1. See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/309149477
Latent fingerprint wavelet transform image
enhancement technique for optical coherence
tomography
Conference Paper · September 2016
DOI: 10.1109/ICAIPR.2016.7585203
CITATIONS
0
READS
3
3 authors:
Sisanda Makinana
Council for Scientific and Industrial Research…
6 PUBLICATIONS 4 CITATIONS
SEE PROFILE
Nontokozo P. Khanyile
Council for Scientific and Industrial Research…
8 PUBLICATIONS 12 CITATIONS
SEE PROFILE
Rethabile Khutlang
Council for Scientific and Industrial Research…
10 PUBLICATIONS 104 CITATIONS
SEE PROFILE
All in-text references underlined in blue are linked to publications on ResearchGate,
letting you access and read them immediately.
Available from: Sisanda Makinana
Retrieved on: 07 November 2016

2. Latent Fingerprint Wavelet Transform Image
Enhancement Technique for Optical Coherence
Tomography
Sisanda Makinana
Modelling and Digital Science
Council of Scientific and Industrial Research
Pretoria, South Africa
Email: smakinana@csir.co.za
Portia N. Khanyile and Rethabile Khutlang
Modelling and Digital Science
Council of Scientific and Industrial Research
Pretoria, South Africa
Abstract—In crime investigation, fingerprint identification
plays a major role in identifying culprits. However, traditional
procedure of acquiring latent fingerprint tends to be destructive
and leads to a limitation of not being able to do further analyses
like DNA. Being able to acquire latent fingerprints without
physical contact with the surface could be advantageous. These
advantages are as follows; being able to acquire the imprint
multiple times, there is no physical or chemical processing of a
substrate, the substrate can be concurrently analysed for DNA
and can provide a non-destructive lifting of the fingerprint. An
Optical Coherence Tomography machine is one of the promising
technology that may be used for imaging latent fingerprint
without contacting or destructing the fingerprint impression left
on a substrate. However, owing to the coherent nature of the
image formation process, OCT images suffer from speckle noise
which limits the contrast of OCT images. In this paper, an
algorithm that enhances this OCT latent fingerprint image to
ensure reliable extraction of features is proposed. To test the
proposed algorithm latent prints were collected and stored as
a database. Two statistical and biometric system measurement
namely False Match Rate (𝐹 𝑀 𝑅) and Equal Error Rate (𝐸𝐸𝑅)
were used. The results of these two measures gives the 𝐹 𝑀 𝑅 of
3% and 𝐸𝐸𝑅 of 1.9% for denoised images which is better than
non-denoised images where the 𝐸𝐸𝑅 is 8.7%.
Index Terms—Forensic; Latent fingerprints; Optical coherence
tomography (OCT); Enhancement;
I. INTRODUCTION
Fingerprint recognition has rapidly become the widely used
technology in biometrics and forensic application [1]. In a
crime scene, fingerprints play an important role in terms of
identification of criminals. Latent prints are very important
in forensic as they are evidence of interaction between
an individual and the surface containing the fingerprint
impression. Various research has been conducted on lifting
these prints using contactless acquisition devices to overcome
disadvantages of traditional dactyloscopic technique [2], [3]
[4]. Most importantly, alteration from traditional fingerprint
processing may contaminate the evidence and even rule out
further evaluation from other perspective.
The fingerprints obtained in crime scenes are known as
latent fingerprints. Latent fingerprints are either visible or not
visible to human naked eye [5]. Forensic investigators use
various techniques to make invisible prints visible. However,
these techniques rely on adhesives and chemicals to detect,
visualize and preserve latent fingerprints on surfaces [6].
While such techniques are effective, they require surfaces
to be physically contacted and that may cause destruction
to the fingerprint resulting in no further analyses like DNA
testing. Furthermore, the lifting of a fingerprint from a surface
is an irreversible, unrepeatable procedure that separates the
fingerprint from its related surface.
Contactless acquisition of latent fingerprint where the
fingerprint is imprinted is advantageous in various ways [2].
Some of these advantages include: being able to preserve
both the surface and any latent prints in their unaltered form
and to secure them for further analysis [7]. Furthermore, the
substrate can be concurrently be analysed further for DNA.
On electrically conductive surfaces, the scanning Kelvin
probe can be used to visualize latent fingerprints contact-less.
The acquisition using contactless sensing and image pro-
cessing techniques of untreated latent fingerprint traces is
an upcoming opportunity in crime scene forensics. Optical
Coherence Tomography (OCT) is a high resolution contact-
less technology that may be used for acquisition of latent
fingerprint. In this paper, OCT was used to collect prints from
glass and plastic substrate. However, the OCT imaging process
produces speckle noise which makes it difficult for adequate
amount features to be extracted for recognition. The main is to
propose a Wavelet Transform Image Enhancement approach to
remove the speckle noise in order to improve feature extraction
performance.
The structured of the paper is as follows: Section II de-
scribes the acquisition device (OCT machine), the acquisition
process using the OCT machine and the latent print process
stages. Section III describes the proposed enhancement algo-
rithm based based phase preserving using wavelets transform.
Section IV describes further image filtering applied to latent
print. Section V provides experimental results and discussion.
ISBN: 978-1-4673-9187-0 ©2016 IEEE 7

3. Conclusion of the paper are provided in Section VI.
II. ACQUISITION PROCESS
A. Optical Coherence Tomography
In this work, to acquire latent fingerprint from a substrate
surfaces, a swept source OCT system namely; OCS1300SS
Thorlabs, was used. The swept laser optical source has a
central wavelength of 1325 nm and a spectral bandwidth of
100 nm. It has an average power output of 10.0 mW, and an
axial scan rate of 16 kHz. The system has a maximum imaging
depth of 3mm. Glass and plastic were used as the substrates.
It is an off-the shelf device that has not yet been adapted for
this specific task of laten print acquisition.
B. Acquisition Process
The latent fingerprint acquisition was done in the following
manner:
∙ Each individual placed a finger on substrate and leave a
fingerprint behind i.e. glass and plastic.
∙ The OCT system would start acquiring a fingerprint
impression left on the substrates.
∙ The OCT machine produces 512 3D volume data per
fingerprint, which must be rendered to produce corre-
sponding 2D images.
The example of the 3D volume is illustrated in Figure 1.
Fig. 1. Example of 512 3D data volume slices produced by OCT machine
C. Processing Stages of the OCT Latent images
The OCT machine used outputs a 3D fingerprint (as shown
in Figure 1) image which then is processed using [6] to
produce a 2D fingerprint. The 3-D lifted scan is processed
on a per cross-sectional image basis. First the cross-sections
are filtered to reduce the effects of speckle noise, then the
one dimensional Sobel edge detection is applied horizontally.
The detected edge represents the substrate surface plus the
latent fingerprint impression left on it. They are concatenated
together to form a 2-D segmented image of the lifted fin-
gerprint. The 3D to 2D conversion produces two 2D images:
the original 2D image and its compliment. Fig. 2 shows an
example of the two images produced.
Fig. 2. Original images (a) & (c) and their complements (b) & (d)
III. OCT LATENT FINGERPRINT ENHANCEMENT
A. Phase Preserving
In this paper, Peter Kovesi’s [8] phase preserving algorithm
was used as a benchmark. In order to preserve the phase data
at each point of latent print, extraction of the local phase
and amplitude information was executed. This was done by
applying (a discrete implementation of) the continuous wavelet
transform. Analysis of the signal was done by convolving the
signal with each wavelet. If I represents the OCT fingerprint
image, and 𝐺
(𝑒)
𝑛𝑑 and 𝐺
(o)
nd represents the even and odd Gabor
wavelets at a scale of n and direction d. The response is given
by:
[𝐼
(1)
𝑛𝑑 (𝑥, 𝑦), 𝐼
(2)
𝑛𝑑 (𝑥, 𝑦)] = [𝐺
(1)
𝑛𝑑 ∗ 𝐼(𝑥, 𝑦), 𝐺
(2)
𝑛𝑑 ∗ 𝐼(𝑥, 𝑦)] (1)
The values 𝐺
(𝑒)
𝑛𝑑 and 𝐺
(o)
nd may be defined as real and
imaginary parts of complex valued frequency components. For
every 𝑛 and 𝑑, we obtain a vector, and the local phase is given
by:
𝜑 𝑛𝑑 = 𝑎𝑡𝑎𝑛2(𝐼
(1)
𝑛𝑑 (𝑥, 𝑦), 𝐼
(1)
𝑛𝑑 (𝑥, 𝑦)) (2)
and amplitude is given by:
𝐴 𝑛𝑑(𝑥, 𝑦) =
√
𝐼
(1)
𝑛𝑑 (𝑥, 𝑦)2 + 𝐼
(2)
𝑛𝑑 (𝑥, 𝑦)2 (3)
ISBN: 978-1-4673-9187-0 ©2016 IEEE 8

4. In the frequency domain, the denoising structure entails
the process of determining a noise threshold at each scale
and shrinking the magnitudes of the filter response vectors
appropriately, and all this is done while the phase is kept
unchanged.
B. Implementing denosing algorithm
Image denoising is a process of recovering the image by
eliminating noise which is usually an aspect of electronic noise
[8]. Phase information is of critical significance to human vi-
sual perception. The most crucial part of the denoising process
is the thresholding. In this paper, a self-adaptive thresholding
was used. The technique of applying the deniosing algorithm
used in this paper is illustrated in Figure 3:
Fig. 3. The Denosing Process
The 3D volume latent print is processed by using the
technique in [6] to produce 2D latent print. Gaussian filters
are applied to the 2D image to remove ’salt and pepper’ noise
to the image. The image was transformed into a frequency
domain using Fast Fourier Transform [9], where the noise
component is more distinct and easily identified. Then the
latent print was convolved with Gabor wavelets with vari-
ous scale and direction. A thresholding operation was then
determined and applied to remove the noise, and finally the
transformation was inverted to reconstruct a noise-free image.
Fig. 4 shows an example of the results of the denoised images.
Fig. 4. FFT denoised fingerprints (a) & (c) with their complements (b) & (d)
IV. IMAGE FILTERING
The denoised images were then passed through a gaussian
filter followed by a median filter to further remove residual
noise from the images. The valleys were then morphologically
dilated to ensure that the was enough separation between
ridges and valleys, i.e. ridges do not touch. Finally, the contrast
was adjusted to increase the distinction between ridges and
valleys. Fig. 5 shows an example of the results of the further
processed image.
Fig. 5. Images filtered using Gaussian filter and median filter and contrast
adjusted
Ridges are assigned a value of 0 to make them black and
ISBN: 978-1-4673-9187-0 ©2016 IEEE 9

5. hence even more distinct. The complement image was then
inverted (valleys are changed to ridges) to make sure that the
features correspond with the original fingerprints. This makes
the complement image useful for minutiae validation during
the feature extraction step. Fig. 6 shows an example of the
results of the final enhanced image.
Fig. 6. Final enhanced images
V. EXPERIMENTAL RESULTS
Latent fingerprints were collected from 20 volunteers. La-
tent fingerprints were left on two plane object made of glass
and plastic where collection was done using OCT and stored
on a database. Since the OCT machine outputs a 3D volume
data and .ISO standard feature extractor uses 2D images,
the OCT prints were then converted to 2D latents prints.
Two 2D prints were randomly selected and are illustrated
in Figure 7. The denoised algorithm was applied to the
2D OCT prints and stored as a separate database. Features
were manually extracted from the 2D OCT latent print. The
manually extracted features were used as a ground truth in
assessing the performance of the feature extractor on denoised
and non-denoised images.
A recognised commercial feature extractor namely Digi-
talPersona was used to extract features on denoised and non-
denoised images. There were two sets of data inputs used;
denoised 2D images and non-denoised 2D images for feature
extraction and features were marked as illustrated in Figure 8
- 9. In the images, the blue 0s represent ridge bifurcations and
the lime 0s represent ridge endings. Comparison was made
on features extracted from denoised and non-denoised images
with the manually extracted features and results are illustrated
in Table I.
TABLE I
PERFORMANCE ANALYSIS OF NON-DENOISED AND DENOISED BASED ON
FEATURE EXTRACTION
Non-denoised Denoised
EER 0.087 0.019
FMR 0.031 0.003
Fig. 7. Two 2D latent prints randomly selected
The results of applying the phase preserving denoising
algorithm to the latent print, using k = 2 to set the thresh-
olding along with features marked from the denoised image
is illustrated in Figure 8.
Fig. 8. Features marked from denoised using phase preserving for image
The features marked by the commercial feature extractors
on images with no deniosing applied in Figure 7 (a and b) are
illustrated in Figure 9. It may be noted that there are many
false minutiae detected by feature extractor which may have
an impact in performance of matching algorithm.
Fig. 9. Features marked from enhanced image using commercial feature
extractor
ISBN: 978-1-4673-9187-0 ©2016 IEEE 10

6. To asses and compare the two images, a similarity score [10]
of matching features to the manually extracted is calculated.
Table II gives performance statistics of the two images based
on denoised using phase preserving of OCT image and non-
denoised image of Figure 7. The similarity score is between 0
and 2, where 0 is a non-match and 2 is a perfect match. These
results illustrate that the proposed phase preserving denoising
process improves the quality of the image and hence features
are clearly identifiable by feature extractors. Thus, improving
matching performance.
TABLE II
SIMILARITY SCORES OF IMAGE (A) AND (A)
denoised Non-denoised denoised Non-denoised
(a) (a) (b) (b)
Similarity 1.2 0.79 1.084 0.622
Matched 18 12 11 7
VI. CONCLUSION
A contactless acquisition of OCT is a promising technology
which may be used to identify criminals. This technique of
acquisition is non-destructive and one may perform multiple
acquisition and be able to preserve evidence in the from it
was discovered. In this paper, OCT has been identified to be
such technology and has proven to be able to acquire latent
prints. However, OCT produces 3D volume data which is not
an .ISO standard, therefore requires processing. A technique
to convert 3D volume to 2D latent print by [6] was used.
However, feature extraction is difficult as the reluctant 2D
image is full of noise. In this paper, a denoising algorithm
based on decomposition of using wavelets is proposed. This
technique preserve the phase information which is the crucial
part of the image. The denoising algorithm presented here
served to inform that phase is an important part of information
that should not have noise. The results proved that denoising
OCT latent print using phase preserving does improve feature
extraction as illustrated by experimental results. Thus improve
latent matching which makes it easier for crime investigators
to catch criminals.
ACKNOWLEDGMENT
We would like to express our gratitude to the CSIR National
Laser Centre for allowing us to capture the latent prints using
their OCT machine. Also, we would like to extend our sincere
thanks to Ann Singh for assisting with the capturing process,
without your help this work would not be possible.
REFERENCES
[1] A. K. Jain, L. Hong, S. Pankanti, and R. Bolle, “An identity-
authentication system using fingerprints,” Proceedings of the IEEE,
vol. 85, no. 9, pp. 1365–1388, 1997.
[2] M. Leich, S. Kiltz, J. Dittmann, and C. Vielhauer, “Non-destructive
forensic latent fingerprint acquisition with chromatic white light sen-
sors,” in IS&T/SPIE Electronic Imaging. International Society for
Optics and Photonics, 2011, pp. 78 800S–78 800S.
[3] R. Merkel, S. Gruhn, J. Dittmann, C. Vielhauer, and A. Br¨autigam, “Gen-
eral fusion approaches for the age determination of latent fingerprint
traces: results for 2d and 3d binary pixel feature fusion,” in IS&T/SPIE
Electronic Imaging. International Society for Optics and Photonics,
2012, pp. 82 900Y–82 900Y.
[4] S. Meissner, R. Breithaupt, and E. Koch, “Fingerprint fake detection by
optical coherence tomography,” in SPIE BiOS. International Society
for Optics and Photonics, 2013, pp. 85 713L–85 713L.
[5] L. R. Cambrea and B. G. Harvey, “Fumeless latent fingerprint detection,”
Nov. 5 2013, uS Patent 8,574,658.
[6] R. Khutlang, F. V. Nelwamondo, and A. Singh, “Segmentation of foren-
sic latent fingerprint images lifted contact-less from planar surfaces with
optical coherence tomography,” in Computer Software and Applications
Conference (COMPSAC), 2015 IEEE 39th Annual, vol. 3. IEEE, 2015,
pp. 30–34.
[7] R. D. Labati, A. Genovese, V. Piuri, and F. Scotti, “Two-view contactless
fingerprint acquisition systems: a case study for clay artworks,” in Bio-
metric Measurements and Systems for Security and Medical Applications
(BIOMS), 2012 IEEE Workshop on. IEEE, 2012, pp. 1–8.
[8] P. Kovesi, “Phase preserving denoising of images,” signal, vol. 4, no. 3,
p. 1, 1999.
[9] S. Makinana, J. J. Van Der Merwe, and T. Malumedzha, “A fourier
transform quality measure for iris images,” in Biometrics and Security
Technologies (ISBAST), 2014 International Symposium on. IEEE, 2014,
pp. 51–56.
[10] N. P. Khanyile, A. de Kock, and M. E. Mathekga, “Similarity score
computation for minutiae-based fingerprint recognition,” in Biometrics
(IJCB), 2014 IEEE International Joint Conference on. IEEE, 2014, pp.
1–8.
ISBN: 978-1-4673-9187-0 ©2016 IEEE 11