Fédérale de Lausanne, Institute of Imaging and Applied Optics, CH-1015 Lausanne,
Switzerland
Received March 20, 2006; revised June 21, 2006; accepted June 26, 2006; posted July 5, 2006 (Doc. ID 69126)
The concept of numerical parametric lenses (NPL) is introduced to achieve wavefront reconstruction in digital
holography. It is shown that operations usually performed by optical components and described in ray geo-
metrical optics, such as image shifting, magnification, and especially complete aberration compensation (phase
aberrations and image distortion), can be mimicked by numerical computation of a NPL. Furthermore, we
demonstrate that automatic one-dimensional or two-dimensional fitting procedures allow adjustment of the
NPL parameters as expressed in terms of standard or Zernike polynomial coefficients. These coefficients can
provide a quantitative evaluation of the aberrations generated by the specimen. Demonstration is given of the
reconstruction of the topology of a microlens. © 2006 Optical Society of America
OCIS codes: 090.1760, 090.1000, 100.5070
.
1. INTRODUCTION
Digital holographic microscopy (DHM) permits the recon-
struction of the amplitude and the phase of an object
wavefront from the acquisition of a single digital holo-
gram. The principle consists of digitizing—with a CCD or
other type of image sensor such as a complementary
metal-oxide semiconductor (CMOS)—the interference be-
tween a reference and an object wave. Then, the wave-
front is propagated from the hologram to the image plane
within the Fresnel approximation by a numerical process.
Two different numerical formulations are principally
used: the single Fourier transform formulation (SFTF) or
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3177
1084-7529/06/123177-14/$15.00 © 2006 Optical Society of America
the convolution formulation (CF) (for SFTF and CF, see
Ref. 1). For both formulations, several parameters must
be adjusted or calibrated
2–4
to achieve a correct recon-
struction. Generally, the object wave contains aberrations
including the tilt due to the off-axis geometry, the phase
curvature introduced by the microscope objective (MO)
used to increase the spatial resolution, and all the optical
aberrations of the setup. It has previously been deter-
mined that the wavefront curvature introduced by the
MO and lenses can be successfully removed,
3,5
as well as
spherical aberration,
6
chromatic aberration,
7
astigmatism,
8,9
anamorphism,
10,11
and longitudinal im-
age shifting introduced by a beam splitter cube.
12
Fur-
thermore, a recent paper demonstrates that an automatic
procedure for performing the adjustment of parameters
associated with a standard polynomial model of aberra-
tions allows one to achieve a complete compensation for
phase aberrations in the image plane. This procedure can
be applied without prior knowledge of physical param-
eters of the setup including, e.g., wave vector components,
focal lengths, and positions of the optical components.
13
On the other hand, the two reconstruction formulations
SFTF and CF each have several advantages and disad-
vantages. In particular, within the framework of the
SFTF, the scaling of the reconstructed region of interest
(ROI) inside the reconstructed wave front is dependent on
the reconstruction distance, the pixel number of the holo-
gram, and the wavelength,
14
whereas the CF allows a
scaling-free reconstruction of the ROI if there is no chro-
matic aberration in the setup. Consequently, different so-
lutions have been proposed to control the scaling in SFTF
to maintain the size of the ROI for a sequence of digital
holograms recorded at different distances and to solve the
problem of superimposition in multiwavelength methods
for color holography,
15–18
tomographic holography,
19–25
or
optical diffraction tomography.
26
Ferraro et al. proposed
to control the scaling in SFTF by padding the holograms
with zeros before the reconstruction.
14
This approach has
the drawback of increasing the computational load be-
cause the number of hologram pixels is no longer a power
of 2. Indeed, the standard fast Fourier transforms are op-
timized to compute the Fourier transform in time
O共N log N兲 instead of O共N
2
兲 with N=2
n
. Zhang et al. pro-
posed another method to keep the original pixel number
in SFTF.
27
A two-stage reconstruction algorithm controls
the scale of the reconstructed image by placing between
the hologram and the image planes a numerical lens with
a focal length and a position defined by the chosen scale.
The disadvantage of this method is the requirement of the
computation of two propagations.
Finally, the curvature and the propagation direction of
the object wave are sensitive to the wavelength used if the
optics are not completely achromatic; thus a different
scale and position of the ROI can also occur with CF.
We define in this paper a numerical parametric lens
(NPL) placed in the hologram plane and/or in the image
plane that achieves a complete compensation for aberra-
tions (phase aberrations and image distortion) in SFTF or
CF. The NPL shape is defined by standard or Zernike
polynomial models whose parameters are adjusted auto-
matically by a two-dimensional (2D) fitting procedure ap-
plied on specimen areas known to be flat instead of con-
sidering one-dimensional (1D) profiles as presented in
Ref. 13. We demonstrate that the Zernike polynomial
model of the NPL achieves quantitative measurements of
specimen aberration properties. Then we demonstrate
that placing the NPL in the hologram plane has several
advantages. First, the correction of the tilt in the holo-
gram plane allows an automatic centering of the ROI in
the image plane that avoids any aliasing in CF or in
SFTF with small reconstruction distance. Second, the
complete aberration compensation in the hologram plane
is preserved for any reconstruction distance.
We illustrate the technique by compensating for astig-
matism produced by a cylindrical lens used as a MO and
for high-order aberrations produced by a ball lens and a
field lens introduced into the setup. Furthermore, chosen
shift and magnification operations are demonstrated in
CF by automatic computing of NPLs in hologram and im-
age plans with the advantages of maintaining constant
the original hologram pixel number and of using a unique
numerical propagation.
2. EXPERIMENTAL SETUPS
Figure 1 presents the optical setups of transmission [Fig.
1(a)] and reflection [Fig. 1(b)] digital holographic micro-
scopes. In both cases the basic architecture is that of a
modified Mach–Zehnder interferometer. The light source
depends on the targeted application: in Refs. 2 and 3 a
He–Ne laser is used; low coherence sources
22
or a tunable
source such as an optical parametric amplifier system
28
can be used.
In both configurations, a MO collects the object wave O
transmitted or reflected by the specimen and produces a
magnified image of the specimen at a distance d behind
the CCD camera. As explained in detail in Ref. 3, this
situation can be considered equivalent to a lensless holo-
graphic setup with an object wave O emerging directly
from the magnified image of the specimen and not from
the specimen itself.
In order to improve the sampling capacity of the CCD,
a lens can be optionally introduced into the reference arm
RL to produce in the CCD plane a spherical reference
wave with a curvature very similar to the curvature cre-
ated by the MO. At the exit of the interferometer, the in-
terference between the object wave O and the reference
wave R creates the hologram intensity
I
H
共x,y兲 = 共R + O兲共R + O兲
*
= 兩R兩
2
+ 兩O兩
2
+ R
*
O + RO
*
.
共1兲
This hologram is digitized by a black and white CCD cam-
era and then recorded on a computer. The digital holo-
gram I
H
共k ,l兲 is an array of N⫻ N (usually 512⫻ 512 or
1024⫻1024) 8-bit-encoded numbers resulting from the
2D sampling of I
H
共x ,y兲 by the CCD camera:
I
H
共k,l兲 =
冕
k⌬x−⌬x/2
k⌬y+⌬y/2
冕
l⌬y−⌬y/2
l⌬y+⌬y/2
I
H
共x,y兲dxdy, 共2兲
3178 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
where k, l are integers, and ⌬x, ⌬y define the sampling in-
tervals in the hologram plane (pixel size). The different
terms of this hologram are the zeroth order of diffraction
共兩R兩
2
+兩O兩
2
兲, the real image 共RO
*
兲, and the virtual image
共R
*
O兲. This hologram can be numerically filtered as
shown in Ref. 29 to produce a hologram containing only
the virtual image term
I
H
F
= R
*
O. 共3兲
We now introduce the NPL into the reconstruction pro-
cess. NPLs mimic pure phase objects and are defined as
two arrays of unit-amplitude complex numbers ⌫
H
and ⌫
I
placed respectively in the hologram 共H兲 and image 共I兲
planes. The numerically reconstructed wavefronts ⌿ are
also computed in the SFTF or CF as follows:
⌿
SFTF
共m,n兲 = ⌫
I
共m,n兲A exp
冋
i
d
共m
2
⌬
2
+ n
2
⌬
2
兲
册
⫻ FFT
再
⌫
H
共k,l兲I
H
F
共k,l兲
⫻exp
冋
i
d
共k
2
⌬x
2
+ l
2
⌬y
2
兲
册
冎
, 共4兲
⌿
CF
共m,n兲 = ⌫
I
共m,n兲A ⫻ FFT
−1
兵FFT关⌫
H
共k,l兲I
H
F
共k,l兲兴
⫻exp关− i
d共
k
2
+
l
2
兲兴其, 共5兲
where FFT is the fast Fourier transform; m, n, k, l are in-
tegers 共−N /2⬍m ,n, k ,l ⱕN /2兲; d is the reconstruction
distance; A= exp共 i2
d/ 兲 /共id兲; is the wavelength;
k
=k /共N⌬x兲,
l
=l /共N⌬y兲 are the spatial frequency coordi-
nates; and ⌬
and ⌬
are the sampling intervals in the
image plane defined as
⌬
= ⌬
=
d
N⌬x
. 共6兲
In SFTF, a scaling factor results between the hologram
size and the reconstructed ROI defined by the scale fac-
tors
␣
and given by
␣
=
⌬x
⌬
=
N⌬x
2
d
,
␣
=
⌬y
⌬
=
N⌬y
2
d
. 共7兲
We should remark that the particular case of ⌫
H
=R
and ⌫
I
=1 corresponds to the standard numerical expres-
sion of the Fresnel propagation.
2
Let us define a NPL as an array of unit amplitude com-
plex numbers that can be defined by standard or Zernike
polynomials:
⌫
S
共m,n兲 = exp
冉
− i
2
兺
␣
=

=0
␣
+

=o
P
␣
m
␣
m

冊
, 共8兲
⌫
Z
共m,n兲 = exp
冉
− i
2
兺
␣
=0
o
P
␣
Z
␣
冊
, 共9兲
where P
␣
and P
␣
are the NPL parameters and o is the
polynomial order. The Zernike polynomials, further desig-
nated by Z
␣
, are defined in Table 1 following the ZEMAX
classification.
30
We recall that the Zernike polynomials
are defined in a unit circle.
Now we represent the parametric numerical lenses in
the planes P=H,I as the multiplication of three different
lenses used to shift (Sh), magnify (M), and compensate (C)
for aberrations with the polynomial model PM=S,Z:
⌫
P
= ⌫
S
P,Sh
⌫
S
P,M
⌫
PM
P,C
. 共10兲
The numerical multiplication of these complex arrays is
achieved pixel-by-pixel and is therefore commutative.
However, the procedure to define them is not. The appli-
cation order of the NPL is first to compensate for the ab-
errations, then to do numerical magnification, and finally
to apply the numerical shift. But to simplify the explana-
tion, we present the different methods in the reverse or-
der.
Fig. 1. Digital holographic microscope, (a) transmission and (b)
reflection setups. O object wave; R reference wave; BS beam
splitter; M1, M2 mirrors; MO microscope objective, RL lens in
the reference wave, OC condenser in the object wave. For dem-
onstration purposes a tilted plate is introduced between the BS
and the CCD to intentionally produce aberrations. (c) Detail of
the off-axis geometry.
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3179
3. PRINCIPLE OF AUTOMATIC
PROCEDURES
In Ref. 13, a simple procedure has already been presented
that is performed in the image plane and permits adjust-
ment of the standard polynomial parameters P
␣
bya1D
least-squares fitting procedure applied to profiles ex-
tracted from areas of the specimen known to be flat.
Drawbacks of the 1D procedure are that the method is
limited to a standard polynomial model and requires the
use of a reference hologram (without specimen) to com-
pute the cross terms of the standard polynomial (as x
␣
y

,
␣
,

⫽ 0). We present here a more general and efficient 2D
fitting procedure applied on specimen areas known to be
flat. The fitting procedure is illustrated in the image
plane of a United States Air Force (USAF) test target ho-
logram recorded with a reflection setup [see Fig. 1(b)]
where a tilted thick plate is introduced between the beam
splitter BS and the CCD camera in order to produce ab-
errations.
In the assumed flat specimen area F defined by the mo-
saic of white rectangle on the surface [see Fig. 2(a)] N
pts
points 共
␥
m
,
n
兲 are selected 关共
␥
m
,
n
兲苸 F 兴. The N
pts
mea-
sured phase values are converted to optical path lengths
(OPL) Y共
␥
m
,
n
兲 that satisfy the N
pts
following equations
depending on the model used [Eqs. (8) and (9)]:
Y共
␥
m
,
n
兲 =
兺
␣
=

=0
␣
+

=o
a
␣
S
␣
, S
␣
=
␥
m
␣
n

, 共11兲
Y共
␥
m
,
n
兲 =
兺
␣
=0
␣
=o
a
␣
Z
␣
, 共12兲
where
␥
m
and
n
are computed from the pixel position
共m ,n兲 to satisfy the condition that F is inscribed in the
unit circle and o is the polynomial order. Equations (12)
and (11) define two linear systems with N
pts
equations
and, respectively, a number of unknown coefficients o +1
and 共 o
2
+3o+2兲/2 (for example, for a second order of the
standard polynomial the six unknown coefficients are a
00
,
a
10
, a
01
, a
20
, a
02
, and a
11
). Because a great number of
points can be selected in an image the system is always
Fig. 2. Two-dimensional fitting procedure with standard poly-
nomial model (left column) and Zernike polynomial model (right),
(a) the reconstructed amplitude contrast with the assumed flat
areas F situated inside the white rectangles. (b) and (f) recon-
struction with initial parameters computed with 1D procedure.
(c) and (g) 2D unwrap of (b) and (f), respectively (d) and (e) re-
spectively, the corrections with six and ten adjusted standard
polynomial coefficients 共o =2,3兲. (h) and (i) respectively, the cor-
rection with six and eight adjusted Zernike polynomial coeffi-
cients 共o =5,7兲.
Table 1. Zernike Standard Coefficients in ZEMAX
Classification
Polynomial Cartesian Form Description
Z
0
1 Piston
Z
1
冑
4x
Tilt x
Z
2
冑
4y
Tilt y
Z
3
冑
3共2x
2
+2y
2
−1兲
Power
Z
4
冑
6共2xy兲
Astig y
Z
5
冑
6共x
2
−y
2
兲
Astig x
Z
6
冑
8共3x
2
y+3y
3
−2y兲
Coma y
Z
7
冑
8共3x
3
+3xy
2
−2x兲
Coma x
Z
8
冑
8共3x
2
y−y
3
兲
Trefoil y
Z
9
冑
8共x
3
−3xy
2
兲
Trefoil x
Z
10
冑
5共6x
4
+12x
2
y
2
+6y
4
−6x
2
−6y
2
+1兲
Primary
Spherical
Z
11
冑
10共4x
4
−3x
2
+3y
2
−4y
4
兲
2
ary
Astig x
Z
12
冑
10共8x
3
y+8xy
3
−6xy兲
2
ary
Astig y
Z
13
冑
10共x
4
−6x
2
y
2
+y
4
兲
Tetrafoil x
Z
14
冑
10共4x
3
y−4xy
3
兲
Tetrafoil y
Z
15
冑
12共10x
5
+20x
3
y
2
+10xy
4
−12x
3
−12xy
2
+3x兲
2
ary
Coma x
Z
16
冑
12共10x
4
y+20x
2
y
3
+10y
5
−12x
2
y−12y
3
+3y兲
2
ary
Coma y
Z
17
冑
12共5x
5
−10x
3
y
2
−15xy
4
−4x
3
+12xy
2
兲
2
ary
Trefoil x
Z
18
冑
12共15x
4
y+10x
2
y
3
−5y
5
−12x
2
y+4y
3
兲
2
ary
Trefoil y
Z
19
冑
12共x
5
−10x
3
y
2
+5xy
4
兲
Pentafoil x
Z
20
冑
12共5x
4
y−10x
2
y
3
+y
5
兲
Pentafoil y
Z
21
冑
7共20x
6
+60x
4
y
2
+60x
2
y
4
+20y
6
−30x
4
−60x
2
y
2
−30y
4
+12x
2
+12y
2
−1兲
2
ary
Spherical
Z
22
冑
14共30x
5
y+60x
3
y
3
+30xy
5
−40x
3
y−40xy
3
+12xy兲
3
ary
Astig y
Z
23
冑
14共15x
6
+15x
4
y
2
−20x
4
+6x
2
−15x
2
y
4
−15y
6
+20y
4
−6y
2
兲
3
ary
Astig x
3180 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
overdetermined. For example assuming 3% flat area in a
256⫻256 image, 1966 equations can be defined, for ex-
ample, to compute 28 unknown factors for o= 6 in the
standard polynomial model. These systems are solved by
computing in the least-squares sense the solution of
M ⫻ A
M
= Y , 共13兲
where M =S, Z is the matrix of fitting polynomials in the
standard or Zernike model, Y is the vector of the OPL
measured values Y共
␥
m
,
n
兲, and A
M
is the vector of the un-
known coefficients a
␣
or a
␣
.
As already developed in Ref. 13, an iterative procedure
can be used to adjust the parameter vector P
M
:
P
M
共i兲
= P
M
共i−1兲
+ A
M
共i兲
. 共14兲
The NPLs for aberration compensation are therefore
given by
⌫
M
P,C
= exp
冋
− i
2
P
M
· M
册
. 共15兲
Obviously, this iterative procedure fails if there are phase
jumps in the areas F due to initial NPLs parameters be-
ing too different from the optimal ones. Therefore a
simple first-step procedure consists of computing initial
parameters P
M
共0兲
with a 1D fitting method
13
:
P
S
共0兲
=
关
0 P
10
1D
P
01
1D
P
20
1D
P
02
1D
兴
, 共16兲
P
Z
共0兲
=
冋
0
P
10
1D
2
P
01
1D
2
册
, 共17兲
where P
␣
1D
are the parameters adjusted by the 1D fitting
procedure [see Figs. 2(b) and 2(f)].
The second step consists of performing a 2D unwrap on
the resulting reconstructed phase in order to suppress
possible remaining phase jumps due to aberrations [see
Figs. 2(c) and 2(g)].
Finally, the 2D fitting procedure is applied by increas-
ing the polynomial order when necessary. Figures 2(d),
2(e), 2(h), and 2(i) present the reconstructed phase ob-
tained with the NPL adjustment in, respectively, the
standard and Zernike polynomial models. The polynomial
order used for Figs. 2(d) and 2(e) are, respectively, o=2
(six parameters) and o =3 (ten parameters), and those
used for Figs. 2(h) and 2(i) are, respectively, o =5 (six pa-
rameters) and o= 7 (eight parameters).
4. RESULTS
A. Application to Quantitative Aberration Measurement
As already established in Ref. 13, the automatic adjust-
ment of NPL in the image plane can be used to compen-
sate for the specimen curvature. Here, we demonstrate
that the Zernike polynomial model not only allows us to
compensate for the curvature of the specimen but also
measures quantitatively the aberrations in term of
Zernike coefficients. Figure 3 presents different represen-
tations of the same microlens recorded in a transmission
setup. The first step consists of compensating for the
setup aberration by applying the 2D fitting procedure in
the Zernike model on areas around the microlens where
the surface is known to be flat [areas shown by white lines
in Fig. 3(a)]. The resulting image allows a perspective
representation of the curvature induced by the microlens
in Fig. 3(b) by applying a 2D phase unwrap on the image
in Fig. 3(a). Now, the 2D fitting procedure applied in the
area of the microlens [dashed white circle in Fig. 3(c)] al-
lows us to compensate for the specimen curvature. Figure
3(c) corresponds to the adjustment of Zernike coefficients
up to Z
9
. The result of the adjustment of the next Zernike
coefficient Z
10
is shown in Fig. 3(d). The “flattening” op-
eration performs better and reveals that this microlens
generates important spherical aberrations (see Table 1).
Finally, Fig. 3(e) shows that the increase of the polyno-
mial order up to Z
20
does not provide a better aberration
compensation.
Figure 4 summarizes the repartition of the aberrations
of the microlens. It shows an important astigmatism (Z
4
and Z
5
; more in direction y than x), a coma amplitude (Z
6
and Z
7
) equivalent in the two directions, a trefoil (Z
8
and
Z
9
) negligible in direction y, and finally an important pri-
Fig. 4. Repartition of Zernike coefficients for an adjustment of
21 coefficients. The absolute coefficient values are plotted. Black
and gray patterns indicate, respectively, negative and positive
values.
Fig. 3. (a) Microlens phase by applying 2D fitting procedure
with Zernike polynomial on points included in areas indicated by
white lines. (b) Perspective representation of 2D phase unwrap of
(a). (c)–(e) Microlens shape compensation with Zernike formula-
tion with (c) 10 parameters, (d) 11 parameters, (e) 21 parameters.
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3181
mary spherical aberration 共Z
10
兲. In addition the represen-
tation of the “flattened” microlens and the coefficient pa-
rameters provides a wealth of data on the microlens such
as surface topography, radius of curvature, lens height,
and surface roughness.
31
B. Automatic Region of Interest Centering
As a result of the off-axis geometry of the holographic set-
ups, the carrier frequencies of the real or virtual images
are not in the center of the spectrum as presented in Fig.
5(a). This results in a spatial separation of the different
diffraction orders during the reconstruction process.
32
In
our setups (Fig. 1), the object and reference waves propa-
gate, respectively, collinearly and with an angle
from
the normal vector to the hologram plane during the re-
cording process [Fig. 6(a)]. Let us consider now the wave-
front reconstruction from a filtered hologram containing
only a virtual image in two different ways.
The first way [Fig. 6(b)] corresponds to the reconstruc-
tion process with the digital reference wave outside the
Fresnel integral, as described in Ref. 13. In this case, the
reconstructed wavefront in the hologram plane is R
*
O
and propagates at an angle −
. The ROI is therefore
shifted in the image plane [see Fig. 7(a) with SFTF and
Fig. 7(e) with CF].
The second way [Fig. 6(c)] corresponds to the optical re-
construction process with the reference wave R as illumi-
nating wave. Digitally this amounts to the same thing as
computing Eq. (4) or (5) with ⌫
H,C
=R. In this case, the
wavefront in the hologram plane is O, which propagates
normally to the hologram plane. Therefore, the ROI is
centered in the reconstructed wavefront.
The shift of the ROI in the first reconstruction method
is not convenient in CF because aliasing appears as pre-
sented in Figs. 7(e) and 7(f). In the case of SFTF, it may
not be a problem if the scale factors defined in Eq. (7) are
sufficiently small to avoid any aliasing [Figs. 7(a) and
7(b)]. But because these scale factors are inversely pro-
portional to the reconstruction distance d, aliasing could
nevertheless appear when d becomes too small (Fig. 8).
Therefore for any formulation, it is more judicious to sup-
press the shift or the ROI as presented in Figs. 7(g) and
7(h) for CF and in Fig. 8(d) for SFTF with small recon-
struction distance.
The procedure to shift the ROI to the center can be
achieved with two methods. The first one, called spectrum
centering, consists of shifting the carrier frequency of the
virtual (real) image to the center of the filtered hologram
spectrum, applying an inverse Fourier transform of the
resulting spectrum, and then propagating the wavefront.
A simpler procedure consists of detecting the position of
the amplitude maximum corresponding to the carrier fre-
quency of the virtual image [see Fig. 5(a)]. This position is
then shifted to the center of the spectrum [see Fig. 5(b)].
This method has two main drawbacks. The first is that
the shifting amplitude is limited by the pixel accuracy.
The second concerns the central frequency spreading [see
Fig. 5(c)] that results from a difference of curvature be-
tween the reference and object waves and makes the vir-
tual image carrier frequency impossible to center by a
simple maximum-amplitude detection.
We propose to compute automatically the tilt param-
eters P
10
H,C
and P
01
H,C
of ⌫
S
H,C
by selecting profiles or areas
known to be flat in the hologram plane and then proceed-
ing with the fitting procedure. Because the image in the
hologram plane is defocused flat areas would seem to be
difficult to define. In fact, the contributions of the phase
diffraction pattern are averaged and are therefore negli-
gible if the selected profiles or areas are sufficiently far
from the specimen. Figure 9 presents the hologram plane
phase image before [Fig. 9(a)] and after [Fig. 9(c)] the tilt
adjustment and their corresponding image plane ampli-
tude [Figs. 9(b) and 9(d)] reconstructed in SFTF. In this
example, the 1D procedure is applied to the selected black
profiles of Fig. 9(a). The resulting phase curvature in
Fig. 9(c) corresponds to the noncorrected curvature in-
duced by the MO. The ROI is centered in the image plane
[Fig. 9(d)] and we have ⌫
H,C
=R.
C. Manual Shifting in Convolution Formulation
It may be interesting to shift the ROI manually in a spe-
cific region, for example, in order to compensate for a
specimen translation between two hologram acquisitions.
For this purpose, we show how to define the shifting
NPLs ⌫
H,Sh
and ⌫
I,Sh
in the CF. The procedure has three
principal steps. First, the operator draws two points de-
fining the desired shift (arrows in Fig. 10). The second
Fig. 5. Procedure of spectrum centering. (a) Initial filtered spec-
trum, (b) spectrum centered. The arrow represents the shift be-
tween the amplitude maximum of the frequencies associated
with the virtual image and the center of the entire spectrum. (c)
Spectrum of a hologram for which the curvatures of the reference
and object waves are different, inducing a nonpunctual central
frequency in the spectrum.
Fig. 6. Principle of digital reconstruction process to center the
ROI. (a) Hologram recording, (b) reconstruction with a digital
reference wave U = 1 (the ROI is not centered), (c) reconstruction
with a digital reference wave U =R (the ROI is centered).
3182 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
step consists of computing the parameters P
10
H,Sh
and P
01
H,Sh
of ⌫
H,Sh
. These parameters can be easily computed by con-
sidering Fig. 11. Let us define the chosen shift in the two
directions by
⌬S
j
= N
Sj
⌬j, 共18兲
where j =x, y and N
Sj
is the number of pixels to shift in the
j direction. The shifting NPL is written as
⌫
S
H,Sh
共x兲 = exp
冋
i
2
S
ˆ
x
册
= exp
冋
i
2
共S
x
m⌬x + S
y
n⌬y兲
册
,
共19兲
where S
ˆ
is the unit shift vector. The components of the
vector S
ˆ
are
S
j
= sin共
j
兲 = sin
冋
arctan
冉
⌬S
j
d
冊
册
. 共20兲
The parameters P
10
H,Sh
and P
01
H,Sh
are also
P
10
H,Sh
= − sin
冋
arctan
冉
⌬S
x
d
冊
册
⌬x,
P
01
H,Sh
= − sin
冋
arctan
冉
⌬S
y
d
冊
册
⌬y. 共21兲
Obviously this shift introduced into the hologram plane
produces a tilt in the image plane that should be compen-
sated for. We introduce therefore a predicted compensat-
ing shifting NPL in the image plane defined as
⌫
S
I,Sh
共m,n兲 = exp
冋
− i
2
共P
10
I,Sh
m + P
01
I,Sh
n兲
册
, 共22兲
where P
10
I,Sh
=−P
10
H,Sh
and P
01
I,Sh
=−P
01
H,Sh
. Figures 10(c) and
10(d) show the shifted amplitude and phase reconstruc-
tions.
It is important to note that this shifting method is lim-
ited by the chosen shift and by the reconstruction dis-
tance. Indeed, the Nyquist sampling criterion requires
that the highest spatial frequency introduced by the shift
should be less than the cutoff frequency 1/共2⌬x兲. In other
words, it means that the shifting angle
may not exceed
the maximum value
max
given by
ⱕ
max
= arcsin
冉
2⌬x
冊
. 共23兲
The shifting is also limited by Eq. (23), which gives the
inequality
Fig. 8. Aliasing appears when the reconstruction distance is too
small. (a) d =11 cm, (b) d=5 cm, (c) aliasing at d=3.3 cm. With
ROI centering, the reconstruction (d) can be achieved without
aliasing.
Fig. 7. Comparison between SFTF [(a)–(d)] and CF [(e)–(h)], with [(c), (d), (g), (h)] or without [(a), (b), (e), (f)] ROI centering. (a), (c), (e),
(g) are amplitude images and (b), (d), (f), (h) the corresponding phase reconstructions.
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3183
arctan
冉
⌬Sj
d
冊
ⱕ
max
= arcsin
冉
2⌬x
冊
. 共24兲
For example, with a sampling ⌬x= 6.7
m, a wavelength
= 633 nm, and a reconstruction distance d =1 cm, the
maximum number of pixels to shift is N
Smax
=d /⌬x tan关arcsin共 /2⌬x兲兴=70.58 pixels. This limitation is
not a problem, because the chosen shift is usually limited
to a maximum of a few dozen pixels and the reconstruc-
tion distance is also usually ⬇5 cm, which corresponds to
N
Smax
=352 pixels.
The automatic and manual shift methods are very effi-
cient in comparing or superposing different wavefront re-
constructions that appear usually in different areas of the
image plane. In particular, these methods are very useful
in polarization imaging with DHM
33
in which two fringe
patterns are recorded on the same hologram. For this ap-
plication, two different orthogonal, polarization-state ref-
erence waves with two different propagation directions
are used in order to separate spatially the two recon-
structed virtual images. The automatic and manual shifts
constitute a very simple way to perform a subpixel super-
position of the two wavefronts so as to compute the polar-
ization parameters.
D. Numerical Magnification in Convolution Formulation
We propose here to adjust the magnification of the ROI by
computing the parametric focal length of the NPL charac-
terized by ⌫
H,M
and ⌫
I,M
. This method keeps constant the
pixel number of the hologram and is based on a single
propagation. Let us define the hologram plane H where
the NPL with focal distance f is placed, the original image
plane I defined by the reconstruction distance d (position
of the reconstructed virtual image), and the final image
plane I
⬘
defined by the reconstruction distance d
⬘
.By
these definitions, the real object (having the same size as
the virtual image) is at a distance d from the hologram
plane. The magnification M is also calculated from the
real object and image distances:
M =−d
⬘
/共− d兲 = d
⬘
/d. 共25兲
The lens equation gives
1/f =1/共− d兲 +1/d
⬘
. 共26兲
Let us define now the magnification NPL described by a
thin lens transmittance
32
or from Eq. (8):
⌫
H,M
共m,n兲 = exp
冋
i
2
1
2f
共m
2
⌬x
2
+ n
2
⌬y
2
兲
册
, 共27兲
⌫
H,M
共m,n兲 = exp
冋
− i
2
共P
20
H,M
m
2
+ P
02
H,M
n
2
兲
册
, 共28兲
where P
20
H,M
=P
02
H,M
are the magnification parameters asso-
ciated with the focal length of the lens:
P
20
H,M
= P
02
H,M
=
⌬x
2
2f
. 共29兲
Finally, with Eqs. (25), (26), and (29), the new reconstruc-
tion distance and the parameter P
02
H,M
can be computed
Fig. 9. Adjustment of the tilt parameters of the NPL ⌫
H
by ap-
plying 1D procedure along black profiles. (a) The initial phase in
the hologram plane, (b) the corresponding amplitude reconstruc-
tion in SFTF, (c) tilt-corrected phase in the hologram plane, (d)
the corresponding centered amplitude reconstruction.
Fig. 10. Shifting procedure: (a) and (b) show, respectively, the
amplitude and phase reconstructions after tilt compensation.
The arrows define the chosen translation of the ROI. (c) and (d)
show the respective amplitude and phase shifted
reconstructions.
Fig. 11. H, hologram plane; I, image plane; ⌬S
x
, chosen shift in
the direction x; d, reconstruction distance;
x
, shifting angle.
3184 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
from M and the initial reconstruction distance d:
d
⬘
= Md, 共30兲
P
02
H,M
= P
20
H,M
=
冉
1
M
−1
冊
⌬x
2
2d
. 共31兲
Obviously, as for the shifting method, the phase curva-
ture introduced in the hologram plane by the NPL has to
be compensated for in the new image plane I
⬘
. The pre-
dicted compensation for the magnification NPL in the im-
age plane is
⌫
I,M
共m,n兲 = exp
冋
i
2
1
2共f − d
⬘
兲
共m
2
⌬x
2
+ n
2
⌬y
2
兲
册
, 共32兲
⌫
I,M
共m,n兲 = exp
冋
− i
2
共P
20
I,M
m
2
+ P
02
I,M
n
2
兲
册
, 共33兲
where the parameters are
P
20
I,M
= P
02
I,M
=
⌬x
2
共M −1兲
2M
2
d
. 共34兲
An example of application of this method is presented
in Fig. 12. Two different holograms of the same object
have been recorded with two different wavelengths
1
=480 nm [Figs. 12(a) and 12(b)] and
2
=700 nm [Figs.
12(c)–12(f)]. We note that the size of the observed object is
different because of the nonachromatic MO used in the
setup (difference between the dashed and solid white rect-
angles). The ratio of the rectangle sizes defines a magni-
fication M =1.0038. The magnification procedure allows
us to achieve the scaled reconstruction presented in Figs.
12(e) and 12(f).
We can mention here that a different scaling in the two
directions can be done by applying two different magnifi-
cations in the corresponding directions.
The shifting and magnification procedure can be ap-
plied in the context of submicrometer optical tomography
by multiple wavelength DHM. The principle consists of
recording several holograms at different wavelengths
(typically 20 holograms with wavelengths between
480 nm and 700 nm) with a reflection digital holographic
microscope. The reconstruction of these holograms and
their processing allows tomographic imaging.
34
An impor-
tant point for the tomographic reconstruction process is
that the size of the ROI on each reconstructed image
should be identical. Because of the presence of chromatic
aberration and/or laser pointing changes for each wave-
length, the reconstruction distance, the size, and the po-
sition of the ROI change as shown in Figs. 12(a)–12(d).
Figure 13 compares the mean amplitude computed from
the 20 holograms [Figs. 13(a) and 13(b)] and the mapping
of it on the 3D topography of the specimen [Figs. 13(c) and
13(d)] when the magnification and the shift are either ap-
plied to all 20 superimposed, reconstructed images [Figs.
13(b) and 13(d)] or not applied [Figs. 13(a) and 13(c)]. We
can see clearly that the image in Fig. 13(a) is blurred
whereas that in Fig. 13(b) is not. The improvement of the
method is also visible in Fig. 13(d) where the noise on the
specimen edges is clearly diminished.
E. Complete Aberration Compensation
Let us assume that the specimen does not introduce ab-
errations but only a phase delay
共x ,y兲. In the known flat
Fig. 12. Amplitude and phase reconstructions are presented, re-
spectively, on the left and on the right. The reconstructions are
done from a hologram recorded with (a), (b)
1
=480 nm; (c)–(f)
2
=700 nm. The white rectangle defines the reference size. The
white dashed rectangle defines the size of the same object with-
out performing magnification. Images in (e), (f) are reconstructed
from the same hologram as in (b), (c) after performing a magni-
fication procedure M=1.0038 defined by the ratio of the rectangle
sizes.
Fig. 13. (a), (b) Mean amplitude reconstructed from 20 holo-
grams recorded with different wavelengths and (c), (d) mapping
of it on the 3D topography of the specimen. Images in (b), (d) are
processed with shift and magnification compensation.
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3185
areas this term
共x ,y兲=c (c a constant) for any plane, ne-
glecting the diffraction pattern due to the specimen in de-
focused planes. The reference and object waves can be de-
fined more generally by introducing the respective phase
aberration terms W
R
and W
O
:
R共x,y兲 = 兩R兩exp关i共k
x
x + k
y
y兲兴exp关iW
R
共x,y兲兴, 共35兲
O共x,y兲 = 兩O兩exp关i
共x,y兲兴exp关iW
O
共x,y兲兴. 共36兲
These phase aberration terms can be astigmatism, defo-
cus aberration, spherical aberration, and so on. Here, we
assume that the amplitude is not affected by the aberra-
tions. The filtered hologram of Eq. (3) becomes
I
H
F
= 兩R储O兩exp关− i共k
x
x + k
y
y兲兴exp关i共
+ W
O
− W
R
兲兴.
共37兲
The method of suppressing the aberration term W =W
O
−W
R
consists simply of applying the 1D or the 2D fitting
procedure on the filtered hologram phase. The adjustment
of the standard or Zernike polynomial parameters of ⌫
H
is
achieved by considering the known flat areas in the holo-
gram plane.
We have already shown that the tilt adjustment in this
plane permits us to place the ROI in the center of the im-
age plane [see Figs. 14(a)–14(c)]. By increasing the order,
it is possible to “flatten” the phase in the hologram plane
as presented in Fig. 14(d). Because of the compensation
for the curvature of the object wave, the NPL works also
as a magnification lens: The reconstruction distances with
or without NPL are different, consistent with the equa-
tions presented in Subsection 4.D.
In Fig. 14 the initial reconstruction distance is d
=17.46 cm [Figs. 14(b) and 14(c)], the adjusted term P
02
H
=1.24558⫻ 10
−10
provides a magnification M =0.5122, and
a new reconstruction distance d =8.78 cm is used to recon-
struct Figs. 14(e) and 14(f). Because the reconstruction
was achieved in SFTF, no magnification of the ROI ap-
pears. It is important to note that no NPL is applied in
the image plane for the reconstruction of the images in
Figs. 14(c) and 14(f). We see that the correction in the ho-
logram plane avoids the utilization of the NPL in the im-
age plane for any reconstruction distance. Indeed, the
term ⌫
H
I
H
F
is similar to a plane wave modulated by the
phase delay induced by the specimen. The propagation of
this plane wave therefore conserves a constant phase
value in the areas known to be flat, and the phase aber-
rations are also corrected for any reconstruction distance.
5. APPLICATIONS AND DISCUSSION
A. Compensation for Astigmatism Induced by a
Cylindrical Lens
Grilli et al. present theoretically the potentialities of
DHM for astigmatism evaluation and compensation.
9
Furthermore, De Nicola et al. present a method with two
different reconstruction distances to achieve astigmatism
compensation.
10
Here we demonstrate experimentally
that astigmatism introduced by a cylindrical lens can be
compensated for by a NPL in the hologram plane only.
This cylindrical lens is introduced in a reflection setup in
the place of the MO [see Fig. 1(b)]. Figure 15 presents the
hologram of a USAF test target recorded with this setup.
The fringe pattern is unusual and corresponds to the in-
terference between an ellipsoidal wave and a plane wave.
Figure 16 presents the amplitude reconstruction with
SFTF along the z direction for different reconstruction
distances. Because of astigmatism of the cylindrical lens,
there are two partial focal points, one for each direction,
localized at d= ⬁ and d= 5.5 cm (the amplitude reconstruc-
tion shows a vertical line). The image is almost focused at
d= 23.3 cm as also shown on Fig. 17(b). This astigmatism
can be revealed better by using CF. Because aliasing ap-
pears in CF [Fig. 17(a)] because of a larger magnification
of the cylindrical lens in the horizontal direction, a nu-
merical magnification M=0.3 is applied; the results can
Fig. 15. Hologram of USAF test target recorded with a cylindri-
cal lens as MO.
Fig. 14. (a) Correction of the tilt in the hologram plane and re-
spective (b) amplitude and (c) phase reconstructions in SFTF at a
distance d =17.46 cm without NPL in image plane. (d) High-order
correction in the hologram plane and respective (e) amplitude
and (f) phase reconstruction at a distance d=8.78 cm. The correc-
tion in the hologram plane is preserved along the direction of
propagation and a numerical lens is no longer necessary in the
image plane.
3186 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
be observed in Fig. 17(c). The inset in Fig. 17(c) shows
clearly that the vertical edges of the USAF step are not
focused.
Let us define a new NPL written ⌫
H,A
that is dedicated
to astigmatism compensation and defined by two second-
order standard polynomial coefficients P
02
H,A
and P
20
H,A
. Fig-
ures 17(d) and 17(e) present, respectively, the compensa-
tion for this astigmatism by the manual adjustment of
P
20
H,A
=0.11⫻ 10
−10
共P
02
H,A
=0兲 and by the adjustment of two
reconstruction distances d
1
=6.99 cm and d
2
=7.92 cm as
explained in Ref. 10. We can see in the insets that the two
methods correct the astigmatism very well, but there is a
very small difference between the ROI sizes: The image in
Fig. 17(e) is larger in the horizontal direction.
Figure 18 reveals that the two astigmatism compensa-
tions used for the amplitude image are not sufficient to
compensate for the other phase aberrations if no NPL is
applied in the image plane [see Fig. 18(a)]. Therefore, the
NPL ⌫
I,C
is adjusted in the image plane for the different
cases of astigmatism correction: Fig. 18(b) without correc-
tion, Fig. 18(c) with P
20
H,C
=0.11⫻ 10
−10
, and Fig. 18(d) with
two reconstruction distances. One should note that the
NPL method preserves the geometry [the step length is
equal between Figs. 18(b) and 18(c)], whereas this is not
the case for the two-reconstruction-distance technique of
Fig. 18(d).
Finally, we compare the two astigmatism compensation
methods when the 2D fitting procedure is applied in the
hologram plane to adjust ⌫
S
H,C
. It is also important to re-
mark that, as established for the magnification method,
the introduction of ⌫
H,A
when ⌫
H,C
has already been ad-
justed introduces a phase curvature in the image plane
that can be compensated with the predicted NPL ⌫
I,A
in
the image plane with P
20
I,A
or P
02
I,A
computed from Eqs. (31)
and (33):
P
20
I,A
=
⌬x
2
共⌬x
2
/2P
20
H
兲 − d
. 共38兲
Figure 19 presents the phase image in the hologram
plane before [Fig. 19(a)] and after [Fig. 19(b)] the 2D fit-
ting procedure for ⌫
S
H,C
. The “flattening” operation in the
hologram plane increases the astigmatism as presented
in Figs. 20(a) and 20(b). Indeed, two very different recon-
struction distances allow focus along the horizontal [Fig.
20(a), d= 13.3 cm] or vertical direction [Fig. 20(b),
d= − 6.9 cm]. The astigmatism is therefore compensated
by the two-reconstruction-distance method [Fig. 20(c)] or
by the adjustment of ⌫
H,A
for the cases of two reconstruc-
tion distances: Fig. 20(d) corresponds to d= −7.1 cm,
P
02
H,A
=−4.7⫻10
−10
and Fig. 20(e) to d =7.95 cm, P
20
H,A
=4.7
⫻10
−10
.
In short, we show that the two astigmatism compensa-
tion methods are not equivalent in terms of geometry con-
servation. Indeed, the two-reconstruction-distance
method deforms the reconstructed images, whereas this is
not the case with our method. Furthermore, our method
involving aberration compensation in the hologram plane
achieves astigmatism compensation for the amplitude
and phase images for any reconstruction distances. These
Fig. 17. Amplitude reconstruction with different parameters,
(a) CF, M =1, d=23.3 cm; (b) SFTF, d =23.3 cm, (c) CF, M=0.3, d
=6.99. The astigmatism shown in detail in (c) is compensated by
(d) the adjustment of P
20
H,A
=0.11⫻10
−10
or by (e) defining two re-
construction distances d
1
=6.99 cm and d
2
=7.92 cm.
Fig. 18. (a) Phase reconstruction with P
20
H,A
=0.11⫻10
−10
with-
out ⌫
I,C
; the other images are compensated with ⌫
I,C
,(b)P
20
H,A
=0, (c) P
20
H,A
=0.11⫻10
−10
, and (d) two reconstruction distances.
The black lines have the same length and reveal a dilatation in
the image in the horizontal direction for (d).
Fig. 16. Amplitude reconstruction for different distances of re-
construction. Because of the astigmatism of the cylindrical lens,
there are two different focal points located at d =5.5 cm and d
=⬁. The reconstructed image is focused at d =23.3 cm.
Fig. 19. Phase image in hologram plane: (a) without ⌫
S
H,C
ad-
justment, (b) after adjustment of standard polynomial order o
=3. The straight black lines define the profiles used to set the ini-
tial values of 2D fitting parameters, and the curved white lines
delimit the areas excluded from the areas known to be flat.
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3187
results show that a cylindrical lens can be used advanta-
geously instead of a MO to study specimens with different
characteristic length and width such as optical fibers or
waveguides.
B. Ball Lens as Microscope Objective
To illustrate further the different techniques of aberration
compensation, we introduce a ball lens (Edmund ball lens
SF8 of 2 mm diameter, n =1.689) as MO and a field lens
between the BS and the CCD camera in a transmission
setup [Fig. 1(a)]. The positions of the ball lens, the field
lens, and the CCD are adjusted to produce very strong ab-
errations. A liquid of index n= 1.6 is used as immersion
fluid. The specimen is a USAF test target. Figure 21 pre-
sents the comparison between different methods of aber-
ration compensation. In the first column, only the tilt is
compensated for in the hologram plane. It is evident that
aberrations are introduced by the ball lens that deforms
the USAF test target [Figs. 21(b) and 21(c)]. Furthermore,
the NPL applied in the image plane does not succeed in
correctly “flattening” the phase image. In the second col-
umn, a seventh-order standard polynomial 2D fitting is
applied to the hologram. The correction of distortion is
good, and the phase aberrations are well compensated.
The residual distortion comes from the nonexact as-
sumption of a nonaberrated amplitude of the reference
and object waves. Indeed, some part of the phase aberra-
tions introduced by the ball lens and/or the field lens is
converted to amplitude aberrations by the optical propa-
gation of the wave in the path of the ball lens, the field
lens, and the CCD camera. Because the NPL compensates
only for phase aberrations in the hologram plane, the re-
sidual amplitude aberrations in the hologram plane are
not compensated for by the automatic adjustment. This
residual amplitude aberration in the hologram plane is
converted to distortion in the image plane as shown in
Figs. 22(b) and 22(c).
Fig. 21. (a)–(c) The correction of the tilt is done in the hologram
plane, and the aberration compensation is performed in the im-
age plane. (d),(e) Compensation with ⌫
S
H,C
with seventh-order
standard polynomial 2D fitting. (a), (d) Hologram, plane phase
images. (b), (e) and (c), (f), respectively, amplitude and phase im-
ages in the image plane. The image distortion clearly visible in
(b), (c) is compensated in (e), (f).
Fig. 20. Amplitude (left) and phase (right) reconstructions after
⌫
S
H,C
adjustment: (a) d =13.3 cm, (b) d = −6.9 cm, (c) d
1
=13.3 cm
and d
2
=−6.9 cm, (d) M =0.56 共d=−7.1 cm兲 and P
02
H
=−4.7⫻10
−10
,
(e) M=0.56 共d=7.95 cm兲 and P
20
H
=4.7⫻10
−10
.
3188 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
To overcome this residual distortion, a NPL could be
placed in the plane where the aberrations are introduced,
or in other words in the plane where there are phase-only
aberrations. We do not treat this solution further for two
reasons. First, it is not evident that such a plane exists,
because the aberrations are produced by different optics
(here principally by the ball lens and the field lens, but
the other optics do contribute also). Second, even if this
plane existed and its position could be defined, two nu-
merical propagations would need to be performed to re-
construct the corrected wavefront (from the hologram
plane to the phase-only aberration plane and then from
the phase-only aberration plane to the image plane), and
that would not be suitable in terms of time-consuming re-
construction.
To keep to a single numerical propagation, the distor-
tion compensation is applied in the hologram plane by ad-
justing manually the NPL parameters to minimize the
distortion in the image plane. In Fig. 22(d), the primary
spherical term parameter of the NPL ⌫
Z
H,C
is adjusted
共Z
10
=9.83⫻ 10
−7
兲 to compensate for the distortion [see
Figs. 22(e) and 22(f)] that is not yet totally compensated
for by the 2D fitting procedure method [see Figs. 22(b)
and 22(c)]. We note that the introduced phase term in the
hologram plane [see Fig. 22(d)] produces a phase defor-
mation in the image plane that is compensated automati-
cally by a sixth-order NPL ⌫
S
I,C
as presented in the phase
image [see Fig. 22(f)].
6. CONCLUSION
We have presented in this paper numerical methods to
compensate for all aberrations. Classically, aberrations
are minimized by use of different well-designed optical
components placed successively in the optical path. Our
technique is similar but has the advantage of using a
maximum of two numerical parametric lenses placed in
the hologram and in the image plane. Furthermore, we
demonstrated that these numerical lenses can be com-
puted to achieve a numerical magnification and shift of
the region of interest. This last feature gives us the ability
to compensate for chromatic aberrations, the scaling com-
ing from different reconstruction distances, and the speci-
men shift that can occur between two hologram acquisi-
tions.
In addition, the technique has the advantage of mini-
mizing the number of parameters that should be adjusted
by the operator. Indeed, automatic fitting procedures
showed that phase aberrations and image distortion can
be suppressed, in particular with the compensation for
aberration introduced by the use of a cylindrical lens or a
ball lens as MO. This feature allows low-cost setups that
could be constructed with inexpensive optical components
that produce aberrations.
ACKNOWLEDGMENTS
The authors thank M. Gu and H. Tiziani for fruitful dis-
cussion and acknowledge gratefully research grant
205320-103885/1 from the Swiss National Science Foun-
dation and research grant 7152.1 from the Innovation
Promotion Agency (KTI/CTI).
Corresponding author T. Colomb’s e-mail address is
tristan.colomb@a3.epfl.ch.
REFERENCES
1. U. Schnars and W. P. O. Jüptner, “Digital recording and
numerical reconstruction of holograms,” Meas. Sci. Technol.
13, R85–R101 (2002).
2. E. Cuche, F. Bevilacqua, and C. Depeursinge, “Digital
holography for quantitative phase-contrast imaging,” Opt.
Lett. 24, 291–293 (1999).
3. E. Cuche, P. Marquet, and C. Depeursinge, “Simultaneous
amplitude-contrast and quantitative phase-contrast
microscopy by numerical reconstruction of Fresnel off-axis
holograms,” Appl. Opt. 38, 6994–7001 (1999).
4. P. Ferraro, S. De Nicola, A. Finizio, G. Coppola, S. Grilli, C.
Magro, and G. Pierattini, “Compensation of the inherent
wave front curvature in digital holographic coherent
microscopy for quantitative phase-contrast imaging,” Appl.
Opt. 42, 1938–1946 (2003).
5. G. Pedrini, S. Schedin, and H. J. Tiziani, “Aberration
compensation in digital holographic reconstruction of
microscopic objects,” J. Mod. Opt. 48, 1035–1041 (2001).
6. A. Stadelmaier and J. H. Massig, “Compensation of lens
aberrations in digital holography,” Opt. Lett. 25,
1630–1632 (2000).
7. S. De Nicola, A. Finizio, G. Pierattini, D. Alfieri, S. Grilli, L.
Fig. 22. (a)–(c) Compensation with ⌫
S
H,C
with eighth-order stan-
dard polynomial 2D fitting; (d)–(f) after adding manual adjust-
ment of primary spherical Zernike term Z
10
=9.83⫻10
−7
and a
compensation of the resulting phase deformation in the image
plane by an automatic adjustment of ⌫
S
I,C
(six orders). (a) and (d)
show the hologram plane phase image; (b), (e) and (c), (f), respec-
tively, show the amplitude and phase images in the image plane.
The image distortion clearly visible in (b), (c) is totally compen-
sated in (e), (f).
Colomb et al. Vol. 23, No. 12 /December 2006 / J. Opt. Soc. Am. A 3189
Sansone, and P. Ferraro, “Recovering correct phase
information in multiwavelength digital holographic
microscopy by compensation for chromatic aberrations,”
Opt. Lett. 30, 2706–2708 (2005).
8. S. De Nicola, P. Ferraro, A. Finizio, and G. Pierattini,
“Wave front reconstruction of Fresnel off-axis holograms
with compensation of aberrations by means of phase-
shifting digital holography,” Opt. Lasers Eng. 37, 331–340
(2002).
9. S. Grilli, P. Ferraro, S. De Nicola, A. Finizio, G. Pierattini,
and R. Meucci, “Whole optical wavefields reconstruction by
digital holography,” Opt. Express 9, 294–302 (2001).
10. S. De Nicola, P. Ferraro, A. Finizio, and G. Pierattini,
“Correct-image reconstruction in the presence of severe
anamorphism by means of digital holography,” Opt. Lett.
26 974–976 (2001).
11. S. De Nicola, A. Finizio, G. Pierattini, P. Ferraro, and D.
Alfieri, “Angular spectrum method with correction of
anamorphism for numerical reconstruction of digital
holograms on tilted planes,” Opt. Express 13, 9935–9940
(2005).
12. S. De Nicola, P. Ferraro, A. Finizio, S. Grilli, and G.
Pierattini, “Experimental demonstration of the
longitudinal image shift in digital holography,” Opt. Eng.
(Bellingham) 42, 1625–1630 (2003).
13. T. Colomb, E. Cuche, F. Charrière, J. Kühn, N. Aspert, F.
Montfort, P. Marquet, and C. Depeursinge, “Automatic
procedure for aberration compensation in digital
holographic microscopy and applications to specimen shape
compensation,” Appl. Opt. 45, 851–863 (2006).
14. P. Ferraro, S. De Nicola, G. Coppola, A. Finizio, D. Alfieri,
and G. Pierattini, “Controlling image size as a function of
distance and wavelength in Fresnel-transform
reconstruction of digital holograms,” Opt. Lett. 29, 854–856
(2004).
15. J. Kato, I. Yamaguchi, and T. Matsumura, “Multicolor
digital holography with an achromatic phase shifter,” Opt.
Lett. 27, 1403–1405 (2002).
16. I. Yamaguchi, T. Matsumura, and J. Kato, “Phase-shifting
color digital holography,” Opt. Lett. 27, 1108–1110 (2002).
17. P. Almoro, M. Cadatal, W. Garcia, and C. Saloma, “Pulsed
full-color digital holography with a hydrogen Raman
shifter,” Appl. Opt. 43, 2267–2271 (2004).
18. B. Javidi, P. Ferraro, S.-H Hong, S. D. Nicola, A. Finizio, D.
Alfieri, and G. Pierattini, “Three-dimensional image fusion
by use of multiwavelength digital holography,” Opt. Lett.
30, 144–146 (2005).
19. G. Indebetouw and P. Klysubun, “Optical sectioning with
low coherence spatiotemporal holography,” Opt. Commun.
172, 25–29 (1999).
20. M. K. Kim, “Tomographic three-dimensional imaging of a
biological specimen using wavelength-scanning digital
interference holography,” Opt. Express 7, 305–310 (2000).
21. A. Dakoff, J. Gass, and M. K. Kim, “Microscopic three-
dimensional imaging by digital interference holography,” J.
Electron. Imaging 12, 643–647 (2003).
22. P. Massatsch, F. Charrière, E. Cuche, P. Marquet, and C.
Depeursinge, “Time-domain optical coherence tomography
with digital holographic microscopy,” Appl. Opt. 44,
1806–1812 (2005).
23. A. Thelen, J. Bongartz, D. Giel, S. Frey, and P. Hering,
“Iterative focus detection in hologram tomography,” J. Opt.
Soc. Am. A 22, 1176–1180 (2005).
24. L. Martínez-León, G. Pedrini, and W. Osten, “Applications
of short-coherence digital holography in microscopy,” Appl.
Opt. 44, 3977–3984 (2005).
25. L. Yu and M. K. Kim, “Wavelength scanning digital
interference holography for variable tomographic
scanning,” Opt. Express 13, 5621–5627 (2005).
26. F. Charrière, A. Marian, F. Montfort, J. Kühn, T. Colomb,
E. Cuche, P. Marquet, and C. Depeursinge, “Cell refractive
index tomography by digital holographic microscopy,” Opt.
Lett. 31, 178–180 (2006).
27. F. C. Zhang, I. Yamaguchi, and L. P. Yaroslavsky,
“Algorithm for reconstruction of digital holograms with
adjustable magnification,” Opt. Lett. 29, 1668–1670
(2004).
28. F. Montfort, “Tomography using multiple wavelengths in
digital holographic microscopy,” Ph.D. dissertation (Ecole
Polytechnique Fédérale de Lausanne, Lausanne, 2005).
29. E. Cuche, P. Marquet, and C. Depeursinge, “Spatial
filtering for zero-order and twin-image elimination in
digital off-axis holography,” Appl. Opt. 39, 4070–4075
(2000).
30. “
ZEMAX: Optical Design Program, User’s Guide, Version
10.0” (Focus Software, Tucson, 2001), pp. 126–127.
31. F. Charrière, J. Kühn, T. Colomb, F. Monfort, E. Cuche, Y.
Emery, K. Weible, P. Marquet, and C. Depeursinge,
“Characterization of microlenses by digital holographic
microscopy,” Appl. Opt. 45, 829–835 (2006).
32. B. E. Saleh and M. C. Teich, Fundamentals of Photonics
(Wiley, 1991).
33. T. Colomb, F. Dürr, E. Cuche, P. Marquet, H. Limberger,
R.-P. Salathé, and C. Depeursinge, “Polarization
microscopy by use of digital holography: application to
optical fiber birefringence measurements,” Appl. Opt. 44,
4461–4469 (2005).
34. F. Montfort, T. Colomb, F. Charrière, J. Kühn, P. Marquet,
E. Cuche, S. Herminjard, and C. Depeursinge,
“Submicrometer optical tomography by multiple
wavelength digital holographic microscopy,” Appl. Opt. (to
be published).
3190 J. Opt. Soc. Am. A / Vol. 23, No. 12 / December 2006 Colomb et al.
