Description
A method and an apparatus for verifying an identity of a
person
The invention relates to image processing, and more
particularly, to verifying the identity of the person under
varying illumination conditions.
Identification methods to identify a person are generally
implemented for security reasons in a national high security
area as well as in a privately owned area. But identification
methods are generally not safe stand alone due to identity
theft cases. So, to keep up the security aligned, the
identity verification methods are implemented to enhance the
security measures. There are various identity verification
methods known which are based on biometrics of a person.
Biometrics provide more accurate verification as biometrics
have a lesser tendency of being stolen or tampered with due
to there direct association with the part of the body of the
person holding an identity. Biometrics includes retina,
finger prints, hand, face, etc. Generally, most types of
biometrics require the person to be cooperative. On the other
hand, face based identity verification generally does not
require the person's cooperation or active participation and
the identity verification can even be done without being
noticed by the identity holder.
Face based verification of an identity of a person has a
drawback in verifying a person in varying illumination
condition and it generally results in increase in probability
of both missed detection and false detection. Thus, the
illumination is required to be corrected in respect to a
reference image of the person, so as to provide the optimal
verification.
US 2006/0280344 Al discloses a face recognition apparatus and
method using illumination normalization where a plurality of
basis vectors is generated in reference to a plurality of
images in a training set for various illumination conditions,
an illumination normalizing coefficient is generated from a
first image by using the basis vectors, an illumination
normalized image is generated from a second face image using
the basis vectors and the illumination normalizing
coefficient and the recognition is done by matching between
the illumination normalized image and the first image.
It is an object of the invention to verify an identity of a
person irrespective of the varying illumination conditions
efficiently and cost effectively.
The said objective is achieved by a method for verifying an
identity of a person of claim 1 and an identity verification
system of claim 10.
According to an embodiment, the method for verifying an
identity of a person includes capturing a digital image of
the person wherein the image is made up of a plurality of
pixels on an area of the image, providing illumination
coefficients which are chosen such that a linear combination
of scalar basis functions defined on the area of the image
with the illumination coefficients results in a function
which determines illumination correction factors for the
brightness of pixels of the image for compensating
illumination differences between the image and the reference
image, correcting the brightness of the pixels of the image
based on the corresponding illumination correction factors to
generate a corrected image, providing the reference image of
a reference person, comparing the corrected image with the
reference image to verify the person as the reference person.
The underlying idea of the present invention is to provide
the reference image which reduces the step of comparing the
digital image to each of the other reference images for
verification, as the illumination compensated image would be
compared to the corresponding reference image only rather
than the entire database. It helps to reduce the processing
time for verification, thus provide a cost effective method.
Also, the illumination coefficients generated are based on
the various illumination conditions, so the verification is
done using these illumination coefficients irrespective of
the illumination conditions.
In one embodiment, the method for verifying the identity of
the person further includes providing the identity from the
person, identifying the person on the basis of the identity
from a set of identity, and choosing the reference image
linked to the identity. The process of identifying the person
and choosing the reference image accordingly makes the
verification process efficient and reduce a missed detection
and also a false detection.
In one embodiment, comparing the corrected image with the
reference image includes determining a similarity measure
between the corrected image and the reference image and
verifying the person as the reference person if the
similarity measure is higher than a threshold value. It
provides flexibility to the comparison to provide
approximation of the comparison of the image and the
reference image.
In a further embodiment, the method further includes
generating a false detection if the similarity measure is
greater than the threshold value and both the corrected image
and the reference image are from different persons and
processing the false detection to optimize the threshold
value. The number of false detection can thus be reduced.
In a still further embodiment, the method further comprises
generating a missed detection rate if the similarity measure
is lesser than the threshold value and both the corrected
image and the reference image are from the same persons, and
processing the missed detection rate to optimize the
threshold value. The number of missed detection can thus be
reduced.
In an exemplary embodiment, the correcting of the brightness
of the pixels of the image includes multiplying the
brightness of the pixels of the image with the corresponding
illumination correction factors. This embodiment provides
more precision to the corrected image.
In a further embodiment, the method further comprises
capturing the digital image from the person, providing the
reference image of the person, and defining the illumination
coefficients such that the linear combination of scalar basis
functions defined on the area of the image with the
illumination coefficients results in a function which
determines the illumination correction factors for the
brightness of the pixels of the image for compensating
illumination differences between the captured image and the
reference image. Defining the illumination coefficients in
such a way provides the illumination correction factors which
corrects the image more accurately.
In a one embodiment, the method further includes storing the
reference image in the database and linking the reference
image with the identity. This allows the reference image to
be retrieved from the database for comparing with the
corrected image which is linked to the identity of the
person.
In an exemplary embodiment, the method is included for the
verification of the person at a boarding point of an aircraft
at an airport, as the accurate verification is required for
national security issues. The method includes storing the
reference image in the database at a check-in point at the
airport and verifying the identity of the person at a
boarding point of an aircraft at the airport.
The above-mentioned and other features of the invention will
now be addressed with reference to the drawings of preferred
embodiments of the method and the system for verifying an
identity of a person. The illustrated embodiments of the
method and the system for verifying an identity of the person
are intended to illustrate, but not to limit the invention.
The drawings contain the following figures, in which like
numbers refers to like parts, throughout the description and
drawings.
FIG. 1 is a schematic diagram of an identity verification
system of a person in accordance with an embodiment of the
present invention,
FIG. 2 is a schematic diagram showing the system to verify an
identity of the person at an airport.
FIG. 1 illustrates an identity verification system 46 for a
person 2. According to the system 46, an image capturer 48
captures a digital image 4 of the person 2, an illumination
coefficient provider 50 provides the illumination
coefficients 6 which are defined such that a linear
combination of scalar basis functions 8 defined on the area
10 of the image 4 with the illumination coefficients 6
results in a function 36 which determines illumination
correction factors 12, an image processor 52 which corrects
the brightness of the pixels 14 by using the illumination
correction factors 12 to produce the corrected image 18, a
database 26 for storing and providing a reference image 16 of
a reference person 20 linked to the identity 22, and image
comparator 54 which compares the corrected image 18 with the
reference image 16 to verify the person 2 as the reference
person 20.
The digital image 4 is a 2-D image, but the image 4 can also
be a 3-D image. If a 3-D image is used than the system 46
should also be adapted to store, process and compare the 3-D
images. The image 4 can also be an analog image captured by
an image capturer 48 adapted to capture an analog image. If
the analog image is captured than the system 46 should also
be adapted to store, process and compare the analog image.
The image 4 is a combination of the pixels 14 on which the
linear combination of scalar basis functions 8 is defined.
The pixels 14 of image 4 are corrected for compensating the
illumination difference between the image 4 and the reference
image 16 by using the illumination correction factors 12.
The image capturer 48 is a camera, but it need not be a
camera rather it can be any optical device like mirrors,
lenses, telescopes, microscopes, etc having image capturing
features which could captures the image 4. The image capturer
48 captures the image 4 of the person 2 particularly of his
face whenever the person 2 poses before the image capturer
48. The image capturer 48 is not limited to capture the image
4 of face of the person 2 rather it can capture the image 4
of any of the essential features from a body part of the
person 2 together including the face or just the part of the
body, or even it can capture the image 4 of any part of the
face, like eyes, forehead, chin, etc. which would be used for
purpose of comparing by the comparator 54.
The image 4 comprises of pixels 14. The image 4 can further
be represented as a product of the illumination coefficients
6 and a fixed reflectance angle over each of the pixels 14 of
the image 4:

where L can be denoted as a linear combination of the scalar
basis functions 8 representing an illumination of the image 4
and R can be denoted as a combination of functions
representing a fixed reflectance angle in the image 4.
The illumination coefficient provider 50 provides the
illumination coefficient (A) 6 which are defined such that a
linear combination of the scalar basis functions (L) 8
results in a function 36 which determines the illumination
correction factors 12 for correcting the brightness of the
pixels 14 of the image 4 for compensating illumination
differences between the image 4 and the reference image 16.
The linear basis functions 8 is a Legendre basis functions or
any other functions which can represent the illumination
coefficient 6 either accurately or approximately according to
security requirements or performance requirements or any
specific requirements. (NA) represents the Legendre basis
function defined over the area 10 of the image 4.
Let Pk(x) denote k th legendre basis function. Then for N? =
2k+1, A = [?0, . . .?NA] T, than the illumination correction factor
12 at the pixel 14 of the image 4 is computed as:

Rewriting T0 and T1 as vectors, we get [T1]vec = [T0]Vec +
[To]vecPA so that when ? = 0, T1 = T0. refers to
multiplying each row of P by T0(x, y) . Given T1 and T0, the
Legendre coefficients ?1 that generates function 36 which
determines the illumination correction factors 12 for the
brightness of the pixels 14 of the image 4 for compensating
illumination differences between the captured image (T1) 4
and the reference image (To) 16 can be computed by solving
the least squares problem:

The system 46 also works for more than one person and for
various illumination conditions.
The image processor 52 corrects the brightness of the pixels
14 of the image (T1) 4 using equation (B) by multiplying each
pixel 14 of the image 4 with the illumination correction
factors 12 generated from the function P(x,y) 36.
The identifier 56 identifies the person 2 on the basis of the
identity 22 provided by the person 2. The identifier 56
identifies the identity 22 from a set of identity 24 and
chooses the reference image 16 from a set of reference image
68 linked to the identity 22 from the database 26. The
identifier 56 is a computer processor which can match the
identity 22 with entire set of identity 24 to identify the
matching identity 22 from the set of identity 24. The
identifier 56 also provides the reference image 16 chosen to
the comparator 54. The identifier 56 can also be any other
processor like microcontroller or a calculation device or a
manual calculator sufficient enough to identify the identity
22 from the set of identity 24 and choose the reference image
16 from a set of reference image 68 to be provided to the
comparator 54.
The database 26 stores the reference image 16 of a person 2
linked to the identity 22 in a storage device 64 and provides
the reference image 16 of the reference person 20 to the
comparator 54 for the purpose of comparing the reference
image 16 with the corrected image 18. The database 26
represents an integral collection of the reference images 16
of the reference person 20 as a set of reference images 68
and a set of identity 24, wherein each of the identity 22 is
linked to each reference image 16. The linking of the
identity 22 is done by the linker 68 by adding a pointer to
the identity 22 pointing to the reference image 16 stored in
the database 26. The linking of the identity 22 can also be
done in any other way such as indexing the identity 22 and
establishing a relation with the image 4, etc. When a query
is raised on the basis of the identity 22 than the database
26 provides the reference image 16 linked to the identity 22.
The database 26 is a relational database created in a storage
device 64 to store the reference image 16 in relation to the
identity 22. The database 26 can also be a data warehouse, a
real time database, a distributed database, or any other such
database which can store the reference image 16 linked to the
identity 22 and can provide the reference image 16 to the
image comparator 54 for comparing it with the image 4.
The storage device 64 is a memory unit such as computer based
memory units like CD ROM, DVD ROM, USB data storage or
magnetic based memory unit like magnetic tape, or optical
storage unit or any such memory storage unit which has a
storage capacity to store the reference image 16 and which
can provide the reference image 16 from the database 26 to
the image comparator 54. The storage device writes the
reference image 16 and the identity 22 once and provides the
reference image 16 to the image comparator 54 whenever the
comparator 54 requires the reference image 16 for comparing
the reference image 16 with the image 4. The storage device
64 can also write more than once into the memory space left
inside the storage device 64 or the storage device 64 can
delete the reference image 16 stored and re-write into the
memory space a new reference image and the identity 22 linked
to it. The storage device 64 can also provide the reference
image 16 once to the image comparator 54 for comparing the
reference image 16 with the image 4. In all the storage
device 64 should at least be able to write onto memory space
once to store the reference image 16 in the database 26 and
provide the reference image 16 once to the image comparator
54. The storage device 64 can also be a volatile storage
device to store the image 4 for a small period of time to
meet the security requirements or the personal information
secrecy requirements or any such requirements.
The image comparator 54 compares the corrected image 18 with
the reference image 16 of the reference person 20 to verify
the person 2 as the reference person 20. The reference image
16 of the reference person 20 is provided from the database
26 when the database 26 is queried in reference to the
identity 22, while the image comparator 54 receives the
corrected image 18 of the person 2 from the image processor
52.
The image comparator 54 also compares the corrected image 18
with the reference image 16 of the reference person 20 to
determine a similarity measure 28 on a basis of a threshold
value 30, so that the person 2 is verified as the reference
person 20 if the similarity measure 28 is less than or equal
to threshold value 30. The similarity measure 28 is
determined by creating a similarity matrix by comparing each
pixel of the reference image 16 with each pixel of the
corrected image 18 on a basis of the threshold value 30. Each
entry of the similarity matrix represents each pixel 14 of
the image 4. The threshold value 30 is a constant value to be
represented for each of the pixel 14, but the threshold value
30 need not be a constant value rather it can vary from pixel
to pixel. Also, the threshold value 30 can be represented as
matrix where each entry of the threshold matrix will
represent the threshold value 30 for each pixel 14. The
similarity measure 28 can also be determined as an average by
averaging all entries of the similarity measure by either
taking arithmetic mean, median or any other ways of
averaging.
The identity verification system 46 also has a false
detection generator 58 which generates a false detection 32
to optimize the threshold value 30 by the threshold value
optimizer 62. The false detection generator 58 generates a
false detection 32 when the similarity measure 28 is greater
than the threshold value 30 and both the corrected image 18
and the reference image 16 are from different persons.
The identity verification system 46 also has a missed
detection generator 60 which generates a missed detection 34
to optimize the threshold value 30 by the threshold value
optimizer 62. The missed detection generator 60 generates a
missed detection 34 when the similarity measure 28 is lesser
than the threshold value 30 and the corrected image 18 and
the reference image 16 are from the same person.
The threshold optimizer 62 can use both the false detection
32 and the missed detection 34 one by one sequentially at the
same time or both together in combination to optimize the
threshold value 30.
The system 46 can also be trained to work for various
illumination conditions. In such a case, the image capturer
48 would capture the image (T1) 4 and the corresponding
reference image (T0) 16 for each of the N individual and
equation (D) can be used to find the illumination
coefficients {?1,.....,?N} 6 for all the individual. Generally
the underlying distribution of the illumination coefficients
(?s) 6 is continuous. But on the basis of requirement the
illumination coefficients 6 obtained for different person
could be condensed down to represent one illumination
coefficient 6 for one illumination condition. Even on the
basis of accuracy required, even one illumination coefficient
6 can represent a continuous set of illumination conditions.
A k-means clustering of the illumination coefficients
{?1,.....,?N} 6 can be carried out to yield the centroid set of
{c1, ....., ck}. Even the k-clustering could be used to
determine another centroid set of where each centroid can be
used as a representation of a set of continuous illumination.
The k-clustering need not be done rather any other averaging
or mean calculating or any such technique can be used to
condense down the number of illumination coefficients 6.
The system 4 6 can also be tested to determine whether or not
the system 46 trained, work for various illumination
conditions. In such a case, the image capturer 48 will
capture the image (T1) 4 and the corresponding reference
image (T0) 16 for each of the N person 2, 20. The image
processor 52 will correct the brightness of the pixels 14 of
the Image (T1) 4 on the basis of various illumination
correction factors 12 for each of the N person 2, 20. The
image (T1) 4 will be compared by the comparator 54 with the
corresponding reference image (T0) 16 for each of the N
individuals to verify the person 2 as the reference person
20. The image (T1) 4 can also be compared by the comparator
54 with the corresponding reference image (T0) 16 to
determine the similarity measure 28. The image (T1) 4 can
also be compared by the comparator 54 with the corresponding
reference image (T0) 16 on a basis of a threshold value 30 to
determine the similarity measure 28. The threshold optimizer
62 can also be used to optimize the threshold value 30 while
testing the system 46.
FIG. 2 illustrates a schematic diagram of the identification
verification system 46 to verify the identity 22 of the
person 2 at an airport 40. The identification system 46 as
disclosed in FIG. 1 is implemented at the airport 40. The
image capturer 48 captures the reference image 16 from the
reference person 20 at the check-in point 38 and also the
identity 22 is issued to the reference person 20 at the
check-in point 38. The reference image 16 is stored in a
database 26 inside the storage device 64. The identity 22 is
also stored in the database 26 by linking the reference image
16 by the linker 68 inside the storage device 64. When the
person 2 appears at the boarding point 42 of the aircraft 44,
the image capturer 48 captures the image 4 of the person 2
and the identity 22 provided by the person 2 is verified. At
the time of verification, the image 4 is corrected by the
image processor 52 according to the illumination conditions
to provide the corrected image 18 and the corrected image 18
is compared to the reference image 16. The reference image 16
is provided by the database 26 when the database 26 is
queried on the basis of the identity 22. The comparator 54
compares the corrected image 18 and the reference image 16 to
verify the person 2 as the reference person 20 by verifying
the identity 22.
WE CLAIM
1. A method for verifying an identity (22) of a person (2),
comprising:
- capturing a digital image (4) of the person (2) wherein the
image (4) comprises a plurality of pixels (14) on an area
(10) of the image (4),
- providing illumination coefficients (6) which are defined
such that a linear combination (8) of scalar basis functions
defined on the area (10) of the image (4) with the
illumination coefficients (6) results in a function (36)
which determines illumination correction factors (12) for the
brightness of the pixels (14) of the image (4) for
compensating illumination differences between the image (4)
and a reference image (16),
- correcting the brightness of the pixels (14) of the image
(4) based on the corresponding illumination correction
factors to generate a corrected image (18),
- providing the reference image (16) of a reference person
(20),
- comparing the corrected image (18) with the reference image
(16) to verify the person (2) as the reference person (20).
2. The method according to claim 1, wherein the method
further comprises:
- providing the identity (22) from the person (2),
- identifying the person (2) on the basis of the identity
(22) from a set of identity (24),
- choosing the reference image (16) from a set of reference
images (68) linked to the identity (22) from a database (26).
3. The method according to claims 1 or 2, wherein comparing
the corrected image (18) with the reference image (16)
comprises:
- determining a similarity measure (28) between the corrected
image (18) and the reference image (16) and
- verifying the person (2) as the reference person (20) if
the similarity measure (28) is higher than a threshold value
(30) .
4. The method according to the claim 3, wherein the method
further comprises:
- generating a false detection (32) if the similarity measure
(28) is greater than the threshold value (30) and both the
corrected image (18) and the reference image (16) are from
different persons,
- processing the false detection (32) to optimize the
threshold value (30).
5. The method according to the claims 3 or 4, wherein the
method further comprises:
- generating a missed detection (34) if the similarity
measure (28) is lesser than the threshold value (30) and both
the corrected image (18) and the reference image (16) are
from the same persons,
- processing the missed detection (34) to optimize the
threshold value (30).
6. The method according to claims 1 to 5, wherein the
correcting of the brightness of the pixels (14) of the image
(4) comprises multiplying the brightness of the pixels (14)
of the image (4) with the corresponding illumination
correction factors (12).
7. The method according to claims 1 to 6, further comprising:
- capturing the digital image (4) from the person (2),
- providing the reference image (16) of the person (2),
- defining the illumination coefficients (6) such that a
linear combination of scalar basis functions (8) defined on
the area (10) of the image (4) with the illumination
coefficients (6) results in a function (36) which determines
the illumination correction factors (12) for the brightness
of the pixels (14) of the image (4) for compensating
illumination differences between the captured image (4) and
the reference image (16).
8. The method according to claims 1 to 7, further comprising:
- storing the reference image (16) in the database (26),
- linking the reference image (16) with the identity (22).
9. The method according to claims 1 to 9, further comprising:
- storing the reference image (16) in the database at a
check-in point (38) at an airport (40),
- verifying the identity (22) of the person (2) at a boarding
point (42) of an aircraft (44) at the airport (40).
10. An identity verification system (46) for a person (2),
comprising:
- an image capturer (48) adapted to capture an image (4) of
the person (2) wherein the image (4) comprises a plurality of
pixels (14) on an area (10) of the image (4),
- an illumination coefficient provider (50) adapted to
provide illumination coefficients (6) which are defined such
that a linear combination of scalar basis functions (8)
defined on the area (10) of the image (4) with the
illumination coefficients (6) results in a function (36)
which determines illumination correction factors (12) for
brightness of the pixels (14) of the image (4) for
compensating illumination differences between the captured
image (4) and the reference image (16),
- a image processor (52) adapted to correct brightness of the
pixels (14) of the image (4) with the corresponding
illumination correction factors (12) to generate a corrected
image (18) ,
- a database (26) for storing and providing a reference image
(16) of a reference person (20) linked to the identity (22),
- a comparator (54) adapted to compare the corrected image
(18) with the reference image (16) of the reference person
(20) to verify the person (2) as the reference person (20).
11. The system (46) according to claim 10, wherein the
identity (22) is provided by a person (2), the system (46)
further comprising:
- an identifier (56) adapted to identify the person (2) on
the basis of the identity (22) from a set of identity (24)
and adapted to choose the reference image (16) from a set of
reference image (68) linked to the identity (22) from the
database (26).
12. The system (46) according to claim 10 or 11, wherein the
comparator (54) is further adapted to determine a similarity
measure (28) between the corrected image (18) and the
reference image (16) of the reference person (20) on the
basis of a threshold value (30) such that the person (2) is
verified as the reference person (20) if the similarity
measure (28) is less than or equal to threshold value (30).
13. The system (46) according to claim 12, further
comprising:
- a false detection generator (58) adapted to generate a
false detection (32) if the similarity measure (28) is
greater than the threshold value (30) and both the corrected
image (18) and the reference image (16) are from different
Dersons.
- a threshold optimizer (62) adapted to process the false
detection (32) to optimize the threshold value (30).
14. The system (46) to claims 12 or 13, further comprising:
- a missed detection generator (60) adapted to generate a
missed detection (34) if the similarity measure (28) is
lesser than the threshold value (30) and the corrected image
(18) and the reference image (16) are from the same person,
- a threshold optimizer (62) adapted to process the missed
detection (34) to optimize the threshold value (30).
15. The system (46) according to claims 10 to 14, wherein the
processor (52) is adapted to correct the brightness of the
image (4) by multiplying the brightness of the pixels (14) of
the image (4) with the corresponding illumination correction
factors (12).
16. The system (46) according to claims 10 to 15, wherein:
- the image capturer (48) is adapted to capture the reference
image (16) of the person (2),
- the database(26) is adapted to provide the reference image
(16) of the person (2),
- the illumination coefficient provider (50) is further
adapted to define the illumination coefficients (6) such that
a linear combination of scalar basis functions (8) defined on
the area (10) of the image (4) with the illumination
coefficients (6) results in a function (36) which determines
the illumination correction factors (12) for brightness of
the pixels (14) of the image (4) for compensating
illumination differences between the captured image (4) and
the reference image (16).
17. The system (46) according to claims 10 to 16, further
comprising:
- a storage device (64) adapted to store the reference image
(16) in the database (26).
- a linker (68) adapted to link the reference image (16) with
the identity.
18. The system (46) according to claim 10 to 17, wherein:
- the storage device (64) is adapted to store the reference
image (16) in the database (26) at the check-in point (38) at
the airport (40) ,
- the comparator (54) is adapted to verify the identity (22)
of the person (2) at the boarding point (42) of the
aircraft(44) at the airport (40).

The invention relates to a method and system for verifying an
identity (22) of a person (2). The method comprises of
capturing a digital image (4) of the person (2) wherein the
image (4) comprises a plurality of pixels (14) on an area
(10) of the image (4), providing illumination coefficients
(6) which are defined such that a linear combination (8) of
scalar basis functions defined on the area (10) of the image
(4) with the illumination coefficients (6) results in a
function (36) which determines illumination correction
factors (12) for the brightness of the pixels (14) of the
image (4) for compensating illumination differences between
the captured image (4) and a reference image (16), correcting
the brightness of the pixels (14) of the image (4) based on
the corresponding illumination correction factors to generate
a corrected image (18), providing the reference image (16) of
a reference person (20), comparing the corrected image (18)
with the reference image (16) to verify the person (2) as the
reference person (20). The method further comprises of
storing the reference image (16) in the database at a check-
in point (38) at an airport (40) and verifying the identity
(22) of the person (2) at a boarding point (42) of an
aircraft (44) at the airport (40) .