forensic discrimination of latent fingerprints using laser...
TRANSCRIPT
Article
Forensic Discrimination of LatentFingerprints Using Laser-InducedBreakdown Spectroscopy (LIBS) andChemometric Approaches
Jun-Ho Yang and Jack J Yoh
Abstract
A novel technique is reported for separating overlapping latent fingerprints using chemometric approaches that combine
laser-induced breakdown spectroscopy (LIBS) and multivariate analysis. The LIBS technique provides the capability of real
time analysis and high frequency scanning as well as the data regarding the chemical composition of overlapping latent
fingerprints. These spectra offer valuable information for the classification and reconstruction of overlapping latent fin-
gerprints by implementing appropriate statistical multivariate analysis. The current study employs principal component
analysis and partial least square methods for the classification of latent fingerprints from the LIBS spectra. This technique
was successfully demonstrated through a classification study of four distinct latent fingerprints using classification methods
such as soft independent modeling of class analogy (SIMCA) and partial least squares discriminant analysis (PLS-DA). The
novel method yielded an accuracy of more than 85% and was proven to be sufficiently robust. Furthermore, through
laser scanning analysis at a spatial interval of 125 mm, the overlapping fingerprints were reconstructed as separate two-
dimensional forms.
Keywords
Laser-induced breakdown spectroscopy (LIBS), latent fingerprint, overlapping fingerprint, multivariate analysis
Date received: 3 November 2017; accepted: 20 February 2018
Introduction
Each human fingerprint consists of unique and distinctive
patterns of ridges and valleys. These patterns do not change
over the lifetime of an individual and the probability of fin-
gerprints from two different individuals being identical is
known to be only about one in a million.1 Additional infor-
mation such as eating habits, behavioral habits, and physical
growth can be obtained from the chemical composition of
the oil present on the latent fingerprints.2–6 As a result, an
individual’s fingerprints are widely recognized as a means of
personal authentication, and detection of latent fingerprints
in an actual crime scene can play a very important role.
However, in a real crime scene, obtaining a complete
fingerprint is often a challenging task. The fingerprints
may be present in a damaged or overlapped form and add-
itional efforts are necessary, as two or more overlapping
fingerprints are difficult to distinguish clearly. Various
studies on separating overlapping fingerprints have been
conducted. The majority of mainstream research focuses
on identifying the direction of the fingerprint as a vector
component after extracting images of overlapping finger-
prints, and to proposing an algorithm that separates lines
that do not have vector continuity.7–12 One such method
proposed by Nakamura et al. has been developed to ana-
lyze the laser spectrum of overlapping fingerprints using a
continuous wave green laser.10 Meanwhile, the method
developed by Bradshaw et al. utilizes matrix-assisted laser
desorption ionization mass spectrometry imaging with
multivariate statistical analysis for detecting the endogen-
ous and exogenous species on fingerprints.11 In the study
reported by Tang et al.,12 the molecular signal of the latent
fingerprint was enhanced using gold nanoparticles, and the
two latent fingerprints were classified using mass spectrom-
etry imaging.
Department of Aerospace Engineering, Seoul National University, Republic
of Korea
Corresponding author:
Jack J Yoh, Department of Aerospace Engineering, Seoul National
University, 1 Gwanakro, Gwanakgu, Seoul 151-742, Republic of Korea.
Email: [email protected]
Applied Spectroscopy
0(0) 1–10
! The Author(s) 2018
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0003702818765183
journals.sagepub.com/home/asp
Laser-induced breakdown spectroscopy (LIBS) is a
method of analyzing atomic components of the sample by
creating plasma using a high-power laser beam. When a
high-power laser is irradiated on to a targeted region,
laser ablation occurs, plasma is generated, and a high tem-
perature is generated at this time. In the plasma, the sample
is atomized and ionized after vaporization. The atoms and
ions being investigated attain an excited state due to the
energy absorbed from the plasma and return to the ground
state as they emit energy in electromagnetic form. This
energy is emitted with an intrinsic wavelength depending
on the kind of the element and the excited state. As a
result, LIBS can detect an atomic element whenever
plasma generation is allowed. Additionally, all elements in
the sample can be measured at the same time as the specific
signal of each element is analyzed independently in any state
of solid, liquid, or gas. The LIBS method offers several
advantages over other laser spectroscopic techniques
such as simplicity of setup in combination with other
optics, ability to perform real time, remote analysis, and
it does not need chemical pretreatment of the sample.
Consequently, LIBS has proven quite useful in various
fields that include flame diagnostics, space exploration,
and forensic science, in particular.13–16
The study by Taschuk et al. detected the sodium content
of the ridge portion of the latent fingerprint using 120 fs
laser pulse after engraving the latent fingerprint on a Si
wafer.17 This study was the first to detect the shape of
the latent fingerprint using LIBS and confirmed that the
shape of the fingerprint can be derived for an area of
5 mm width and 1 mm height as a two-dimensional (2D)
image. In the study by Godwal et al.,18 a 5 lJ ultraviolet
laser pulse was used and a portable LIBS device was fabri-
cated to produce realistic latent fingerprint detection sys-
tems. This study demonstrated that with a dual-pulse
technique the sensitivity of micro-LIBS can be increased
by magnitude of ablation pulse power. Other fingerprint
research using LIBS did not detect latent fingerprint but
detected explosives on the fingerprint and restored them
on the fingerprint as a 2D image. In the studies by
Abdelhamid et al.,19,20 optical catapulting was utilized in
combination with LIBS to analyze solid aerosol, explosive,
and confusant materials such as trinitrotoluene (TNT), dini-
trotoluene (DNT), and sodium maleonitriledithiolate
(MNT) were classified. Despite being widely utilized for
fingerprint detection, LIBS has so far been used only for
detection of a single fingerprint.17–20
Recently, the multivariate chemometrics method has
been combined with LIBS data and used in many diverse
fields. The multivariate chemometrics has been used to
identify soil constituents and classify the soil according to
composition. Various analytical and multivariate methods
such as partial least squares regression (PLSR), soft inde-
pendent modeling of class analogy (SIMCA), partial least
square discriminant analysis (PLS-DA), linear discrimination
analysis, and principal component regression been used to
analyze the soil elemental components.21–24 The multivari-
ate analysis combined with LIBS was also utilized in arche-
ology to classify coins with archeological significance
through principal component analysis (PCA) and conse-
quently to estimate the age and location of use for the
coin.25,26 Oztoprak et al. transformed medieval ceramics
into LIBS laser spectral data to classify archeological cer-
amic samples, glaze, paint, and clay via PCA.27 Dingari et al.
conducted a study on the classification and estimation of
chemical components of unknown substances through
multivariate analysis for previously specified chemical com-
ponents,28 even when LIBS spectroscopy was used for
unknown substances. In the study by Vance et al.,29 it was
also possible to classify protein components into 2D images
using the nonlinear classification method known as support
vector machine (SVM) together with LIBS data. Neiva et al.
used a linear classification method such as SIMCA and
PLS-DA analysis as an index to judge and classify the quality
of leather of sheep and cattle.30 In the study by Zhang
et al.,31 qualitative and quantitative analysis, and classifica-
tion of, the chemical composition of 20 slags was per-
formed by multivariate analysis such as SVM and PLS
combined with LIBS. In addition, there have been studies
to confirm the high degree of accuracy of classification of
explosives by analyzing the suspended matter containing
explosives and classifying them according to the situation.32
Highly reliable analysis as well as classification of pigments
and inks has been conducted using Raman spectroscopy,
which can detect molecular components differently from
LIBS, combined with the LIBS analysis method.33
The current study is focused on a methodology for sep-
aration of two overlapping latent fingerprints using a novel
technique which combines LIBS with multivariate analysis.
This research assumes the chemical composition of the
latent fingerprints to be different and extracts the latent
fingerprints from the four different individuals. The study
also includes a linear and nonlinear multivariate analysis
using PCA, SIMCA, and PLS-DA. In order to validate the
LIBS results, the classification was confirmed by conducting
a sensitivity test as well as a robustness test. The new tech-
nique was found to be suitable for identifying two different
overlapping fingerprints in a real crime scene using LIBS
combined with multivariate analysis and for classifying
them into two distinct fingerprint images.
Material and Methods
Preparation of Latent Fingerprint Sample
All of the latent fingerprints were extracted from four
Korean and Indian males in their 20 s and 30 s, and were
prepared spontaneously. This experiment is based on the
assumption that there are differences in the chemical com-
ponents existing in the latent fingerprint, and that the
2 Applied Spectroscopy 0(0)
classification and detection based on the difference in
chemical components can be performed through multivari-
ate analysis and utilized as a 2D chemical distribution.
Therefore, the same procedure was used to collect four
different latent fingerprints. Because the fingers and fore-
head oil can also be easily affected by external factors such
as humidity and contaminations, the task of cleaning using
the ethanol with the tissues was prioritized to index finger
and forehead. After 5 min, the index finger was rubbed for
10 s from the forehead of the fingerprint extractor, and
then pressed on the aluminum plate with a slight force
for 10 s to leave the fingerprint on the aluminum plate. A
fingerprint sample was prepared in the same order as in
Figure 1a. The strong fluorescent light is illuminated in the
case of Figure 1 to visualize the clear fingerprint. In the case
of overlapping fingerprint samples, the other two finger-
prints were individually extracted, and after washing with
ethanol like in the previous single fingerprint extraction,
two fingerprints were successively placed on the aluminum.
When fingerprinting two people, samples were prepared so
that overlapping parts existed. In preparation of the alumi-
num plate, vinyl was attached to prevent contact with the
outside before leaving the fingerprint, and the vinyl was
removed before the fingerprint was imprinted.
Experimental Setup
A Q switched neodymium doped yttrium aluminum garnet
(Nd:YAG) laser with a 1064 nm wavelength was used as a
LIBS experimental device (RT250-Ec, Applied Spectra Inc.)
for inducing laser ablation and generating plasma. The laser
has a pulse duration of 5 ns. In this study, we aim at mini-
mizing the level of destruction of the fingerprint in all
experiments through laser ablation, enhancing the reso-
lution at the time of image processing. Therefore, laser
pulse energy was maintained at approximately 10 mJ.
When the laser is irradiated, the gate delay is adjustable
from 0.1 to 0.5 ls, and the gate width is set to 1.05 ms. In
the case of the spectrometer, a six-channel charge-coupled
device spectrometer (Applied Spectra Inc.) is used to
obtain a high resolution (0.1 nm from ultraviolet to visible
range, 0.12 nm for visible to near-infrared range), and a
laser spectrum can be detected by dividing the wavelength
in the range 198–1050 nm into about 13 000 wavelength
channels. A 100 mm uncoated quartz lens was used as a
focal lens to allow laser light to generate plasma through
focusing laser light.
For minimizing laser crater, beam expander was used to
decrease beam diameter. First of all, entrance aperture of
12.7 mm focuses the wide beam from Nd:YAG laser. After
that, using 8.6 mm diameter lens, laser beam was expanded
to straight laser beam. To obtain regular and reasonable
LIBS data, the distance between the fingerprint sample
and the laser was kept constant at 15 cm through
the stage which can move in the xyz axis inside the
LIBS device. The experimental setup is the same as in
Figure 1b. Laser spot size and laser irradiance are 125 lm
and 6.366� 1013 W m�2 s�1.
Multivariate Chemometrics Methods
The principal component analysis (PCA) technique can
transfer high-dimensional data such as laser spectra to
low-dimensional data. When the data is arranged on PCA
plane, the axis having the greatest dispersion is set as the
first principal component (PC) to analyze and divide each
data, and the allocated PCs are sequentially set according to
the magnitude of dispersion about laser spectrum data.
Through PCA analysis, grouping based on the area of PCs
score plot becomes the basis of additional SIMCA and PLS-
DA methods.
Soft independent modeling of class analogy is used to
identify local models for possible groups through PCA ana-
lysis and to predict possible class memberships for new
Figure 1. (a) Sequence of preparation of latent fingerprint. (b) LIBS experimental setup.
Yang and Yoh 3
observations. Firstly, the global PCA is performed accord-
ing to the available data structure for the entire data set to
identify the observation group. Then, regional models are
evaluated for each model and classified as one of the exist-
ing class models. The number of major components
retained for each class is generally freely configurable, but
the signal or information contained in the model for the
class may be distorted if it has too few components, and if
there are too many PCs, since the signal is reduced, it is
important to determine the number of PCs that must be
retained for each class.
Partial least squares discriminant analysis can derive the
maximum separation between classes where each latent
fingerprint laser spectrum data is divided using PLS compo-
nents to sharpen the separation between observation
groups. Partial least squares discriminant analysis consists
of a PLSR analysis where the response variable is a categor-
ical variable representing the hierarchical membership of
the statistical unit. Therefore, PLS-DA does not allow any
other response variable that is a response variable to define
an individual group, and consequently all measurement vari-
ables play the same role in relation to class assignment. In
practice, the PLS component is constructed by attempting
to find the appropriate trade-off between the two object-
ives of explaining the set of explanatory variables and pre-
dicting the response variable. Classification based on PLS
has an advantage in constructing prediction capability, and
this method can be evaluated as a more advanced method
than the classical SIMCA classification analysis method,
which works more for unit reallocation of predefined
classes.
In the current study, sensitivity (cross-validation) and
robustness (validation) tests were performed for the
SIMCA and PLS-DA methods.28 The sensitivity test was
conducted by randomly splitting 100 sample laser spectra
into two groups, viz. training group of 30 spectra and test
data of 70 spectra. After that, the 12 test laser spectra are
comprised of three randomly chosen ones for each latent
fingerprint. Robustness test was conducted in a similar
sequence as the sensitivity test with a key difference of
removal of one latent fingerprint class from the calibration
set and repetition of the sequence for each latent
fingerprint class. In order to improve the accuracy of the
analysis, 100 independent iterations are performed by re-
splitting the whole data into training and test data for both
the sensitivity and robustness tests. Current study assigns
the term ‘‘correct classification’’ for the instances when the
data of each latent fingerprint is correctly assigned, ‘‘mis-
classification’’ for the cases with incorrect assignment from
the original latent fingerprint, and ‘‘unclassification’’ for the
cases where no class is assigned.
The Unscrambler X 10.1 and Matlab R2016a software
were used for data analysis and statistical multivariate ana-
lysis. In addition, Origin Pro 8.5.1 (Origin Lab Corporation)
was used as the graphing software.
Results and Discussion
Laser Spectrum Analysis and PCA of LatentFingerprints
The LIBS spectrum was extracted from four different per-
sons. To improve reliability of the laser spectrum, in each
human’s laser spectrum, the mean value of 100 laser spec-
tral data was used. In Figure 2, the y-axis value is offset by
35 000. It is confirmed that the signal of the latent finger-
print has a relative high and low signal intensity at several
specific wavelengths.
The wavelengths with relatively high signal intensities in
the oil fingerprint laser spectrum are approximately
238 nm, 239 nm, 274 nm, 308 nm, 309 nm, 393 nm,
394 nm, 396 nm, 588 nm, 589 nm, 747 nm, 766 nm, and
777 nm. As a result, the signals for Fe, Al, O, K, Na, and
Ca were confirmed, as shown in Figure 2. The composition
of this signal was analyzed and checked using the NIST
atomic spectra database.34
There are differences in body chemical composition due
to various factors such as eating habits, environment, and
inherited family history, and this is also checked from the oil
presented on human skin surface. Therefore, when the
chemical component of latent fingerprint is measured
using LIBS, there is a difference in the laser spectrum ten-
dency depending on the difference of the chemical compo-
nents. However, these results are only a result of visual
confirmation through naked eyes, and PCA is required to
separate or classify the latent fingerprints as reliable data.
The PCA analyzes the overall LIBS spectral region, extracts
a specific portion having a main emission line having a rela-
tively high SNR (signal to noise ratio) value capable of ana-
lyzing the PCs, and extracts a specific value based on the
extracted PC spectrum LIBS spectral results are inter-
preted. In this research, PCA was performed on 100
laser spectra data for each latent fingerprint. Figure 3
shows the results obtained from the PCA and PLS about
four different latent fingerprints. It can be seen that the
location of each latent fingerprints shows the independent
position on the 2D plane. Also, as shown in Figure 3, it can
be seen that about 90% of the total spectral range is also
explained when using only PC 1 and PC 2 (PC 1: 53.80%,
PC 2: 35.90%). In case of PLS analysis, 95% of the total
spectral range is explained when using only factor 1 and 2
(factor 1: 86%, factor: 7%).
As shown in Figure 3(b), the emission line wavenumber
classifications in PCA are Mg, Fe, Na, Ca, and K. PCs 1–7
show almost the same wavelength; however, the intensities
at some wavelengths are reversed by each PC, respectively.
It can be seen that this makes a difference.
Multivariate Analysis
For the first SIMCA analysis, 100 laser spectra were used
for each latent fingerprint. In the case of the sensitivity test,
4 Applied Spectroscopy 0(0)
Figure 2. LIBS spectra lines of four types of latent fingerprints.
Figure 3. (a) Separation via the principal component analysis, (b) the results of extracted PCs, (c) separation via the partial least square
(PLS) method, and (d) the PLS loading plot.
Yang and Yoh 5
randomly selected 30 fingerprints from the 100 fingerprint
laser spectra obtained at one time were set as a test
sample, and the remaining 70 fingerprint data were classi-
fied through PCA. The results are shown in Table 1, when
SIMCA analysis is performed on 30 test samples per
fingerprint. In this case, it was confirmed that the class
was correctly assigned except for the four data. For the
robustness test, one latent fingerprint class is deleted from
calibration set and repeated for each latent fingerprint class.
For the data analyzed by PCA, data such as PCA analysis
data were used for the sensitivity test. Figure 4 shows the
model distance between PCA models. This value means
how different two models are from each other. A larger
model to model distance express that two model can be
discriminated more clearly.
In order to obtain statistical reliability, not only SIMCA
analysis results were obtained for the whole data, but also
laser spectra were extracted randomly for every data of
each of the latent fingerprints and repeated 100 iterations.
This research conducted sensitivity and robustness test
about SIMCA. The difference between sensitivity and
robustness is whether the test sample comes from the
first 100 fingerprint data or the additional extracted spec-
tral data. The results are shown in Table 1. In the case of
the sensitivity test, it was confirmed that the average
classification accuracy was 90.36%, 6.18% were inadequate,
Table 1. The classification probability of SIMCA.
Type
Correct
classification Misclassification Unclassification
(a) Sensitivity test
Latent fingerprint 1 0.9010 0.0320 0.0670
Latent fingerprint 2 0.9327 0.0333 0.0300
Latent fingerprint 3 0.8750 0.1000 0.0250
Latent fingerprint 4 0.9060 0.0820 0.0120
(b) Robustness test
Latent fingerprint 1 0.8440 0.1140 0.0420
Latent fingerprint 2 0.8427 0.1313 0.0260
Latent fingerprint 3 0.8227 0.1433 0.0340
Latent fingerprint 4 0.8540 0.1320 0.0140
Figure 4. Model distance using SIMCA.
6 Applied Spectroscopy 0(0)
and 3.46% were not classified. In the robustness test, it
was found that the average of the test was 84.10%,
13.01%, and 2.89%, respectively. However, since robustness
experiments have a probability of 82% or more, it is pos-
sible to classify them using SIMCA in the case of latent
fingerprints of different origins, and it can be separated
based on the chemical composition of overlapping latent
fingerprints.
For the PLS-DA analysis, the same laser spectra were
used for the four latent fingerprints subjected to the
SIMCA analysis, after that, PLS regression was performed
for each type of fingerprint to classify each latent finger-
print. Although the PLS-DA analysis is basically stable and
robust compared with the SIMCA analysis, the actual
Table 2. The classification probability of PLS-DA.
Type
Correct
classification Misclassification Unclassification
Sensitivity test
Latent fingerprint 1 1.0000 0.000 0.0000
Latent fingerprint 2 0.8667 0.1230 0.0000
Latent fingerprint 3 0.9667 0.0210 0.1200
Latent fingerprint 4 0.9000 0.0920 0.0080
Robustness test
Latent fingerprint 1 0.9333 0.0333 0.0333
Latent fingerprint 2 0.7980 0.1667 0.0353
Latent fingerprint 3 0.8667 0.0987 0.0346
Latent fingerprint 4 0.8333 0.0667 0.1000
Figure 5. (a) A process for the separation of overlapping fingerprints. (b, c) Schematic figures showing separation of overlapping
fingerprints.
Yang and Yoh 7
results (Table 2) are not significantly different from the
SIMCA analysis. For the sensitivity test, the correct classi-
fication was performed with an average probability of
93.33%, misclassification was 5.9%, and unclassification
was 3.2%. This leads to correct classification with relatively
high probability compared to SIMCA analysis, while mis-
and unclassification also have a small probability of drop.
In the robustness experiment, correct, mis-, and unclassifi-
cation were obtained with 85.78%, 9.13%, and 5.07% prob-
ability, respectively. The probability of correctly assigned
PLS-DA method was higher than that of SIMCA method,
and misclassification was also less probable. However,
unclassification has been confirmed to have a higher prob-
ability of PLS-DA scheme.
Separation of Overlapping Latent Fingerprints
It is confirmed that the classification is conducted through
multivariate analysis methods such as PCA, SIMCA, and
PLS-DA when the sources of the latent fingerprints are
different. In this section, we present result of separating
overlapping fingerprints in terms of 2D chemical distribu-
tions by comparing the laser spectrum tendencies, and then
the multivariate analysis method is applied to reconstruct
each fingerprint’s line. The fingerprint classification process
is shown in Figure 5. First of all, we extract 30 laser spectra
of each latent fingerprint. After that, the predicting class is
created through PCA analysis. Thereafter, the laser is irra-
diated at regular intervals, and the laser spectra of each
Figure 6. Separation of overlapping latent fingerprints using classification methods.
8 Applied Spectroscopy 0(0)
point is classified and analyzed to finally separate the over-
lapping fingerprints like Figure 6. As a result, it was con-
firmed that even in the case of actual overlapping
fingerprints, multivariate analysis can be used to classify
and separate each latent fingerprints.
Conclusion
The chemometric approach of combining LIBS with multi-
variate analysis for separating overlapping latent fingerprints
has been validated. The LIBS data was obtained using a
Q-switched Nd:YAG 1064 nm wavelength laser with the
laser energy, pulse duration, and the beam width optimized
for highly reliable spectra which were collected in the ultra-
violet to visible range. The PCA and PLS methods were
utilized to extract the independent regions corresponding
to each of the two overlapping fingerprints. Furthermore,
sensitivity testing of the new methodology established
probability of accurate detection of the fingerprint to be
higher than 90%. Similarly, the robustness of the method
against perturbations was recognized as the method yielded
an accuracy of approximately 85%. Finally, the separations
of overlapping fingerprints were performed using LIBS scan-
ning technique and multivariate analysis. Based on the ease
of application and accuracy of the results, the novel tech-
nique was confirmed to be useful in separating overlapping
fingerprints in real life scenarios.
Conflict of Interest
None declared.
Funding
This work was supported by the Korea National Research
Foundation (grant numbers NRF-2014M1A3A3A02034903 and
NRF-2016R1D1A1A02937421) through IAAT at Seoul National
University.
References
1. E.E. Hueske. Firearms and Fingerprints. New York: Infobase Publishing,
2008.
2. N.E. Archer, Y. Charles, J.A. Elliott, S. Jickellss. ‘‘Changes in the Lipid
Composition of Latent Fingerprint Residue with Time after Deposition
on a Surface’’. Forensic Sci. Int. 2005. 154(2): 224–239.
3. R.S. Croxton, M.G. Baron, D. Butler, T. Kent, V.G. Searss. ‘‘Variation in
Amino Acid and Lipid Composition of Latent Fingerprints’’. Forensic
Sci. Int. 2010. 199(1): 93–102.
4. J.S. Day, H.G. Edwards, S.A. Dobrowski, A.M. Voices. ‘‘The Detection of
Drugs of Abuse in Fingerprints Using Raman Spectroscopy I: Latent
Fingerprints’’. Spectrochim. Acta, Part B. 2004. 60(3): 563–568.
5. K.M. Antoine, S. Mortazavi, A.D. Miller, L.M. Millers. ‘‘Chemical
Differences are Observed in Children’s Versus Adults’ Latent
Fingerprints as a Function of Time’’. J. Forensic Sci. 2010. 55(2):
513–518.
6. C. Weyermann, C. Roux, C. Champods. ‘‘Initial Results on the
Composition of Fingerprints and its Evolution as a Function of Time
by GC/MS Analysis’’. J. Forensic Sci. 2011. 56(1): 102–108.
7. Q. Zhao, A.K. Jains. ‘‘Model Based Separation of Overlapping Latent
Fingerprints’’. IEEE Trans. Inf. Forensics Secur. 2012. 7(3): 904–918.
8. M. Schott, R. Merkel, J. Dittmann, ‘‘Sequence Detection of
Overlapping Latent Fingerprints Using a Short-Term Aging Feature’’.
In Proceedings of the IEEE International Workshop on Information
Forensics and Security (WIFS). Tenerife, Spain: December 2–5,
2012. Pp. 85–90.
9. J. Feng, Y. Shi, J. Zhous. ‘‘Robust and Efficient Algorithms for
Separating Latent Overlapped Fingerprints’’. IEEE Trans. Inf.
Forensics Secur. 2012. 7(5): 1498–1510.
10. A. Nakamura, H. Okuda, T. Nagaoka, N. Akiba, et al. ‘‘Portable
Hyperspectral Imager with Continuous Wave Green Laser for
Identification and Detection of Untreated Latent Fingerprints on
Walls’’. Forensic Sci. Int. 2015. 254: 100–105.
11. R. Bradshaw, W. Rao, R. Wolstenholme, M. Clench, et al. ‘‘Separation
of Overlapping Fingermarks by Matrix Assisted Laser Desorption
Ionisation Mass Spectrometry Imaging’’. Forensic Sci. Int. 2012.
222(1): 318–326.
12. H.-W. Tang, W. Lu, C.-M. Che, K.-M. Ngs. ‘‘Gold Nanoparticles and
Imaging Mass Spectrometry: Double Imaging of Latent Fingerprints’’.
Anal. Chem. 2010. 82(5): 1589–1593.
13. S.H. Lee, H. Do, J.J. Yohs. ‘‘Simultaneous Optical Ignition and
Spectroscopy of a Two-Phase Spray Flame’’. Combust. Flame. 2016.
165: 334–345.
14. J.-J. Choi, S.-J. Choi, J.J. Yoh. ‘‘Standoff Detection of Geological
Samples of Metal, Rock, and Soil at Low Pressures Using Laser-
Induced Breakdown Spectroscopy’’. Appl. Spectrosc. 2016. 70(9):
1411–1419.
15. K.-J. Lee, S.-J. Choi, J.J. Yoh. ‘‘Stand-Off Laser-Induced Breakdown
Spectroscopy of Aluminum and Geochemical Reference Materials at
Pressure Below 1 Torr’’. Spectrochim. Acta, Part B. 2014. 101:
335–341.
16. J.-H. Yang, S.-J. Choi, J.J. Yoh. ‘‘Towards Reconstruction of Overlapping
Fingerprints Using Plasma Spectroscopy’’. Spectrochim. Acta, Part B.
2017. 134: 25–32.
17. M. Taschuk, Y. Tsui, R. Fedosejevss. ‘‘Detection and Mapping of Latent
Fingerprints by Laser-Induced Breakdown Spectroscopy’’. Appl.
Spectrosc. 2006. 60(11): 1322–1327.
18. Y. Godwal, M. Taschuk, S. Lui, Y. Tsui, R. Fedosejevss. ‘‘Development
of Laser-Induced Breakdown Spectroscopy for Microanalysis
Applications’’. Laser Part. Beams. 2008. 26(1): 95–104.
19. M. Abdelhamid, F. Fortes, M. Harith, J. Lasernas. ‘‘Analysis of Explosive
Residues in Human Fingerprints Using Optical Catapulting–Laser-
Induced Breakdown Spectroscopy’’. J. Anal. At. Spectrom. 2011.
26(7): 1445–1450.
20. M. Abdelhamid, F. Fortes, J. Laserna, M. Harith. ‘‘Optical Catapulting
Laser Induced Breakdown Spectroscopy (OC-LIBS) and Conventional
LIBS: A Comparative Study’’. AIP Conf. Proc. 2011. 1380(1): 55–59.
21. J. Feng, Z. Wang, L. West, Z. Li, W. Nis. ‘‘A PLS Model Based
on Dominant Factor for Coal Analysis Using Laser-Induced
Breakdown Spectroscopy’’. Anal. Bioanal. Chem. 2011. 400(10):
3261–3271.
22. A. dos Santos Augusto, E.F. Batista, E.R. Pereira-Filhos. ‘‘Direct
Chemical Inspection of Eye Shadow and Lipstick Solid Samples Using
Laser-Induced Breakdown Spectroscopy (LIBS) and Chemometrics:
Proposition of Classification Models’’. Anal. Methods. 2016. 8(29):
5851–5860.
23. M.J.C. Pontes, J. Cortez, R.K.H. Galvao, C. Pasquini, et al.
‘‘Classification of Brazilian Soils by Using LIBS and Variable Selection
in the Wavelet Domain’’. Anal. Chim. Acta. 2009. 642(1): 12–18.
24. T. Zhang, L. Liang, K. Wang, H. Tang, et al. ‘‘A Novel Approach for the
Quantitative Analysis of Multiple Elements in Steel Based on Laser-
Induced Breakdown Spectroscopy (LIBS) and Random Forest
Regression (RFR)’’. J. Anal. At. Spectrom. 2014. 29(12): 2323–2329.
25. S. Awasthi, R. Kumar, G. Rai, A. Rais. ‘‘Study of Archaeological Coins
of Different Dynasties Using LIBS Coupled with Multivariate Analysis’’.
Opt. Lasers Eng. 2016. 79: 29–38.
Yang and Yoh 9
26. S. Moncayo, M. Kocianova, J. Hulık, J. Plavcan, et al. ‘‘Discrimination of
Copper Alloys with Archaeological Interest Using LIBS and
Chemometric Methods’’. In: Processing of the 2014 World Data
System (WDS) Contributed Paper – Physics, Prague, Czech
Republic, June 3-5, 2014, pp. 131–135.
27. B.G. Oztoprak, M. Sinmaz, F. Tuleks. ‘‘Composition Analysis of
Medieval Ceramics by Laser-Induced Breakdown Spectroscopy
(LIBS)’’. Appl. Phys. A. 2016. 5(122): 1–11.
28. N.C. Dingari, I. Barman, A.K. Myakalwar, S.P. Tewari, et al.
‘‘Incorporation of Support Vector Machines in the LIBS Toolbox for
Sensitive and Robust Classification Amidst Unexpected Sample and
System Variability’’. Anal. Chem. 2012. 84(6): 2686–2694.
29. T. Vance, N. Reljin, A. Lazarevic, D. Pokrajac, et al. ‘‘Classification of
LIBS Protein Spectra Using Support Vector Machines and Adaptive
Local Hyperplanes’’. In: Proceedings of the 2010 International Joint
Conference on Neural Networks (IJCNN). Barcelona, Spain: July
18–23, 2010. Pp. 1–7.
30. A.M. Neiva, M.A.C. Jacinto, M.M. de Alencar, S.N. Esteves,
E.R. Pereira-Filhos. ‘‘Proposition of Classification Models for the
Direct Evaluation of the Quality of Cattle and Sheep Leathers Using
Laser-Induced Breakdown Spectroscopy (LIBS) Analysis’’. RSC Adv.
2016. 6(106): 104827–104838.
31. T. Zhang, S. Wu, J. Dong, J. Wei, et al. ‘‘Quantitative and Classification
Analysis of Slag Samples by Laser Induced Breakdown Spectroscopy
(LIBS) Coupled with Support Vector Machine (SVM) and Partial Least
Square (PLS) Methods’’. J. Anal. At. Spectrom. 2015. 30(2): 368–374.
32. F.C. De Lucia Jr, J.L. Gottfried, C.A. Munson, A.W. Mizioleks.
‘‘Multivariate Analysis of Standoff Laser-Induced Breakdown
Spectroscopy Spectra for Classification of Explosive-Containing
Residues’’. Appl. Opt. 2008. 47(31): 112–121.
33. M. Hoehse, A. Paul, I. Gornushkin, U. Pannes. ‘‘Multivariate
Classification of Pigments and Inks Using Combined Raman
Spectroscopy and LIBS’’. Appl. Opt. 2012. 402(4): 1443–1450.
34. P.L. Smith, C. Heise, J.R. Esmond, R.L. Kuruczs. ‘‘Atomic Spectral Line
Database’’. http://cfa-www.harvard.edu/amp/data/kur23/sekur.html
[accessed Mar 8 2016].
10 Applied Spectroscopy 0(0)