The present invention relates to the system and method for lie detection.
More specifically, the invention has directed a system and method for lie detection using
EEG data.
Background
[0002] The background description includes information that may be useful in
understanding the present invention. It is not an admission that any of the information
provided herein is prior art or relevant to the presently claimed invention, or that any
publication specifically or implicitly referenced is prior art.
[0003] Lie detection is an assessment of a verbal statement with the goal to reveal
possible intentional deceit. Lie detection may refer to a cognitive process of detecting
deception by evaluating message content as well as non-verbal cues. Identification of a
statement as truth or lie is a major problem. It has various applications for safety and crime
control. Traditionally physiological activities are monitored during the question-answer round
and compare to a normal level. However, because the subject can control his/her
physiological reactions, therefore, to overcome these brain signals are used to identify the
truth.
[0004] Lie is a ubiquitous social and psychological phenomenon in human society.
The lie becomes a factor influencing the stable group of society and poses a serious threat to
the property and life safety of the people. Psychologists and other related experts have
3
therefore struggled to find effective lie detection methods. The effectiveness of lie detection
has been proven in long-term and extensive application practice at home and abroad. The lie
detection technology has important application value for the detection of criminal detection
cases. In addition, lie recognition is also of great significance for the treatment of
psychological diseases and disorders. In addition, the current international anti-terrorism
struggle situation is still severe, and India also urgently needs to establish an effective antiterrorism means.
[0005] The traditional test method is called a multi-channel physiological signal
tester technology. Although the technology has a certain scientific basis, the test process is
easily affected by factors such as tension, fear, and the like because various collected signals
are based on a peripheral autonomic nervous system, and the accuracy of the test method is
greatly limited. Another important reason is that traditional lie detection techniques have a
potential risk of anti-lie detection since multiple physiological indicators of lie detection can
be changed by conscious cognitive methods and physical control.
[0006] Brain signal is the first to respond to any sensory impulses, which can be
used to identify the person is telling the truth or lying. The most common and long-used
measure is the polygraph. The EEG signals describe the brain signal activity of a person.
There are a wide variety of technologies available for this purpose. In addition to a
polygraph, various physiological measurements are used to identify a lying individual among
a group of individuals. Some of the measures are heart rate, respiratory movement, blood
pressure, galvanic skin response (GSR), electroencephalography (EEG), and functional
magnetic resonance imaging (FMRI). EEG-based systems are being developed by many
companies due to their potential for detecting lies with better accuracy.
4
[0007] Modern lie detection technology mainly utilizes analysis of neural signals of
people facing criminal activities or false information to perform lie detection, for example,
Event-Related Potential (ERP) of electroencephalogram signals is utilized to reflect a
processing process of brain cognition, and Event-related potential is utilized to perform lie
detection analysis. Compared with the traditional lie detection method, the modern lie
detection technology focuses on the information coding and extraction characteristics of the
brain in psychological phenomena such as perception, memory, thinking, imagination, and
the like from the aspect of the research content.
[0008] Research by a plurality of researchers proves that ERP components of EEG
signals comprise an automatic processing process that is difficult to inhibit, so that compared
with the traditional lie detection technology, the modern lie detection technology can
effectively resist the lie detection behavior.
[0009] The current lie detection technology based on ERP has many defects, such
as large stimulation amount, low accuracy, large data superposition, long experiment time,
easy fatigue of a tested person, and meanwhile, along with the increasing times of testing
information, the sensitivity of a criminal or the tested person to false information is greatly
reduced so that the testing effect is influenced.
[00010] Therefore, there is a need for an efficient lie detection system and method
using EEG data with better accuracy and has the potential of providing the faster result.
[00011] All publications herein are incorporated by reference to the same extent as if
each individual publication or patent application were specifically and individually indicated
to be incorporated by reference. Where a definition or use of a term in an incorporated
reference is inconsistent or contrary to the definition of that term provided herein, the
5
definition of that term provided herein applies and the definition of that term in the reference
does not apply.
Objects of the Invention
[00012] An object of the present disclosure is to overcome one or more drawbacks
associated with the conventional mechanisms.
[00013] An object of the present invention is to provide a system and method for lie
detection.
[00014] An object of the present invention is to provide a system and method for lie
detection using EEG data.
[00015] Another object of the invention is to provide a system and method for
extracting relevant information from EEG and using the extracted information for lie
detection.
[00016] Another object of the invention is to provide a system and method of lie
detection that with better accuracy and in a shorter time.
Summary
[00017] The present invention relates to the system and method for lie detection.
More specifically, the invention has directed a system and method for lie detection using
EEG data.
[00018] In an aspect, the present invention provides a method for lie detection, the
method comprising: causing one or more processors which are coupled to a non-transitory
storage device and operable to perform the steps of: receiving, at a computing device
6
(500), from an EEG machine (102), EEG signals comprising of Event-Related Potential
(ERP) signal collected from at least Fz, Cz, and Pz electrode locations on a human scalp;
extracting, at the computing device, a set of features comprising of time-domain features,
frequency domain features, wavelets, nonparametric weighted features, and empirical mode
decomposition (EMD)-based features from the EEG signals; evaluating, at the computing,
correlation coefficients among the Fz, Cz, and Pz electrode sites to examine the similarity of
probe responses; analyzing, at the computing device, the set of features and the correlation
coefficients to detect lie by a machine learning model using the set of features and the
correlation coefficients.
[00019] In an embodiment, the method includes obtaining features for intrinsic mode
function (IMF) from EMD based features; correlating between ERP components of the EEG
signal to determine if whether the components are providing the same or distinct neural
information; and analyzing the set of features based on the result of the determination.
[00020] In an embodiment, the machine learning model is a supervised machine
learning model.
[00021] In an embodiment, the ERP signal comprises a P300 component that is used
as a primary indicator by the machine learning model.
[00022] In an embodiment, the RBF kernel is utilized with SVM.
[00023] In an aspect, the present invention provides a system for lie detection
system, the system comprising: an EEG data receiving module 204 configured, at a
computing device (500), to receive EEG signals comprising of Event-Related Potential (ERP)
signal collected from at least Fz, Cz, and Pz electrode locations on a human scalp; a feature
extraction module 206 configured, at the computing device, to extract a set of features
7
comprising of time-domain features, frequency domain features, wavelets, nonparametric
weighted features, and empirical mode decomposition (EMD)-based features from the EEG
signals; evaluating, at the computing device, correlation coefficients among the Fz, Cz, and
Pz electrode sites to examine the similarity of probe responses; and an SVM classification
and training module 208 configured, at the computing device, to analyze the set of features
and the correlation coefficients to detect lie using a machine learning model that uses the set
of features and the correlation coefficients for lie detection.
[00024] Various objects, features, aspects and advantages of the inventive subject
matter will become more apparent from the following detailed description of preferred
embodiments, along with the accompanying drawing figures in which like numerals represent
like components.
Brief Description of the Drawings
[00025] The detailed description is set forth with reference to the accompanying
figures. In the figures, the left-most digit(s) of a reference number identifies the figure in
which the reference number first appears. The use of the same reference numbers in different
figures indicates similar or identical items.
[00026] FIG. 1 illustrates an exemplary networked environment of Lie detection
system in accordance with an embodiment of the present disclosure.
[00027] FIG. 2 illustrates an exemplary module diagram of a lie detection system in
accordance with an embodiment of the present invention.
8
[00028] FIG. 3 illustrates an example EEG probe waveform at Pz location of a guilty
subject and an innocent subject in accordance with an embodiment of the present invention.
[00029] FIG. 4 illustrates an example training EEG probe signals used for training
SVM classifier used for lie detection in accordance with an embodiment of the present
disclosure.
[00030] FIG. 5 illustrates an exemplary flow chart of a method for lie-detecting using
EEG signals in accordance with an embodiment of the present disclosure.
[00031] FIG. 6 illustrates an example computer system used for the implementation
feature of the present invention.
Detailed Description
[00032] The following discussion provides many example embodiments of the
inventive subject matter. Although each embodiment represents a single combination of
inventive elements, the inventive subject matter is considered to include all possible
combinations of the disclosed elements. Thus, if one embodiment comprises elements A, B,
and C, and a second embodiment comprises elements B and D, then the inventive subject
matter is also considered to include other remaining combinations of A, B, C, or D, even if
not explicitly disclosed.
[00033] The present invention relates to the system and method for lie detection.
More specifically, the invention has directed a system and method for lie detection using
EEG data.
9
[00034] In some embodiments, a lie detection system based on analysis and
classification of EEG signals is described. EEG is the measurement of the electrical signals
generated by the firing of neurons in the brain. Electrical potential, known as event-related
potential (ERP), is generated in the human brain in response to external occurrences in the
environment or decisions made in the brain. It has been observed that the P300 component of
ERP is evoked in the brain in response to the rare and meaningful information presented to
the human. The system receives the EEG signals, which include the P300 component of ERP
that is collected by measuring Fz, Cz, and Pz electrode locations on the human scalp. The
P300 component is extensively investigated to identify if an individual is hiding the
information. Different time-domain features, frequency domain features, wavelets,
nonparametric weighted features and empirical mode decomposition (EMD)-based features
have been used to study various aspects of the EEG signals.
[00035] The system use classification techniques, such as linear discriminant
analysis (LDA), support vector machine (SVM), and artificial neural network (ANN), to
effectively separate the output classes in the neural data. The Fourier transform (FT) is a
commonly used technique for the frequency domain analysis of EEG signals. However, it has
been observed that FT does not provide information about the precise time of occurrence of
the different neural phenomena. In an embodiment, the system may use the short-time
Fourier transform (STFT), which is a sequence of FTs applied on the consequent time
windows of a neural waveform. Sometimes, the STFT does not provide an efficient
resolution of the variable neural waveform at a short and large time as well as space scales
since a fixed window size is needed for examination using STFT. In an embodiment, the
system may use a wavelet method that can adjust the window size of its evaluating functions
over the complete range of time and space scales present in the neural waveform and
provides an effective time-frequency analysis. However, even wavelet analysis is non-
10
adaptive to the local variations of the data. Therefore, a competent data decomposition
method is desired.
[00036] In a preferred embodiment, the system uses EMD based features for analysis
and classification of signals and detect a lie. The EMD can effectively evaluate the
underlying patterns of the non-stationary, time-series data such as an EEG. A number of
amplitude and frequency-modulated oscillatory segments, called intrinsic mode functions
(IMFs), are obtained from the EMD of the EEG data. These IMFs convey valuable
information about the underlying structure of the data under investigation. The correlation
between ERP components provides supplementary information about whether the
components are providing the same or distinct neural information.
[00037] The system uses an SVM classifier to detect a lie from the extracted set of
features. One of the advantages of the SVM classifier is that it can map the nonlinearly
separable input data into a feature space where it can be linearly separated. SVM has been
earlier used in the detection of hidden information in the human brain. In an earlier lie
detection study, time, frequency, and wavelet features have been extracted, and 95.74%
averaged testing accuracy is obtained with the SVM classification model.
[00038] The system extracts features based on the EMD of EEG data along with the
correlation coefficients, combines them with other features, and uses the SVM-based model
to accurately identify a lying person. The system can receive EEG data from multiple users
and identify a lying person among a group of persons.
[00039] The EMD-based features calculated here have not been reported in earlier
works of EEG-based lie detection. In the present work, the analytic presentation of IMFs is
obtained. Afterward, the initial three IMFs have been assessed for extracting time and
11
frequency domain statistical features. Power spectral density (PSD) is calculated to obtain the
frequency domain statistical features.
[00040] Additionally, correlation coefficients have been evaluated among Fz, Cz,
and Pz electrode sites to examine the similarity of probe responses of the subjects at these
sites. Hence, the time and frequency features used in the earlier study are taken together with
the wavelet features, EMD-based features, and the correlation coefficients calculated in the
present study, and all these features collectively represent the proposed hybrid combination
of features. The system uses an SVM-based classification model for analyzing the extracted
features and the correlation coefficients and detecting if a person is lying.
[00041] FIG. 1 illustrates an exemplary networked environment of a lie detection
system in accordance with an embodiment of the present disclosure. As shown in FIG. 1, an
EEG machine 102 can collect EEG data through probe sensors attached to the human scalp. A
lie detection system 104 can be combined with the EEG features to work as a stand-alone
device or can receive the EEG signal from the EEG machine 102. The lie detection system
104 can receive the EEG signal directly from the EEG machine 102 connected through a
suitable wired interface or through a network 106. In an embodiment, users (e.g., doctors,
technicians, nurses, patient, or patient representatives, etc.) can receive a report from the lie
detection system 104 using preapprove user devices 108a-n. A user can access the lie
detection system 104 remotely through network 106, which can be a Local Area Network, a
Wide Area Network, or the Internet.
[00042] The lie detection system 104 extracts a set of features, obtains correlation
coefficient among the Fz, Cz, and Pz components, and applies an SVM-based model on the
set of features and correlation coefficient (collectively referred to as a hybrid set of features)
to detect a lie.
12
[00043] FIG. 2 illustrates an exemplary module diagram of a lie detection system in
accordance with an embodiment of the present invention. A lie detection system 202 includes
an EEG data receiving module 204 configured to receive EEG signals through a wired or
wireless interface. System 202 may use a pre-classified training data set of EEG signals of
people of diverse backgrounds and professions. For example, the pre-classified training data
set can include EEG data collected from males and females from different professional
backgrounds. The EEG signal can be obtained in response to three kinds of stimuli, which
include probe (P) stimuli (related to an offense under investigation), target (T) stimuli
(information not related to the offense but known to all the subjects), and irrelevant (I) stimuli
(unrelated to an offense under investigation). The EEG data can be classified into two groups
of subjects, guilty group and innocent group. The EEG signals can be acquired according to
the International 10-20 electrode placement system from the nine electrode locations: C3, Cz,
C4, P3, Pz, P4, O1, O2, and Oz. The system also collects Horizontal electrooculogram
(HEOG) and vertical electrooculogram (VEOG) data. The EEG signals corresponding to the
Fz electrode site are also collected.
[00044] System 202 applies band-pass filters on all the acquired data in the range of
0.1-30 Hz. The sampling rate can be 500 Hz. The EEG electrodes may use the right earlobe
as the reference. The brain waveforms generated in response to probe stimuli on the Pz
electrode location are used for further processing. System 202 uses a substantial number of
averaged probe waveforms, each in the guilty group and innocent group. It has been observed
that the guilty group waveforms characterize the P300 component, and the innocent group
waveforms characterize the non-P300 component. An averaged probe waveform at the Pz
location corresponding to a guilty subject and an innocent subject is depicted in FIG. 3.
13
[00045] System 202 includes a feature extraction module 206 configured to extract a
set of features from the EEG signals. Features can be extracted in different domains from the
probe waveform of a subject in the time interval of 0.2–1 second. Six morphological features
can be extracted in the time domain, namely maximum amplitude, latency, latency/amplitude
ratio, minimum amplitude, peak-to-peak amplitude, and positive area. Using the
morphological features, various aforementioned characteristics of the P300 component's
waveform can be effectively understood. Additionally, system 202 may perform time-domain
feature analysis, as it is beneficial due to EEG's high temporal resolution. The feature
extraction module 206 can extract three features in the frequency domain. These are
maximum frequency, mean frequency, and the power of the frequency band (0–3.9 Hz)
consisting of the P300 component. The system 202 may use Frequency domain spectral
analysis performed using Burg's method. Information about neural activities and their
underlying causes can be inferred through analysis in various EEG frequency bands. A
combination of different features has been earlier used to classify EEG signals. Wavelet
analysis provides information about the signals overlapping to various extents in the time
domain. Wavelets simultaneously help to understand the variation of the characteristics of
signals with time in the frequency domain. Ten features have been extracted in the timefrequency domain using the wavelet decomposition method. Each probe response has been
separated into seven pairs of wavelet coefficients, using Daubechies 4 (db4) wavelet. The
wavelet coefficients composing the frequency band of 0–3.9 Hz are measured as the ten
wavelet features. Quadratic B-spline-based wavelet transform has been used in earlier work.
[00046] The feature extraction module 206 extracts up to eighteen features based on
the EMD method. EMD-based features have been used to effectively capture the underlying
patterns of the non-stationary EEG data. Using EMD, energy-related to different time scales
14
is extracted in the form of IMFs. IMFs also provide information about the instantaneous
frequencies. The IMFs fulfill the following properties as listed below.
(1) The number of extrema (all maxima and minima) or zero crossings of a waveform should
be equivalent or vary by a maximum one.
(2) At a given point, the average value of the envelope represented by local maxima, and the
envelope represented by local minima is zero.
[00047] A signal y(t) can be decomposed into its constituent IMFs as follows:
(1) Initialize a(t) = y(t).
(2) Find the extrema (all maxima and minima) in the signal a(t).
(3) Join the maxima and minima distinctly utilizing cubic spline interpolation to obtain the
upper envelope eup(t) and lower envelope elow(t).
(4) Measure the local mean as .
(5) Calculate the difference of the original signal and the local mean as a(t) = a(t) - mean(t).
(6) Determine if a(t) satisfies the properties of an IMF or not, as stated above.
(7) Keep performing steps (2) to (6) until an IMF a(t) is found.
When the first IMF is computed, express v1(t) = a(t), which represents the smallest temporal
scale in y(t). In order to obtain other IMFs, calculate the residue res1(t) of the signal as res1(t)
= y(t) - v1(t). This procedure can be carried till the last residue obtained is a constant value,
producing no further IMFs. After the completion of EMD, the signal y(t) is expressed as
15
where N characterizes the number of IMF components and resN is the last residue obtained.
EMD can be performed using the MATLAB codes accessed. Afterward, the analytic
description of the extracted IMFs can be calculated to take out the DC offset from the
frequency constituents of the waveforms. The analytic description of an IMF vj(t) is
calculated as
where H{vj(t)} represents the Hilbert transform of vj(t), which characterizes the jth IMF
calculated after applying EMD to the signal y(t).
In an embodiment, the obtained IMFs are used to calculate the time and frequency domain
statistical features. The time-domain statistical features have been calculated because the
points distributed in the data can be described by their degree of spread, unevenness, and
collection about mean. After the application of Hilbert transform, differences in the
corresponding IMFs can be observed for a guilty subject and an innocent subject. Three timedomain statistical features can be analyzed to understand the differences in the IMFs of the
guilty and innocent subjects. These features are expressed for the individual IMF as
16
where M represents the number of samples in the IMF. The frequency-domain statistical
features have been calculated because of the ability of EMD to provide meaningful spectral
information of the EEG signals.
[00048] In an embodiment, PSD can be evaluated as
where dg[t] describes the autocorrelation of g[t], represented as dg[t] = E(g[n]g*[n]). Three
frequency domain statistical features have been evaluated. In order to understand the
variation of the major frequency of a signal over time, a spectral centroid measure is
calculated. The spectral centroid feature is evaluated as
where P(f) is the amplitude of the frequency in the power spectrum, the variation in the
frequency content is anticipated for the IMFs obtained for the guilty subject and innocent
subject. To capture this difference, the variation coefficient is evaluated as
[00049] Skewness tells about the unevenness in the distribution of data in a dataset.
In order to understand the characteristics of the data, the spectral skew feature is measured as
17
[00050] In this way, the aforementioned six-time and frequency domains statistical
features are calculated for one IMF, resulting in a total of 18 features for the three IMFs.
The feature extraction module 206 determines the correlation coefficient. The correlation
coefficient provides a measure of resemblance between two signals and can therefore serve as
an important feature in providing discriminating information between the guilty subjects and
the innocent subjects. Therefore, three correlation coefficients can be evaluated between FzCz, Fz-Pz, and Cz-Pz electrode locations to understand the influence of P300 among these
sites. The correlation coefficients will reveal interactions among these sites in the neural data.
The correlation coefficient between two time-domain variables {U} = {u1, u2, u3,…, uK} and
{V} = {v1, v2, v3,…, vK} is evaluated as
[00051] Two feature specimen sets representing P300 waveforms of guilty subjects
(class label 1) and P300 waveforms of innocent subjects (class label - 1) are obtained. Forty
features correspond to each feature specimen set. In some embodiments, the system may use
forty features, consisting of six-time domains, three frequency domains, ten wavelets,
eighteen EMD-based, and three correlation coefficient features. The features are scaled to [- 1
1] range before performing classification. During classification, all the forty features are
assigned equal weight.
System 202 includes an SVM classification and training module 208 configured to use the set
of extracted features and correlation coefficients to detect lie from the EEG signal. SVM is a
robust method to classify nonlinear data. SVM develops a decision hyperplane in a way that
the distance of separation between the output classes of data is maximum. Let {ci, di}, i = 1,
18
2,…, m represents the feature-class label values in the training data, where and
, and then, the following optimization problem is resolved
where C>0 describes the penalty attribute of the error value. According to Mercer's theorem,
which provides the eligibility of a kernel function to be used with SVM, two of the main
kernels are polynomial and radial basis function (RBF). Polynomial and RBF kernels are
described as:
where d is equal to degree.
RBF:
Here, ϒ represents the distribution of the data in the kernel.
[00052] The system 202 can include RBF kernel with SVM, as it provides better
results as compared to polynomial kernel. After feature extraction, initially, data are divided
into 70% and 30%. Testing data are taken as 30%. Out of the remaining 70%, 15% validation
data and 55% training data are selected. The validation data can be used to obtain the optimal
values of the classifier and kernel parameters which are further utilized for training the
model. The grid search procedure has been used with tenfold cross-validation to obtain the
optimal values of C and ϒ. The grid search is performed within the range of C = [2-5
, 2-4
,…,
2
15] and ϒ = [2-15, 2-14,…, 23
], with step size being equal to 21
. When 15% validation data are
selected, the optimal values of classification model parameters C and ϒ obtained
19
corresponding to different combinations of features provide the highest average crossvalidation accuracy across the ten divisions of the validation set using tenfold crossvalidation. Moreover, in general, the predicted accuracy obtained with the validation data
closely demonstrates the classification model performance on the unseen testing data. In an
embodiment, the LIBSVM package can be used to apply an SVM-based classification
algorithm.
[00053] System 202 includes a lie detection module 210 that, based on the data
classification performed by module 208, detects lie and notify to a user. The system may
include a buzzer that can play an alert sound of detection of a lie signal. In an embodiment,
the signal classified as a lie can be sent across to a remote user device for presentation and
further diagnostics.
[00054] FIG. 3 illustrates an example EEG probe waveform at Pz location of a guilty
subject and an innocent subject in accordance with an embodiment of the present invention.
FIG. 4 illustrates an example training EEG probe signals used for training SVM classifier
used for lie detection in accordance with an embodiment of the present disclosure. Similar
training data sets can be used by the system to train module 208.
[00055] FIG. 5 illustrates an exemplary flow chart of a method for lie-detecting
using EEG signals in accordance with an embodiment of the present disclosure. The method
500 to lie detection is based on extracting different domain features and combining them with
a customized SVM classifier. The method uses a combination of time, frequency, wavelet,
EMD-based, and correlation coefficient features. Time-domain features effectively capture
the variations in the shape and timing of the waveforms generated in the brain due to external
events. Frequency domain features exhibit the deviations in the spectral content of EEG
signals between guilty and innocent subjects. The neural component patterns which overlap
20
in time are detected precisely by wavelet features. EMD-based features demonstrate better
adaptability to the local characteristic timescale of brain signals. The correlation coefficients
explain the degree of interaction among Fz, Cz, and Pz electrode sites in EEG signals in
response to external events. Primarily, the time and frequency domain statistical features
extracted from EMD of the EEG data have significantly contributed to identifying the
differences in the probe responses of the guilty and innocent subjects. When the neural
information provided by 40 different features is combined and fed into SVM, averaged
training accuracy of 99.94%, averaged testing accuracy of 98.8%, and maximum testing
accuracy of 99.44% are attained. In the proposed methodology, mainly, the probe responses
of the subjects at only the Pz electrode location can be considered. The probe, target, and the
irrelevant responses of the subjects at the other midline electrode locations can also provide
valuable information related to lie detection.
[00056] The method includes steps of extracting different features of probe signal at
electrode location as shown at block 502, combining various features for testing and
classification as shown at block 504, scaling the features to [-1,1] as shown at block 506,
performing SVM classifier training with the optimized values as shown at block 508,
determining classification accuracy using the trained classified as shown at block 510, and
finding the maximum testing classification accuracy by analyzing various accuracies obtained
for different features groups, as shown at block 512. In an embodiment, the grid search
method can be used to get the optimal values of C and ϒ (RBF kernel) using 10-fold crossvalidation, as shown at block 514.
[00057] FIG. 6 illustrates an example computer system used for the implementation
feature of the present invention. As shown in FIG. 6, a computer system includes an external
21
storage device 610, a bus 620, a main memory 630, a read-only memory 640, a mass storage
device 650, a communication port 660, and a processor 670.
[00058] Those skilled in the art will appreciate that computer system 600 may
include more than one processor, 670, and communication ports 660. Examples of processor
670 include, but are not limited to, an Intel® Itanium® or Itanium 2 processor(s), or AMD®
Opteron® or Athlon MP® processor(s), Motorola® lines of processors, FortiSOC™ system
on chip processors or other future processors. Processor 670 may include various modules
associated with embodiments of the present invention.
[00059] Communication port 660 can be any of an RS-232 port for use with a
modem-based dialup connection, a 10/100 Ethernet port, a Gigabit, or 10 Gigabit port using
copper or fiber, a serial port, a parallel port, or other existing or future ports. Communication
port 660 may be chosen depending on a network, such as a Local Area Network (LAN), Wide
Area Network (WAN), or any network to which the computer system connects.
[00060] Memory 630 can be Random Access Memory (RAM) or any other dynamic
storage device commonly known in the art. Read-only memory 640 can be any static storage
device(s), e.g., but not limited to, a Programmable Read-Only Memory (PROM) chips for
storing static information, e.g., start-up or BIOS instructions for processor 670.
[00061] Mass storage 650 may be any current or future mass storage solution, which
can be used to store information and/or instructions. Exemplary mass storage solutions
include, but are not limited to, Parallel Advanced Technology Attachment (PATA) or Serial
Advanced Technology Attachment (SATA) hard disk drives or solid-state drives (internal or
external, e.g., having Universal Serial Bus (USB) and/or Firewire interfaces), e.g., those
available from Seagate (e.g., the Seagate Barracuda 7200 family) or Hitachi (e.g., the Hitachi
22
Deskstar 7K1000), one or more optical discs, Redundant Array of Independent Disks (RAID)
storage, e.g., an array of disks (e.g., SATA arrays), available from various vendors including
Dot Hill Systems Corp., LaCie, Nexsan Technologies, Inc. and Enhance Technology, Inc.
[00062] Bus 620 communicatively couples processor(s) 670 with the other memory,
storage, and communication blocks. Bus 620 can be, e.g., a Peripheral Component
Interconnect (PCI) / PCI Extended (PCI-X) bus, Small Computer System Interface (SCSI),
USB, or the like, for connecting expansion cards, drives, and other subsystems as well as
other buses, such a front side bus (FSB), which connects processor 670 to a software system.
[00063] Optionally, operator and administrative interfaces, e.g., a display, keyboard,
and a cursor control device, may also be coupled to bus 620 to support direct operator
interaction with the computer system. Other operator and administrative interfaces can be
provided through network connections connected through communication port 660. An
external storage device 650 can be any kind of external hard-drives, floppy drives,
IOMEGA® Zip Drives, Compact Disc - Read-Only Memory (CD-ROM), Compact Disc -
Re-Writable (CD-RW), Digital Video Disk - Read Only Memory (DVD-ROM). The
components described above are meant only to exemplify various possibilities. In no way
should the aforementioned exemplary computer system limit the scope of the present
disclosure.
23
Advantages of the Invention
[00064] An advantage of the present disclosure is to overcome one or more
drawbacks associated with the conventional mechanisms.
[00065] The invention provides a system and method for lie detection.
[00066] The invention provides a system and method for lie detection using EEG
data.
[00067] The invention provides a system and method for extracting relevant
information from EEG and using the extracted information for lie detection.
[00068] The invention provides a system and method of lie detection that can detect
lie with better accuracy and in a shorter time.
24

CLAIMS
We Claim:
1. A method for lie detection, the method comprising:
causing one or more processors which are coupled to a non-transitory storage device
and operable to perform the steps of:
receiving, at a computing device (500), from an EEG machine (102), EEG
signals comprising of Event-Related Potential (ERP) signal collected from at
least Fz, Cz, and Pz electrode locations on a human scalp;
extracting, at the computing device, a set of features comprising of time-domain
features, frequency domain features, wavelets, nonparametric weighted
features and empirical mode decomposition (EMD)-based features from the
EEG signals;
evaluating, at the computing, correlation coefficients among the Fz, Cz, and Pz
electrode sites to examine the similarity of probe responses ;
analyzing, at the computing device, the set of features and the correlation
coefficients to detect lie by a machine learning model using the set of features
and the correlation coefficients.
2. The method of claim 1, further comprising
obtaining features for intrinsic mode function (IMF) from EMD based features;
correlating between ERP components of the EEG signal to determine if whether the
components are providing the same or distinct neural information; and
analyzing the set of features based on the result of the determination.
25
3. The method of claim 1, wherein the machine learning model is a supervised machine
learning model.
4. The method of claim 1, wherein the ERP signal comprises a P300 component that is
used as a primary indicator by the machine learning model.
5. The method of claim 1, wherein RBF kernel is utilized with SVM.
6. A lie detection system, the system comprising:
an EEG data receiving module 204 configured, at a computing device (500), to
receive EEG signals comprising of Event-Related Potential (ERP) signal
collected from at least Fz, Cz, and Pz electrode locations on a human scalp;
a feature extraction module 206 configured, at the computing device,
to extract a set of features comprising of time-domain features, frequency domain
features, wavelets, nonparametric weighted features and empirical mode
decomposition (EMD)-based features from the EEG signals;
evaluating, at the computing device, correlation coefficients among the Fz, Cz, and Pz
electrode sites to examine the similarity of probe responses; and
an SVM classification and training module 208 configured, at the computing device,
to analyze the set of features and the correlation coefficients to detect lie using a
machine learning model that uses the set of features and the correlation
coefficients for lie detection.
7. The system of claim 6, wherein the feature extraction module 208 is configured, at the
computing device, to perform:
26
obtain features for intrinsic mode function (IMF) from EMD based features;
correlate between ERP components of the EEG signal to determine if whether the
components are providing the same or distinct neural information; and
analyze the set of features based on the result of the determination.
8. The system of claim 6, wherein the machine learning model, configured at the
computing device, is supervised machine learning model.
9. The system of claim 6, wherein the ERP signal comprises a P300 component that is
used as a primary indicator by the machine learning model.
10. The system of claim 6, wherein RBF kernel is utilized with SVM.