Direct extraction of mean particle size from a digital

hologram

Loı̈c Denis, Corinne Fournier, Thierry Fournel, Christophe Ducottet

Laboratoire Traitement du Signal et Instrumentation

UMR CNRS 5516
Bât. F, 10 rue Barrouin, 42000 Saint-Etienne, France

loic.denis@univ-st-etienne.fr

Dominique Jeulin

Centre de Morphologie Mathématique

Ecole des Mines de Paris
35 rue saint-Honoré, F-77300 Fontainebleau, France

Digital holography, consisting of both acquiring the hologram image in a

digital camera and numerically reconstructing the information, offers new and
faster ways to make the most of a hologram. We describe in this paper a new
method to get the rough size of particles in an inline hologram. It relies on a
property specific to interference patterns in Fresnel holograms: self-correlation
of a hologram gives access to size information.
The proposed method is both simple and fast and gives results with accept-
able precision. It suppresses all “numerical depth of focus” related problems
when analysing large depth volumes.
c 2005 Optical Society of America

OCIS codes: 090.1760, 100.2000, 110.2960, 120.3940, 120.7250.

1
1. Introduction

The potential of holography to measure particle size has been pointed out since its very
beginnings.1 The development of optical holography applications in fluid mechanics2, 3 proved
the capability of holography to give access to both particle size and position in 3D.
Holography is a technique that allows a description of a 3D distribution of particles to be
stored on a two dimension detector. Digital holography suppresses the wet chemical process-
ing step and makes it possible to acquire and analyze volume objects within a short time, the
limiting factor being the numerical processing speed. Many contributions have been brought
over last past years on particle digital holography (see for example papers published in 4). On
the one hand, efforts are put towards precise results through tridimensional image process-
ing of the reconstructed volume. The objective is quality in this quest to reach holography’s
limit.5, 6, 7 On the other hand, the need for fast processes adequate with inline monitoring or
dynamic studies raises new challenges. Speed becomes primordial over precision in this class
of applications. Three dimensional volume reconstruction and analysis are no longer feasible
and information has to be directly extracted from the hologram. We describe in this paper
a novel method corresponding to this latter approach. The aim of the method is to obtain
the mean size of particles from an inline hologram. It is compared with a classical technique
from the former approach.
Digital reconstruction of a hologram is achieved through numerical solving of the diffrac-
tion integral. Fresnel’s transform has been shown8 to be well adapted to the reconstruction
of particle holograms. A focus plane located at a distance zr from the recorded hologram IH
has a complex amplitude Azr given by:8

Azr = IH ∗ hzr

where ∗ stands for bidimensional convolution and hzr denotes Fresnel function with pa-
rameter zr : hzr (x, y) = 1/(jλzr ) exp (jπ(x2 + y 2 )/(λzr )) (Fresnel function is underlined to
emphasize its complex nature, j denotes the imaginary unit).
In the case of particle holograms, the analysis of the reconstructed volume can lead to

2
particle per particle size measurements.9, 6, 10 Global estimation of the particles average size
is however possible in a simpler and therefore faster method through direct analysis of the
hologram. We describe this direct extraction method in the following section. The key step
to get the size information is to compute the hologram self-correlation. The result of this
operation is further analysed in section 3. Results on numerically generated and experimental
holograms are given in section 4.

2. Principle of the method

We describe here how to extract a global size information from an in-line hologram of spher-
ical opaque particles. In-line digital holography setup records the intensity of the incident
coherent beam, diffracted when crossing the object and free propagated towards the digi-
tal sensor (see fig. 1). The reference wave and the object wave are not separate beams but
a unique one which gives a good stability to the setup adapted to industrial application
constraints. However the drawback of this experimental setup however is the restriction to
low concentration particle volumes. Royer11 qualifies inline particle holograms to be of good
quality if the projected particles area is less than 1% of the hologram area.
We derive in this section the hologram intensity expression and show that its self-
correlation carries the particle aperture self-correlation information. This information is con-
nected with the particles size information.

2.A. Approximated expression of the hologram intensity

Let us derive a classical approximation for the hologram intensity expression.12 An opaque
spherical particle may be considered as a diffractive disk (whose aperture will be denoted
ϑri with ri the radius) located at the (xi , yi , zi ) position of the particle center. The diffracted
unitary plane wave at the recording distance in Fresnel approximation is given by:13

Az = ti ∗ hzi

= 1 − ϑri ∗ hzi ∗ δxi ,yi

3
where ti represents the transmittance of the plane zi and δxi ,yi is Dirac’s delta function
located at point (xi , yi ).
If we consider the incoming wave to remain quasi-plane for each particle, which is justified
for low density holograms, the resulting diffracted wave in the recording plane becomes:
N
X
Az ≈ 1 − ϑri ∗ hzi ∗ δxi ,yi (1)
i=1

Sensors only respond to intensity, related to the complex amplitude via the following
relation: IH = |Az |2 . The recorded hologram intensity can therefore be expanded as follows:
N
X N X
X N    
IH = 1 − 2 ϑri ∗ ℜ(hzi ) ∗ δxi ,yi + ϑri ∗ hzi ∗ δxi ,yi × ϑrj ∗ h−zj ∗ δxj ,yj .
i=1 i=1 j=1

where ℜ is used to denote the real part of a complex number.

The second term is predominant over the third one under the hypothesis of both small
particles and low concentration holograms.14 This last approximation results in the following
“pattern-oriented” hologram intensity expression where each particle induces an interference
pattern and where inter-particle interferences are neglected:
N
X
IH ≈ 1 − 2 ϑri ∗ ℜ(hzi ) ∗ δxi ,yi . (2)
i=1

2.B. Hologram information coding

The interference pattern produced by a particle is characteristic of its position and size
(see fig. 2). The pattern center indicates the transversal location (xi , yi ) of the particle. The
frequency modulation of the concentric rings carries the distance information (zi ) whereas
their amplitude modulation depends on the distance to diameter ratio . Size measurements
therefore seem to require digital reconstruction followed by distance estimation as a first
step.

2.C. Self-correlation

As we shall underline later, self-correlation plays a key role to reach direct numerical ex-
traction of size information in a hologram. Self-correlation is commonly used in the image

4
processing field to identify the size of different underlying structures in an image.15 For in-
stance, the relationship between the self-correlation of a binary image of spheres and their
size distribution was studied in depth in the field of stereology.16 It is therefore quite legiti-
mate to apply this technique on our holograms as a potential direct size extraction method.
It should however give us access to pattern sizes (see fig. 3(a)), proportional to λzi /ri ratio
instead of the desired particle size. We show in this section that in the case of these very
special patterns, self-correlation however gives the particle size (fig. 3(b)).
Let us recall a key property of Fresnel functions:17 the duality, which may be written as:
hz ⋆ hz = δ, using ⋆ as a symbol for (complex) correlation. The real part of Fresnel functions
verify a weakened version: ℜ(hz )⋆ℜ(hz ) = 12 (δ+ℜ(h2z )). This implies that the self-correlation
expression of the centered hologram I˜H can be expanded into three terms:
N
I˜H ⋆ I˜H =2
X
ϑr i ∗ ϑr i (3a)
i=1
N
X
+2 ϑri ∗ ϑri ∗ ℜ(h2zi ) (3b)
i=1
XN XN
+4 ϑri ∗ ϑrj ∗ ℜ(hzi ) ∗ ℜ(hzj ) ∗ δxi −xj ,yi −yj (3c)
i=1 j=1
j6=i

An example of such a self-correlation can be seen in figure 4 as a 3D surface representation.

This representation emphasizes the breaking down into 3 terms: the central peak corresponds
to term (3a), the concentric waves to term (3b) and the smaller peaks spread all over the
surface to the final term (3c).
The first term (3a) is the sum of the self-correlation of each particle’s aperture. It is
directly connected with particle sizes. We name the second term (3b) “twin self-correlation”
as it originates from holography’s classical virtual images. The self-correlation operation,
acting as a kind of self-restitution, gives rise to both the “objects” (i.e. particle aperture self-
correlations) and their “twins” that seem located at a distance twice each particle recording
distance. Finally, the third term (3c) arises from inter-particles correlations and therefore
depends on the spatial distribution of the particles. Contrary to the first two, this last term
is of anisotropic nature.

5
 If we consider both (3b) and (3c) as noise, or perturbative terms, the centered hologram
self-correlation unexpectedly depends only on the particle aperture self-correlations. The ob-
tained self-correlation is that of an image with every particle in focus. The direct correlation
of holograms has been studied previously by Poon and Kim and applied to 3D object match-
ing18 and 3D object location.19 To the best of our knowledge, the object size information
has never been connected with the hologram self-correlation before.

2.D. Hologram processing chain

The relative importance of each term in equation 3 is investigated more deeply in section 3.
Methods used to reduce the influence of the perturbative terms (3b) and (3c) are described
there. We focus here on the hologram processing chain we suggest for direct particle size
extraction (fig. 6(b)). This processing chain is compared with that of the classical approach
(fig. 6(a)).
The centered hologram is analyzed through self-correlation calculation. Since useful in-
formation (3a) is isotropic, we can restrict ourselves to a 1D profile of the correlation. The
size information is then directly extracted from this profile (see figure 6(b)). The processing
chain is both simple and fast since the most heavy computing step is that of self-correlation
which is done through two Fast Fourier Transforms. Let us note that as the required range
for self-correlation is limited to its central part (the range width equals the particle diameter)
direct estimation of the correlation (i.e. without Fourier transforms) could be faster in some
cases.
The second step, namely 1D profile extraction, is achieved through a radial averaging
operation. The final and last step consists in computing a best fitting line to get the rough
mean size.
Classical particle size measurements using digital holography is achieved in four steps (fig.
6(a)), requiring quite heavy computing. The reconstruction step (first step in figure 6(a))
is the most computationally expensive. It requires at least 10 Fast Fourier Transform to
reconstruct a volume slice around a particle depth position. In case of the study of large depth

6
volumes, 10 Fast Fourier Transform per particle are necessary. The reconstructed volume
then has to be segmented (step 2) and labeled (step 3). Both these operations are applied on
tridimensional images representing heavy memory loads. The final measurement step gives
the size of each particle which leads to the granulometric distribution. This approach gives
more complete and precise size information at a higher computing cost.

3. Relative importance of self-correlation components

We described in previous section 2.C the origin of each component of the centered hologram
self-correlation. We focus in this section on their relative influence compared with particle
aperture self-correlations (term (3a)).
The approximated expression for the diffracted wave amplitude given in equation 1 makes
it possible to numerically generate realistic holograms. A hologram intensity distribution is
obtained by taking the squared modulus of the computed complex amplitude. The phase
information available in numerical simulations offers the possibility to illustrate the relative
importance of each component of the centered hologram self-correlation. As an illustration,
we generated numerically a hologram (see figure 7(a)) and computed its self-correlation pro-
file (fig. 7(b)). This profile is the sum of different components drawn in figure 8: (a) particles
aperture self-correlations, (b) “twin self-correlations” , (c) particles cross-correlations and
(d) non linear terms.

3.A. Particle self-correlations

Term (3a) is the sum of each particle aperture self-correlation. For a disk of given
radius
r
 r, the expression of
 its self-correlation can be derived analytically: Cr (h) =
h2
h 1−
(2r)2
2r 2 arccos( h ) −  for h < 2r and Cr (h) = 0 for larger values of h. The graphical
2r 2r

representation of Cr (fig. 5) is quite close to its tangent at the origin. The intersection of
this tangent with the abscissa axis is proportional to r and therefore can be used to retrieve
the diameter from the self-correlation profile.
The range [0, 2r] of this term is short compared to the hologram size: the whole information

7
we will search for is located at small values of h (see fig. 8(a)). Since particles are isotropic,
this term also is. In case of particles of different sizes in the hologram field, larger particles
are more influential than smaller ones as their energy (i.e. area of their aperture) is more
important. Moreover, intereference patterns that are partially lost outside the hologram
boundaries tend to reduce the contribution of the corresponding particles.

3.B. Twin self-correlations

Term (3b) is a set of concentric rings whose frequency modulation is roughly that of a
Fresnel function with parameter 2λz̄ (z̄ denoting the average z distance of the particles). It
is negligible in the range [0, 2r] for large enough values of λz̄. It reaches indeed its maximum
√
value close to the maximum of ℜ(h2z ): for h ≈ λz which is well over 2r. It’s maximum
value reduces in proportion to 1/z.
This term is computed in figure 8(b) from the numerically generated complex amplitude
 ∗
 P
Az using the following relation: ℜ Ãz ⋆ Ãz = N i=1 ϑri ∗ ϑri ∗ ℜ(h2zi ). In this example the

maximal value of this term in the range [0, 2r] is around 0.3% the maximum of term (3a).

3.C. Particle cross-correlations

Term (3c) consists in peaks with the same spatial extent 2r as term (3a) but spread on
the whole correlation image (see figure 4). At low densities of particles – compatible with
holographic inline setup – this term is highly anisotropic. It is therefore strongly attenuated
through a radial average. This term depends on the particles spatial distribution.
The curve drawn in figure 8(c) corresponds to both this term and the boundary effects.
It is indeed computed as the difference between the self-correlation of a centered hologram
generated using equation (2) and terms (3a) and (3b). Term (3a) has been obtained using the
analytical expression given in section 3.A and therefore does not take into account boundary
effects appearing when interference patterns go beyond the boundaries of the hologram.
The influence of this term in figure 8(c) is less than 1% the maximum of term (3a).

8
3.D. Non linear terms

Remaining terms come from the linear approximation made in equation 2. These terms might
seem more problematic as they are not easily modeled and then suppressed. Moreover, they
sum up at the origin of the correlation. Their influence for low concentrated holograms of
relatively small particles however appears to be of limited weight: less than 1% in figure 8(d).

4. Results on numerical generated and experimental holograms

4.A. Processing of numerically simulated holograms

Since the perturbative terms listed in the previous paragraphs (b) to (d) are nearly negligible
in our range of interest [0, 2r], we will process in a first approach the centered hologram self-
correlation as if it were formed only of term (3a). To increase robustness to noise, we will
however not use the root of the correlation tangent at the origin (fig. 5) but the root of
the best fitting line computed in a range coarsely corresponding to [r/5, 3r/5]. This range
is chosen to minimize the influence of the perturbative terms. It is refined using the root
found as an estimate for r. The correspondence between the root found and the actual
particle diameter is set up during a calibration step. This step consists in comparing the
root computed through self-correlation to the known calibrating size.
The experimental parameters used for the numerical simulation are described in figure
9(a). Spatial distributions of 10 particles have been randomly generated inside a volume
with a large thickness: 50mm. Each particle has got the same size for a given hologram.
Holograms of particles for 7 different sizes have been generated. For each defined size, 25
holograms corresponding to 25 independent realizations of the particles spatial distribution
were generated leading to the study of 175 holograms.
A sample hologram is shown in figure 9(b). Even though every particle has the same size,
the interference patterns vary from one particle to another because of the large depth of the
studied volume (from 100mm to 150mm). This effect is also pictured in the aforementioned
figure 3.
Correlation profiles for sets of particles at each studied size (70µm to 130µm by 10µm

9
steps) are drawn in figure 9(c). The lowest curve corresponds to the smallest particle size:
70µm while the highest is bound to the maximum particle size: 130µm. The linear dependence
of profiles slopes on particle sizes is clearly visible.
Quantitative results are displayed in table 1. For each diameter appearing on the first line
of the table, 25 holograms were generated. The absolute and relative errors were computed.
Their averages over the 25 realisations are displayed in the second and third rows. A numerical
calibration was carried out with a particle of 90µm. The relative error remains small (less
than 4%) over a wide range of diameters.
Numerical simulation points out that the method is valid when applied to realistic-looking
holograms. The computed errors seem adequate with some real world applications.

4.B. Hologram of a spray

We perform in this section a check of the self-correlation method on an experimental holo-

gram. The setup is described in the next paragraph. The classical approach based on 3D
analysis of the reconstructed planes is carried out as a reference measurement. Our new cor-
relation based approach is then applied to the hologram and the obtained size is compared
with that of the reference (classical) approach.
The studied object is a cloud of water droplets produced by a water spray (the velocity
of the particles remained less than 10m/s). The spray was lightened by a collimated beam
(see inline setup in figure 1) produced by a YAG pulsed laser (λ = 532nm) whose pulse
duration has been set to 7ns. The hologram was therefore almost instantaneously recorded.
The camera used has a dynamic range of 12 bits and its sensor is a 1280 × 1024 CCD matrix
of 6.7µm width full-square pixels. The studied part of this hologram (830 × 744) is shown
on figure 10(a).
The 3D particle localizing and diameter measurements have been done accordingly to the
hologram processing chain depicted in figure 6(a) and presented in section 2.D. The droplets
are located within a bounding volume of 60mm depth and the size particle ranges from 80µm
to 110µm (see table 2 and figure 11 for a 3D reconstruction). The mean diameter of the ten

10
holographed particles is estimated at 84µm ± 7µm.
Direct extraction of the size information was achieved according to the self-correlation
approach. The obtained correlation profile is drawn in figure 10(b). The whole hologram
analysis using this approach takes less than 2s to compute on an average PC. The size found
from the correlation profile is 91µm. This size appears to be over-estimated, which is easily
explained by the size variations from the smallest to the biggest particle. As previously
underlined in section 3.A, larger particles have a higher weight than smaller ones in the
correlation profile. The size found using the correlation approach then is in accordance with
the results obtained through the classical approach.

5. Conclusion

Classical particle size measurement from a digital hologram require the reconstruction of
more than 10 images per particle. This leads to heavy computing unsuitable with inline
monitoring. In this study we have described and applied experimentally a direct and fast
method to extract the mean particle diameter from an inline hologram. The principle of the
method can be seen as an original application of the classical self correlation technique used
in image processing to extract the mean size of objects. The key of the method is the duality
property of Fresnel function which allows a sort of “self reconstruction” when computing the
self correlation. Each interference pattern is indeed convolved with the recording function
stored in its concentric rings.
This approach differs from classical ones in that it does not require to reconstruct the
object but extracts the wanted information directly from the hologram.
The advantages of this method are the simplicity and the stability of the optical setup,
the large depth of field of holography and the speed of the image processing (compared with
the classical approach). It should be well adapted to a large class of industrial applications.
This method can be used as a first step for a refined analysis of the reconstructed volume.
Techniques such as those based on correlation require the knowledge of pattern sizes to
access sub-pixel particle locating.20, 9 The described method gives quickly this necessary size

11
information.
Let us notice that this method can be extended to objects with other shapes than disks
and then allows to obtain rough measurements of sizes and the main orientation in the set
of objects. Studies carried in the field of stereology also describe how to extract more infor-
mation from a correlation profile, namely how to estimate parameters of a given distribution
function of sizes.

12
References

1. B. J. Thompson. A new method of measuring particle size by diffraction techniques.

Japanese Journal of Applied Physics, 4(1):302–307, 1965.
2. C. S. Vikram. Particle field holography. Cambridge University Press, 1992.
3. K.D. Hinsch. Holographic particle image velocimetry. Measurement Science and Tech-
nology, 13:R61–R72, 2002.
4. K. D. Hinsch and S. F. Herrmann. Special Feature : Holographic Particle Image Ve-
locimetry, volume 15 of Measurement Science & Technology. IOP, United Kingdom,
2004.
5. J. Müller, V. Kebbel, and W. Jüptner. Characterization of spatial particle distributions
in a spray-forming process using digital holography. Measurement Science & Technology,
15(4):706–710, 2004.
6. G. Pan and H. Meng. Digital holography of particle fields: reconstruction by use of
complex amplitude. Applied Optics, 42:827–833, 2003. R21.
7. C. Fournier, C. Ducottet, and T. Fournel. Digital in-line holography: influence of the re-
construction function on the axial profile of a reconstructed particle image. Measurement
Science & Technology, 15:686, 2004.
8. T.M. Kreis, M. Adams, and W. Juptner. Methods of digital holography : A comparison.
In Christophe Gorecki, editor, SPIE97, volume 3098, pages 224–233, Munich, Germany,
1997.
9. S. Murata and N. Yasuda. Potential of digital holography in particle measurement.
Optics and Laser Technology, 32(7 8):567–574, 2000.
10. M. Malek, S. Coëtmellec, D. Allano, and D. Lebrun. Formulation of in-line hologra-
phy process by a linear shift invariant system: application to the measurement of fiber
diameter. Optics communications, 223:263–271, 2003.
11. H. Royer. An application of high-speed microholography: the metrology of fogs. Nouv.
Rev. Opt, 5:87–93, 1974.

13
12. L. Onural and P.D. Scott. Digital decoding of in-line holograms. Optical Engineering,
26(11):1124–1132, 1987. Corinne, Holographic Particle Diagnostics Vikram.
13. J.W. Goodman. Introduction to Fourier Optics. Mc Graw-Hill, 1996.
14. C. Buraga-Lefebvre, S. Coëtmellec, D. Lebrun, and C. Özkul. Application of wavelet
transform to hologram analysis: three-dimensional location of particles. Optics and
Lasers in Engineering, 33(6):409–421, 2000.
15. J. Serra. Image analysis and mathematical morphology. Academic Press, London, 1982.
16. D. Jeulin. Random models for the morphological analysis of powders. Journal of Mi-
croscopy, 172(Part 1):13–21, 1993.
17. M. Liebling, T. Blu, and M. Unser. Fresnelets : New multiresolution wavelet bases for
digital holography. IEEE Transactions on image processing, 12(1):29–43, 2003.
18. T.-C. Poon and T. Kim. Optical Image Recognition of Three-Dimensional Objects.
Appl. Opt. , 38:370–381, January 1999.
19. T. Kim and T.-C. Poon. Three-dimensional matching by use of phase-only holographic
information and the Wigner distribution. Optical Society of America Journal A, 17:2520–
2528, December 2000.
20. C. Fournier, C. Barat, C. Ducottet, and T. Fournel. Digital holography applied to piv :
hologram reconstruction and selection of a cloud of particles in the reconstructed volume.
In Pacific Symposium on Flow Visualization and Image Processing, Chamonix, France,
2003.

14
 Table 1. Results of the correlation method on numerically generated holo-
grams. In the first row are displayed the correct diameters used to generate
the holograms. The second and third row give the errors made when retrieving
the diameters.

Diameter (µm) 70 80 90 100 110 120 130

Absolute error⋆ (µm) 1.6 1.7 0.8 1.0 1.7 3.2 4.8
Relative error⋆ (%) 2.3 2.1 0.9 1.0 1.5 2.7 3.7
⋆
Displayed errors are average errors on computed diameters from 25 numerically generated holograms

15
 Table 2. Positions and sizes of the particles in figure 10(a) experimental holo-
gram. Displayed data has been obtained via analysis of the reconstructed vol-
ume using classical techniques.

Particle 1 2 3 4 5 6 7 8 9 10
x (mm) 3.326 2.287 4.460 2.061 3.658 4.694 4.023 3.610 0.737 1.884
Position y (mm) 3.972 4.754 4.046 3.738 2.955 3.692 1.073 1.403 3.428 0.809
z (mm) 40.0 45.8 47.5 49.5 65.0 69.6 69.8 81.0 95.4 101.3
Diameter (µm) 79 91 96 65 79 71 79 81 89 108

16
List of Figures

1 In-line digital holography setup. . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Hologram information coding: a line profile (in black stroke), Fresnel radial
frequency modulation function (dash point line) and the diffraction envelope
(dashed line). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Patterns of a particle with constant radius r for different recording distance
z and their associated self-correlation profile. . . . . . . . . . . . . . . . . . . 20
4 Centered hologram self-correlation (numerical simulation). . . . . . . . . . . 21
5 Self-correlation of a disk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6 Hologram processing chain. . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
7 A hologram and its self-correlation profile. . . . . . . . . . . . . . . . . . . . 24
8 Breaking down of fig. 7(b) correlation profile into the 4 constituent terms: (a)
particle apertures self-correlations term, (b) “twin self-correlations term, (c)
cross-correlations term and (d) non linear terms. . . . . . . . . . . . . . . . . 25
9 Application of the size extraction method on numerically generated holograms. 26
10 Experimental result on a hologram of a water spray. . . . . . . . . . . . . . . 27
11 Reconstructed volume from the experimental hologram of figure 10. . . . . . 28

17
 Beam expander y PC
and spatial filter CCD
Cloud of particles
z x
Laser

{xi,yi,zi}

Fig. 1. In-line digital holography setup.

18
 1

0.8
Hologram (centered) intensity

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8

−1
−1000 −500 0 500 1000
x(µm)

Fig. 2. Hologram information coding: a line profile (in black stroke), Fresnel
radial frequency modulation function (dash point line) and the diffraction
envelope (dashed line).

19
 z = 400r z = 800r z = 1000r z = 1200r z = 1400r

(a) Diffraction pattern produced by a particle located at an increasing distance z

1 1 1 1 1
C(h)

C(h)

C(h)

C(h)

C(h)
0 0 0 0 0
0 7 0 7 0 7 0 7 0 7
h h h h h
(b) Corresponding correlation profiles

Fig. 3. Patterns of a particle with constant radius r for different recording

distance z and their associated self-correlation profile.

20
Fig. 4. Centered hologram self-correlation (numerical simulation).

21
 π 2
energy: 4D

slope: −D

zero: D

0 h
π
root of the tangent at the origin: 4D

Fig. 5. Self-correlation of a disk.

22
 reconstructed segmented labeled
hologram
volume volume volume
numerical
→ restitution → → segmentation → → labeling →
> 10 FFT

granulometry

← measurement

(a) Classical hologram processing chain for size measurements

self-
hologram profile “mean” size
correlation
Self-
linear
→ correlation → → radial mean → → → 84µm
regression
2 FFT

(b) The proposed direct particle size extraction scheme

Fig. 6. Hologram processing chain.

23
 0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 100 200 300 400 500

(a) Numerically generated hologram (b) Corresponding centered self-correlation

Fig. 7. A hologram and its self-correlation profile.

24
 1
0.025
0.9
0.02
0.8
0.015
0.7
0.01

0.6 0.005

0.5 0

0.4 −0.005

0.3 −0.01

0.2 −0.015

0.1 −0.02

0
0 100 200 300 400 500 0 100 200 300 400 500

(a) (b)

−3
x 10
0.02
0
0.015

0.01 −2

0.005
−4
0

−0.005 −6

−0.01
−8
−0.015

−0.02 −10

−0.025
0 100 200 300 400 500 0 100 200 300 400 500

(c) (d)

Fig. 8. Breaking down of fig. 7(b) correlation profile into the 4 constituent
terms: (a) particle apertures self-correlations term, (b) “twin self-correlations
term, (c) cross-correlations term and (d) non linear terms.

25
Source:
1
• wavelength λ = 0.6328µm 0.9

Object: 0.8

0.7
• 10 particles (opaque spheres)
0.6
• same diameter φ
0.5
• spread randomly in a 6mm × 0.4
6mm × 50mm volume
0.3

Sensor: 0.2

0.1
• recording distance: z = 150mm
0
• 1024 × 1024 pixels 0 5 10 15 20

• pixel size: 6.7µm × 6.7µm

(a) Parameters used in the numerical sim- (b) a sample numerically generated holo- (c) Correlation profiles and best fitting
ulations gram (φ = 110µm) curves for holograms with particles from
φ = 70µm to φ = 130µm

Fig. 9. Application of the size extraction method on numerically generated

holograms.

26
 1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
0 50 100 150

(a) Experimental digital hologram of water (b) Correlation profile computed on hologram (a)
droplets in the air

Fig. 10. Experimental result on a hologram of a water spray.

27
 z
x

Fig. 11. Reconstructed volume from the experimental hologram of figure 10.

28