Note: Descriptions are shown in the official language in which they were submitted.
CA 02546296 2008-10-31
APPARATUS, METHOD AND SYSTEM FOR SCREENING
RECEPTACLES AND PERSONS, HAVING IMAGE DISTORTION
CORRECTION FUNCTIONALITY
FIELD OF THE INVENTION
The present invention relates generally to security systems and, more
particularly, to
methods and systems for screening receptacles including, for example, luggage,
mail
parcels or cargo containers, to identify certain objects located therein,
where such
methods and systems implement image distortion correction functionality.
BACKGROUND
Security in airports, train stations, ports, office buildings and other public
or private
venues is becoming increasingly important particularly in light of recent
violent
events.
Typically, security screening systems make use of devices generating
penetrating
radiation, such as x-ray devices, to scan individual pieces of luggage to
generate an
image conveying the contents of the luggage. The image is displayed on a
screen and
is examined by a human operator whose task it is to detect and possibly
identify, on
the basis of the image, potentially threatening objects located in the
luggage. In
1
CA 02546296 2006-05-11
certain cases, some form of object recognition technology may be used to
assist the
human operator.
A deficiency with current systems is that they are mostly reliant on the human
operator to detect and identify potentially threatening objects. However, the
performance of the human operator greatly varies according to such factors as
poor
training and fatigue. As such, the detection and identification of threatening
objects is
highly susceptible to human error. Furthermore, it will be appreciated that
failure to
identify a threatening object, such as a weapon for example, may have serious
consequences, such as property damage, injuries and fatalities.
Another deficiency with current systems is that the labour costs associated
with such
systems are significant since human operators must view the images.
Consequently, there is a need in the industry for providing a method and
system for
use in screening luggage items, cargo containers, mail parcels or persons to
identify
certain objects that alleviate at least in part the deficiencies of the prior
art.
SUMMARY OF THE INVENTION
In accordance with a first broad aspect, the present application seeks to
provide an
apparatus suitable for screening a receptacle. The apparatus comprises an
input for
receiving an image signal associated with the receptacle, the image signal
conveying
an input image related to contents of the receptacle, the image signal having
been
produced by a device that is characterized by introducing distortion into the
input
image. The apparatus also comprises a processing unit in communication with
the
input, and operative for: applying a distortion correction process to the
image signal to
remove at least part of the distortion from the input image, thereby to
generate a
corrected image signal conveying at least one corrected image related to the
contents
of the receptacle; processing the corrected image signal in combination with a
plurality of data elements associated with a plurality of target objects in an
attempt to
detect a presence of at least one of said target objects in the receptacle;
and generating
a detection signal in response to detection of the presence of at least one of
said target
2
CA 02546296 2008-10-31
j
objects in the receptacle. The apparatus further comprises an output for
releasing the
detection signal.
In accordance with a second broad aspect, the present application seeks to
provide an
apparatus for detecting the presence of one or more prohibited objects in a
receptacle.
The apparatus comprises an input for receiving an input image conveying
graphic
information regarding contents of a receptacle, the image having been produced
by a
device that introduces distortion into the input image; a distortion
correction
functional unit operable for processing the input image to remove at least
part of the
distortion from the input image in order to derive at least one corrected
image; means
for processing the at least one corrected image in an attempt to detect
whether at least
one of said one or more prohibited objects is depicted in at least one of the
at least one
corrected image; and an output for releasing a signal in response to detecting
that at
least one of said one or more prohibited objects is depicted in at least one
of the at
least one corrected image.
In accordance with a third broad aspect, the present application seeks to
provide a
method for screening a receptacle, which comprises receiving an image signal
associated with the receptacle, the image signal conveying an input image
related to
contents of the receptacle, the image signal having been produced by a device
that is
characterized by introducing distortion into the input image; applying a
distortion
correction process to the image signal to remove at least part of the
distortion from the
input image, thereby to generate a corrected image signal conveying at least
one
corrected image related to the contents of the receptacle; processing the
corrected
image signal in combination with a plurality of data elements associated with
a
plurality of target objects in an attempt to detect a presence of at least one
of said
target objects in the receptacle; generating a detection signal in response to
detection
of the presence of at least one of said target objects in the receptacle; and
releasing the
detection signal.
In accordance with a fourth broad aspect, the present application seeks to
provide a
computer-readable storage medium comprising computer-readable program code
which, when interpreted by a computing apparatus, causes the computing
apparatus to
3
CA 02546296 2006-05-11
execute a method of screening a receptacle. The computer-readable program code
comprises first computer-readable program code for causing the computing
apparatus
to be attentive to receipt of an image signal associated with the receptacle,
the image
signal conveying an input image related to contents of the receptacle, the
image signal
having been produced by a device that is characterized by introducing
distortion into
the input image; second computer-readable program code for causing the
computing
apparatus to apply a distortion correction process to the image signal to
remove at
least part of the distortion from the input image, thereby to generate a
corrected image
signal conveying at least one corrected image related to the contents of the
receptacle;
third computer-readable program code for causing the computing apparatus to
process
the corrected image signal in combination with a plurality of data elements
associated
with a plurality of target objects in an attempt to detect a presence of at
least one of
said target objects in the receptacle; fourth computer-readable program code
for
causing the computing apparatus to generate a detection signal in response to
detection of the presence of at least one of said target objects in the
receptacle; and
fifth computer-readable program code for causing the computing apparatus to
release
the detection signal.
In accordance with a fifth broad aspect, the present application seeks to
provide an
apparatus for screening a receptacle. The apparatus comprises means for
receiving an
image signal associated with the receptacle, the image signal conveying an
input
image related to contents of the receptacle, the image signal having been
produced by
a device that is characterized by introducing distortion into the input image;
means for
applying a distortion correction process to the image signal to remove at
least part of
the distortion from the input image, thereby to generate a corrected image
signal
conveying at least one corrected image related to the contents of the
receptacle;
means for processing the corrected image signal in combination with a
plurality of
data elements associated with a plurality of target objects in an attempt to
detect a
presence of at least one of said target objects in the receptacle; means for
generating a
detection signal in response to detection of the presence of at least one of
said target
objects in the receptacle; and means for releasing the detection signal.
4
CA 02546296 2008-10-31
In accordance with a sixth broad aspect, the present application seeks to
provide a
system for screening a receptacle. The system comprises an image generation
device
operable to generate an image signal associated with the receptacle, the image
signal
conveying an input image related to contents of the receptacle, the input
image
containing distortion introduced by said image generation device. The system
also
comprises an apparatus in communication with said image generation device, and
operable for: applying a distortion correction process to the image signal to
remove at
least part of the distortion from the input image, thereby to generate a
corrected image
signal conveying at least one corrected image related to the contents of the
receptacle;
lo processing the corrected image signal in combination with a plurality of
data elements
associated with a plurality of target objects in an attempt to detect a
presence of at
least one of said target objects in the receptacle; and generating a detection
signal in
response to detection of the presence of at least one of said target objects
in the
receptacle. The system further comprises an output module for conveying
information derived at least in part on the basis of said detection signal to
a user of the
system.
In accordance with a seventh broad aspect, the present application seeks to
provide an
apparatus suitable for screening a receptacle. The apparatus comprises an
input for
receiving image signal associated with the receptacle, the image signal
conveying an
input image related to contents of the receptacle, the image signal having
been
produced by a device that is characterized by introducing distortion into the
input
image. The apparatus further comprising a processing unit in communication
with the
input. The processing unit being operative for: applying a distortion
correction
process to the image signal to remove at least part of the distortion from the
input
image, thereby to generate a corrected image signal conveying at least one
corrected
image related to the contents of the receptacle; processing the corrected
image signal
in an attempt to detect a presence of at least one target object in the
receptacle; and
generating a detection signal in response to detection of the presence of at
least one
target object in the receptacle. The apparatus also comprising an output for
releasing
the detection signal.
5
CA 02546296 2008-10-31
In accordance with an eigth broad aspect, the present application seeks to
provide a
method for screening a receptacle. The method comprises receiving image signal
associated with the receptacle, the image signal conveying an input image
related to
contents of the receptacle, the image signal having been produced by a device
that is
characterized by introducing distortion into the input image. The method also
comprises applying a distortion correction process to the image signal to
remove at
least part of the distortion from the input image, thereby to generate a
corrected image
signal conveying at least one corrected image related to the contents of the
receptacle.
The method also comprises processing the corrected image signal in an attempt
to
detect a presence of at least one target object in the receptacle and
generating a
detection signal in response to detection of the presence of at least one
target object in
the receptacle. The method also comprises releasing the detection signal.
For the purpose of this specification, the expression "receptacle" is used to
broadly
describe an entity adapted for receiving objects therein such as, for example,
a
luggage item, a cargo container or a mail parcel.
For the purpose of this specification, the expression "luggage item" is used
to broadly
describe luggage, suitcases, handbags, backpacks, briefcases, boxes, parcels
or any
other similar type of item suitable for containing objects therein.
For the purpose of this specification, the expression "cargo container" is
used to
broadly describe an enclosure for storing cargo such as would be used, for
example, in
a ship, train, truck or another suitable type of cargo container.
Other aspects and features of the present invention will become apparent to
those
ordinarily skilled in the art upon review of the following description of
specific
embodiments of the invention in conjunction with the accompanying Figures.
5a
CA 02546296 2006-05-11
BRIEF DESCRIPTION OF THE DRAWINGS
A detailed description of embodiments of the present invention is provided
herein
below, by way of example only, with reference to the accompanying drawings, in
which:
Figure 1 is a high-level block diagram of a system for screening a receptacle
in
accordance with a specific example of implementation of the present
invention;
Figure 2 is a block diagram of an output module suitable for use in connection
with
the system depicted in Figure 1 in accordance with a specific example of
implementation of the present invention;
Figure 3 is a block diagram of an apparatus for processing images suitable for
use in
connection with the system depicted in Figure 1 in accordance with a specific
example of implementation of the present invention;
Figures 4a and 4b depict specific examples of visual outputs conveying the
presence
of at least one target object in the receptacle in accordance with specific
examples of implementation of the present invention;
Figure 5 is a flow diagram depicting a process for detecting a presence of at
least one
target object in the receptacle in accordance with specific examples of
implementation of the present invention;
Figure 6 shows three images associated with a target object suitable for use
in
connection with the system depicted in Figure 1, each image depicting the
target object in a different orientation, in accordance with a specific
example
of implementation of the present invention;
Figure 7 shows a mosaic image including a plurality of sub-images associated
with a
target object suitable for use in connection with the system depicted in
Figure
6
CA 02546296 2006-05-11
1, each sub-image depicting the target object in a different orientation and
scale, in accordance with a specific example of implementation of the present
invention;
Figure 8A is a functional block diagram of a receptacle screening system
including an
optical correlator in accordance with a specific example of implementation of
the present invention;
Figure 8B shows a peak observed in an output of the optical correlator of Fig.
8A;
Figure 8C is a functional block diagram of a receptacle screening system
including an
optical correlator in accordance with another specific example of
implementation of the present invention;
Figure 9 is a block diagram depicting the functioning of the optical
correlator in
accordance with a specific example of implementation of the present
invention;
Figure 10 depicts a Fourier transform, amplitude and phase, of the spatial
domain
image for number 2;
Figure 11 shows two images associated with a person suitable for use in a
system for
screening a person in accordance with a specific example of implementation of
the present invention;
Figure 12 is a block diagram of an apparatus suitable for implementing at
least a
portion of the modules depicted in connection with the apparatus for
processing images shown in Figure 3 in accordance with a specific example of
implementation of the present invention;
Figure 13 diagrammatically illustrates the effect of distortion correction
applied by the
apparatus for processing images;
7
CA 02546296 2006-05-11
Figure 14 diagrammatically illustrates a template for use in a registration
process in
order to model distortion introduced by the image generation device.
In the drawings, the embodiments of the invention are illustrated by way of
examples.
It is to be expressly understood that the description and drawings are only
for the
purpose of illustration and are an aid for understanding. They are not
intended to be a
definition of the limits of the invention.
DETAILED DESCRIPTION
Shown in Figure 1 is a system 100 for screening a receptacle 104 in accordance
with a
specific example of implementation of the present invention. The system 100
includes
an image generation device 102, an apparatus 106 in communication with the
image
generation device 102 and an output module 108. The image generation device
102
generates an image signal 150 associated with the receptacle 104. The image
signal
150 conveys an input image 800 related to contents of the receptacle 104.
It should be noted that the image generation device 102 may introduce
distortion into
the input image 800. More specifically, different objects appearing in the
input image
800 may be distorted to different degrees, depending on the position of the
object in
question within the input image 800 and depending on the height of the object
in
question within the receptacle 104 (which sets the distance between the object
in
question and the image generation device 102).
The apparatus 106 receives the image signal 150 and processes the image signal
150
in combination with a plurality of data elements associated with a plurality
of target
objects in an attempt to detect a presence of one or more target objects in
the
receptacle 104. In a specific non-limiting implementation, the data elements
associated with the plurality of target objects are stored in a database I 10.
In response to detection of the presence of one or more target objects in the
receptacle
104, the apparatus 106 generates a detection signal 160 which conveys the
presence of
one or more target objects in the receptacle 104. Examples of the manner in
which
8
CA 02546296 2006-05-11
the detection signal 160 can be generated are described later on in the
specification.
The output module 108 conveys information derived at least in part on the
basis of the
detection signal 160 to a user of the system.
Advantageously, the system 100 provides assistance to the human security
personnel
using the system 100 in detecting certain target objects and decreases the
susceptibility of the screening process to human error.
In a specific example of implementation, the image generation device 102 uses
penetrating radiation or emitted radiation to generate the image signal 150.
Specific
examples of such devices include, without being limited to, x-ray, gamma ray,
computed tomography (CT scans), thermal imaging and millimeter wave devices.
Such devices are known in the art and as such will not be described further
here. In a
non-limiting example of implementation, the image generation device 102 is a
conventional x-ray machine and the input image 800 related to the contents of
the
receptacle 104 is an x-ray image of the receptacle 104 generated by the x-ray
machine.
The input image 800 related to the contents of the receptacle 104, which is
conveyed
by the image signal 150, may be a two-dimensional (2-D) image or a three-
dimensional (3-D) image, and may be in any suitable format such as, without
limitation, VGA, SVGA, XGA, JPEG, GIF, TIFF and bitmap amongst others.
Preferably, the input image 800 related to the contents of the receptacle 104
is in a
format that can be displayed on a display screen.
In some embodiments (e.g., where the receptacle 104 is large, as is the case
with a
cargo container), the image generation device 102 may be configured to scan
the
receptacle 104 along various axes to generate an image signal conveying
multiple
input images related to the contents of the receptacle 104. Scanning methods
for large
objects are known in the art and as such will not be described further here.
Each of
the multiple images is then processed in accordance with the method described
herein
below to detect the presence of one or more target objects in the receptacle
104.
9
CA 02546296 2006-05-11
In a specific example of implementation, the database 110 includes a plurality
of
entries associated with respective target objects that the system 100 is
designed to
detect. A non-limiting example of a target object includes a weapon. The entry
in the
database 110 that is associated with a particular target object includes a
data element
associated with the particular target object.
In a first non-limiting example of implementation, the data element associated
with
the particular target object comprises one or more images of the particular
target
object. The format of the one or more images of the particular target object
will
depend upon the image processing algorithm implemented by the apparatus 106,
to be
described later. Where plural images of the particular target object are
provided, these
images may depict the particular target object in various orientations. Figure
6 of the
drawings depicts an example of arbitrary 3D orientations of a particular
target object.
In a second non-limiting example of implementation, the data element
associated with
the particular target object comprises the Fourier transform of one or more
images of
the particular target object.
Optionally, for the entry associated with a particular target object,
characteristics of
the particular target object are provided. Such characteristics may include,
without
being limited to, the name of the particular target object, its associated
threat level, the
recommended handling procedure when the particular target object is detected
and
any other suitable information. Optionally still, the entry associated with a
particular
target object is also associated with a respective target object identifier
data element.
Although the database 110 has been shown in Figure 1 to be a component
separate
from the apparatus 106, it will be appreciated that in certain embodiments the
database 110 may be part of apparatus 106 and that such implementations do not
detract from the spirit of the invention. In addition, it will also be
appreciated that in
certain implementations, the database 110 is shared between multiple
apparatuses
106.
CA 02546296 2006-05-11
In a specific example of implementation, the output module 108 conveys to a
user of
the system information derived at least in part on the basis of the detection
signal 160.
A specific example of implementation of the output module 108 is shown in
Figure 2.
As depicted, the output module includes an output device 202 and an output
controller
unit 200.
The output controller unit 200 receives from the apparatus 106 (shown in
Figure 1)
the detection signal 160 conveying the presence of one or more target objects
(hereinafter referred to as "detected target objects") in the receptacle 104.
In a
specific implementation, the detection signal 160 conveys information
regarding the
position and/or orientation of the one or more target detected objects within
the
receptacle 104. Optionally, the detection signal 160 also conveys one or more
target
object identifier data elements, which permit identification of the one or
more
detected target objects. The one or more target object identifier data
elements are
associated with corresponding entries in the database 110.
In a first specific example of implementation, the output controller unit 200
is adapted
to cause a display unit to convey information related to the one or more
detected
target objects. In a non-limiting example of implementation, the output
controller unit
200 generates image data conveying the location of the one or more detected
target
objects within the receptacle 104. Optionally, the output controller unit 200
also
extracts characteristics of the one or more detected target objects from the
database
110 on the basis of the target object identifier data element and generates
image data
conveying the characteristics of the one or more detected target objects. In
yet
another non-limiting example of implementation, the output controller unit 200
generates image data conveying the location of the one or more detected target
objects
within the receptacle 104 in combination with the input image generated by the
image
generation device 102 (shown in Figure 1).
In a second specific example of implementation, the output controller unit 200
is
adapted to cause an audio unit to convey information related to the one or
more
detected target objects. In a specific non-limiting example of implementation,
the
output controller unit 200 generates audio data conveying the presence of the
one or
11
CA 02546296 2006-05-11
more detected target objects, the location of the one or more detected target
objects
within the receptacle 104 and the characteristics of the one or more detected
target
objects.
The output controller unit 200 then releases a signal for causing the output
device 202
to convey the desired information to a user of the system.
The output device 202 may be any device suitable for conveying information to
a user
of the system 100 regarding the presence of one or more target objects in the
receptacle 104. The information may be conveyed in visual format, audio format
or
as a combination of visual and audio formats.
In a first specific example of implementation, the output device 202 includes
a display
screen adapted for displaying in visual format information related to the
presence of
the one or more detected target objects. In a second specific example of
implementation, the output device 202 includes a printer adapted for
displaying in
printed format information related to the presence of the one or more detected
target
objects. Figures 4a and 4b show in simplified form examples of information in
visual
format related to the presence of the one or more detected target objects.
More
specifically, in Figure 4a, the input image generated by the image generation
device
102 is displayed along with a visual indicator (e.g., arrow 404) identifying
the
location of a specific detected target object (e.g., gun 402) detected by the
apparatus
106. Alternatively, in Figure 4b, a text message is provided describing a
specific
detected target object. It will be appreciated that the output device 202 may
produce
additional information without detracting from the spirit of the invention and
that the
examples illustrated in Figures 4a and 4b have been provided for the purpose
of
illustration only.
In a third specific example of implementation, the output device 202 includes
an
audio output unit adapted for releasing an audio signal conveying information
related
to the presence of one or more target objects in the receptacle 104.
12
CA 02546296 2006-05-11
In a fourth specific example of implementation, the output device 202 includes
a set
of visual elements, such as lights or other suitable visual elements, adapted
for
conveying in visual format information related to the presence of one or more
target
objects in the receptacle 104.
The person skilled in the art will readily appreciate, in light of the present
specification, that other suitable types of output devices may be used without
detracting from the spirit of the invention.
The apparatus 106 will now be described in greater detail with reference to
Figure 3.
As depicted, the apparatus 106 includes a first input 310, a second input 314,
an
output 312 and a processing unit. The processing unit comprises a plurality of
functional elements such as a pre-processing module 300, a distortion
correction
module 350, an image comparison module 302 and a detection signal generator
module 306.
The first input 310 is for receiving an image signal 150 associated with the
receptacle
104 from the image generation device 102 (shown in Figure 1). It is recalled
that the
image signal 150 conveys the input image 800 related to the contents of the
receptacle
104. The second input 314 is for receiving data elements from the database
110, such
data elements being associated with target objects. It will be appreciated
that in
embodiments where the database 110 is part of apparatus 106, the second input
314
may be omitted. The output 312 is for releasing the detection signal 160
conveying
the presence of one or more target objects in the receptacle 104. The
detection signal
160 is transmitted towards the output module 108.
Generally speaking, the processing unit of the apparatus 106 receives the
image signal
150 associated with the receptacle 104 from the first input 310 and processes
the
image signal 150 in combination with the data elements associated with target
objects
(received from the database I10 at the second input 314) in an attempt to
detect the
presence of one or more target objects in the receptacle 104. In response to
detection
of one or more target objects (hereinafter referred to as "detected target
objects") in
the receptacle 104, the processing unit of the apparatus 106 generates and
releases at
13
CA 02546296 2006-05-11
the output 312 the detection signal 160 which conveys the presence of the one
or more
detected target objects in the receptacle 104.
The various functional elements of the processing unit of the apparatus 106
implement a process, which is depicted in Figure 5.
Step 500
At step 500, the pre-processing module 300 receives the image signal 150
associated with the receptacle 104 via the first input 310. It is recalled
that the
image signal 150 conveys an input image 800 related to the contents of the
receptacle 104.
Step 501A
At step 501A, the pre-processing module 300 processes the image signal 150
in order to enhance the input image 800 related to the contents of the
receptacle 104, remove extraneous information therefrom and remove noise
artefacts, thereby to help obtain more accurate comparison results later on.
The complexity of the requisite level of pre-processing and the related trade-
offs between speed and accuracy depend on the application. Examples of pre-
processing may include, without being limited to, brightness and contrast
manipulation, histogram modification, noise removal and filtering amongst
others. As part of step 501A, the pre-processing module 300 releases a
modified image signal 170 for processing by the distortion correction module
350 at step 501B. The modified image signal 170 conveys a pre-processed
version of the input image 800 related to the contents of the receptacle 104.
Step 501 B
One recalls at this point that the image generation device 102 may have
introduced distortion into the input image 800 related to the contents of the
receptacle 104. At step 501B, the distortion correction module 350 processes
14
CA 02546296 2006-05-11
the modified image signal 170 in order to remove distortion from the pre-
processed version of the input image 800. The complexity of the requisite
amount of distortion correction and the related trade-offs between speed and
accuracy depend on the application. As part of step 501B, the distortion
correction module 350 releases a corrected image signal 180 for processing by
the image comparison module 302 at step 502. The corrected image signal
180 conveys at least one corrected image related to the contents of the
receptacle 104.
With reference now to Figure 13, distortion correction may be performed by
applying a distortion correction process, which is referred to as TH*-' for
reasons that will become apparent later on. Ignoring for simplicity the effect
of the pre-processing module 300, let the input image 800 be defined by
intensity data for a set of observed coordinates, and let each of a set of one
or
more corrected images 800c be defined by modified intensity data for a set of
new coordinates. Applying the distortion correction process TH*-' may thus
consist of transforming the input image 800 (i.e., the intensity data for the
set
of observed coordinates) in order to arrive at the modified intensity data for
the new coordinates in each of the corrected images 800c.
Assuming that the receptacle 104 were flat (in the Z-direction), one could
model the distortion introduced by the image generation device 102 as a
spatial transformation T on a "true" image to arrive at the input image 800.
Thus, T would represent a spatial transformation that models the distortion
affecting a target object having a given shape and location in the "true"
image,
resulting in that object's "distorted" shape and location in the input image
800.
Thus, to obtain the object's "true" shape and location, it is reasonable to
want
to make the distortion correction process resemble the inverse of T as closely
as possible, so as to facilitate accurate identification of a target object in
the
input image 800. However, not only is T generally unknown in advance, but
moreover it will actually be different for objects appearing at different
heights
within the receptacle 104.
CA 02546296 2006-05-11
More specifically, different objects appearing in the input image 800 may be
distorted to different degrees, depending on the position of those objects
within the input image 800 and depending on the height of those objects
within the receptacle 104 (i.e., the distance between the object in question
and
the image generation device 102). Stated differently, assume that a particular
target object 890 is located at a given height H890 within the receptacle 104.
An image taken of the particular target object 890 will manifest itself as a
corresponding image element 8001 in the input image 800, containing a
distorted version of the particular target object 890. To account for the
distortion of the shape and location of the image element 800, within the
input
image 800, one can still use the spatial transformation approach mentioned
above, but this approach needs take into consideration the height H890 at
which
the particular target object 890 appears within the receptacle 104. Thus, one
can denote the spatial transformation for a given candidate height H by TH,
which therefore models the distortion affects the "true" images of target
objects when such target objects are located at the candidate height H within
the receptacle 104.
Now, although TH is not known, it may be inferred, from which its inverse can
be obtained. The inferred version of TH is denoted TH* and is hereinafter
referred to as an "inferred spatial transformation" for a given candidate
height
H. Basically, TH* can be defined as a data structure that represents an
estimate
of TH. Although the number of possible heights that a target object may
occupy is a continuous variable, it may be possible to granularize this number
to a limited set of "candidate heights" (e.g., such as 5-10) without
introducing
a significant detection error. Of course, the number of candidate heights in a
given embodiment may be as low as one, while the upper bound on the
number of candidate heights is not particularly limited.
The data structure that represents the inferred spatial transformation TH* for
a
given candidate height H may be characterized by a set of parameters derived
from the coordinates of a set of "control points" in both the input image 800
and an "original" image for that candidate height. An "original" image for a
16
CA 02546296 2006-05-11
given candidate height would contain non-distorted images of objects only if
those images appeared within the receptacle 104 at the given candidate height.
Of course, while the original image for a given candidate height is unknown,
it
may be possible to identify picture elements in the input image portion that
are
known to have originated from specific picture elements in the (unknown)
original image. Thus, a "control point" corresponds to a picture element that
occurs at a known location in the original image for a given candidate height
H, and whose "distorted" position can be located in the input image 800.
In one specific non-limiting embodiment, to obtain control points specific to
a
given image generation device 102, and with reference to Figure 14, one can
use a template 1400 having a set of spaced apart holes 1410 at known
locations in the horizontal and vertical directions. The template is placed at
a
given candidate height H1420. One then acquires an input image 1430, from
which control points 1440 (i.e., the holes 1410 present at known locations in
the template) are identified in the input image 1430. This may also be
referred
to as "a registration process". Having performed the registration process on
the input image 1430 that was derived from the template 1400, one obtains
TH1420*, the inferred spatial transformation for the height H1420-
To obtain the inferred spatial transformation TH* for a given candidate height
H, one may utilize a"transformation model". The transformation model that
is used may fall into one or more of the following non-limiting categories,
depending on the type of distortion that is sought to be corrected:
- linear conformal;
- affine;
- projective
- polynomial warping (first order, second order, etc.);
- piecewise linear;
- local weighted mean;
- etc.
17
CA 02546296 2006-05-11
The use of the function cp2tform in the Image Processing Toolbox of Matlab
(available from Mathworks Inc.) is particularly suitable for the computation
of
inferred spatial transformations such as TH* based on coordinates for a set of
control points. Other techniques will now be apparent to persons skilled in
the
art to which the present invention pertains.
The above process can be repeated several times, for different candidate
heights, thus obtaining TH* for various candidate heights. It is noted that
the
derivation of TH* for various candidate heights can be performed off-line,
i.e.,
before scanning of the receptacle 104. In fact, the derivation of TH* is
independent of the contents of the receptacle 104.
Returning now to Figure 13, and assuming that TH* for a given set of
candidate heights has been obtained (e.g., retrieved from memory), one inverts
these transformations and applies the inverted transformations (denoted TH*-a)
to the input image 800 in order to obtain the corrected images 800c. This
completes the distortion correction process.
It is noted that inverting TH* for the various candidate heights yields a
corresponding number of corrected images 800C. Those skilled in the art will
appreciate that each of the corrected images 800c will contain areas of
reduced
distortion where those areas contained objects located at the candidate height
for which the particular corrected image 800c was generated.
It will be appreciated that TH*"' is not always computable in closed form
based
on the corresponding TH*. Nevertheless, the corrected image 800c for the
given candidate height can be obtained from the input image 800 using
interpolation methods, based on the corresponding TH*. Examples of suitable
interpolation methods that may be used include bicubic, bilinear and nearest-
neighbor, to name a few.
The use of the function imtransform in the Image Processing Toolbox of
Matlab (available from Mathworks Inc.) is particularly suitable for the
18
CA 02546296 2006-05-11
computation of an output image (such as one of the corrected images 800c)
based on an input image (such as the input image 800) and an inferred spatial
transformation such as TH*. Other techniques will now be apparent to persons
skilled in the art to which the present invention pertains.
It is noted that certain portions of the corrected image 800c for a given
candidate height might not exhibit less distortion than in the input image
800,
for the simple reason that the objects contained in those portions appeared at
a
different height within the receptacle 104 when they were being scanned.
Nevertheless, if a certain target object was in the receptacle 104, then it is
likely that at least one portion of the corrected image 800c for at least one
candidate height will show a reduction in distortion with respect to
representation of the certain target object in the input image 800, thus
facilitating comparison with data elements in the database 110 as described
later on.
Naturally, the precise numerical values in the transformations used in the
selected distortion correction technique may vary from one image generation
device 102 to the next, as different image generation devices introduce
different amounts of distortion of different types, which appear in different
regions of the input image 800.
Of course, those skilled in the art will appreciate that similar reasoning and
calculations apply when taking into account the effect of the pre-processing
module 300, the only difference being that one would be dealing with
observations made in the pre-processed version of the input image 800 rather
than in the input image 800 itself.
It will also be appreciated that the functionality of the pre-processing
module
300 and the distortion correction module 350 can be performed in reverse
order. In other embodiments, all or part of the functionality of the pre-
processing module 300 and/or the distortion correction module 350 may be
external to the apparatus 106, e.g., such functionality may be integrated with
19
CA 02546296 2008-10-31
the image generation device 102 or performed by external components. It will
also be appreciated that the pre-processing module 300 and/or the distortion
correction module 350 (and hence steps 501A and/or 501B) may be omitted in
certain embodiments of the present invention without detracting from the
spirit
of the invention.
Step 502
At step 502, the image comparison module 302 verifies whether all data
elements in the database 110 have been processed. In the negative, the image
comparison module 302 proceeds to step 503 where the next one of the data
elements is accessed and the image comparison module 302 then proceeds to
step 504. If at step 502 all data elements in the database 110 have been
processed, the image comparison module 302 proceeds to step 508 and the
process is completed.
Step 504
Assuming for the moment that the data elements in the database 110 represent
images of target objects, the data element accessed at step 503 conveys a
particular image of a particular target object. Thus, at step 504, the image
comparison module 302 effects a comparison between at least one corrected
image related to the contents of the receptacle 104 (which are conveyed in the
corrected image signal 180) and the particular image of the particular target
object to determine whether a match exists. It is noted that more than one
corrected image may be provided, namely when more than one candidate
height is accounted for. The comparison may be effected using any image
processing algorithm suitable for comparing two images. Examples of
algorithms that can be used to perform image processing and comparison
include without being limited to:
A- ENHANCEMENT: Brightness and contrast manipulation;
Histogram modification; Noise removal; Filtering.
CA 02546296 2006-05-11
B- SEGMENTATION: Thresholding; Binary or multilevel; Hysteresis
based; Statistics/histogram analysis; Clustering; Region growing;
Splitting and merging; Texture analysis; Watershed; Blob labeling;
C- GENERAL DETECTION: Template matching; Matched filtering;
Image registration; Image correlation; Hough transform;
D- EDGE DETECTION: Gradient; Laplacian;
E- MORPHOLOGICAL IMAGE PROCESSING: Binary; Grayscale;
F- FREQUENCY ANALYSIS: Fourier Transform; Wavelets;
G- SHAPE ANALYSIS AND REPRESENTATIONS: Geometric
attributes (e.g. perimeter, area, euler number, compactness); Spatial
moments (invariance); Fourier descriptors; B-splines; Chain codes;
Polygons; Quad tree decomposition;
H- FEATURE REPRESENTATION AND CLASSIFICATION:
Bayesian classifier; Principal component analysis; Binary tree; Graphs;
Neural networks; Genetic algorithms; Markov random fields.
The above algorithms are well known in the field of image processing and as
such will not be described further here.
In a specific example of implementation, the image comparison module 302
includes an edge detector to perform part of the comparison at step 504.
In another specific example of implementation, the comparison performed at
step 504 includes effecting a "correlation operation" between the at least one
corrected image related to the contents of the receptacle 104 (which are
conveyed in the corrected image signal 180) and the particular image of the
21
CA 02546296 2008-10-31
particular target object. Again, it is recalled that when multiple candidate
heights are accounted for, then multiple corrected images may need to be
processed, either serially, in parallel or a combination thereof.
In a specific non-limiting embodiment, the correlation operation involves
computing the Fourier transform of the at least one corrected image related to
the contents of the receptacle 104 (which are conveyed in the corrected image
signal 180), computing the Fourier transform complex conjugate of the
particular image of the particular target object, multiplying the two Fourier
transforms together and then taking the Fourier transform (or inverse Fourier
transform) of the product. Simply put, the result of the correlation operation
provides a measure of the degree of similarity between the two images.
In a specific example of implementation, the correlation operation is
performed by an optical correlator. A specific example of implementation of
an optical correlator suitable for use in comparing two images will be
described later on in the specification. In an alternative example of
implementation, the correlation operation is performed by a digital
correlator.
The image comparison module 302 then proceeds to step 506.
Step 506
The result of the comparison effected at step 504 is processed to determine
whether a match exists between (I) at least one of the at least one corrected
image 800C related to the contents of the receptacle 104 and (II) the
particular
image of the particular target object. In the absence of a match, the image
comparison module 302 returns to step 502. However, in response to
detection of a match, it is concluded that the particular target object has
been
detected in the receptacle and the image comparison module 302 triggers the
detection signal generation module 306 to execute step 510. Then, the image
comparison module 302 returns to step 502 to continue processing with
respect to the next data element in the database 110.
22
CA 02546296 2006-05-11
Step 510
At step 510, the detection signal generation module 306 generates the
aforesaid detection signal 160 conveying the presence of the particular target
object in the receptacle 104. The detection signal 160 is released via the
output 312. The detection signal 160 may simply convey the fact that the
particular target object has been detected as present in the receptacle 104,
without necessarily specifying the identity of the particular target object.
Alternatively, the detection signal 160 may convey the actual identity of the
particular target object. As previously indicated, the detection signal 160
may
include information related to the positioning of the particular target object
within the receptacle 104 and optionally a target object identifier data
element
associated with the particular target object.
It should be noted that generation of the detection signal 160 may also be
deferred until multiple or even all of the data elements in the database 110
have been processed. Accordingly, the detection signal may convey the
detection of multiple particular target objects in the receptacle 104 and/or
their
respective identities.
It will be appreciated that the correlation operation may be implemented using
a
digital correlator. The correlation operation is computationally intensive
and, in
certain implementations requiring real-time performance, the use of a digital
correlator may not provide suitable performance. Under such conditions, an
optical
correlator may be preferred.
Advantageously, an optical correlator performs the correlation operation
physically
through light-based computation, rather than by using software running on a
silicon-
based computer, which allows computations to be performed at a higher speed
than is
possible with a software implementation and thus provides for improved real-
time
performance. Specific examples of implementation of the optical correlator
include a
23
CA 02546296 2006-05-11
joint transform correlator (JTC) and a focal plane correlator (FPC). Two
specific non-
limiting embodiments of a suitable optical correlator are shown in Figures 8A
and 8C.
In a first embodiment, now described with reference to Figure 8A, the optical
correlator effects a Fourier transformation 840 of a given corrected image
related to
the contents of the receptacle 104. Also, the optical correlator effects a
complex
conjugate Fourier transformation 840' of a particular image 804 of a
particular target
object obtained from the database 110. Image processing and enhancement, as
well as
distortion pre-emphasis, can also be performed on the particular image 804 to
obtain
better matching performance depending on the environment and application. The
result of the two Fourier transformations is multiplied 820. The optical
correlator
then processes the result of the multiplication of the two Fourier transforms
by
applying another optical Fourier transform (or inverse Fourier transform) 822.
The
resulting signal is captured by a camera (or charge coupled device) 912 at
what is
referred to as the correlation plane, which yields the correlation output,
shown at
Figure 8B. The correlation output is released for transmission to the
detection signal
generator 306 where it is analyzed. A peak in the correlation output (see
Figure 8B)
indicates a match between the input image 800 related to the contents of the
receptacle 104 and the particular image 804 of the particular target object.
The result
of the processing is then conveyed to the user by output module 108.
In a second embodiment, now described with reference to Figure 8C, the data
elements in the database I 10 include data indicative of the Fourier transform
of the
images of the target objects that the system 100 is designed to detect. Such
data
elements will be referred to herein after as "templates" (or "filters") for
particular
target objects. In non-limiting examples of implementation, the templates (or
filters)
are digitally pre-computed such as to improve the speed of the correlation
operation
when the system 100 is in use. Image processing and enhancement, as well as
distortion pre-emphasis, can also be performed on the image of a particular
target
object to obtain better matching performance depending on the environment and
application.
24
CA 02546296 2008-10-31
In this second embodiment, the data element accessed at step 503 conveys a
particular
template (or filter) 804' for a particular image 804. Thus, in a modified
version of
step 504, and with continued reference to Figure 8C, the image comparison
module
302 implements an optical correlator for effecting a Fourier transformation
840 of a
given corrected image related to the contents of the receptacle 104. The
result is
multiplied 820 with the (previously computed) template (or filter) 804' for
the
particular image 804, as accessed from the database 110. The optical
correlator then
processes the product by applying the optical Fourier transform (or inverse
Fourier
transform) 822. The resulting signal is captured by the camera 912 at the
correlation
plane, which yields the correlation output, shown at Figure 8B. The
correlation
output is released for transmission to the detection signal generator 306
where it is
analyzed. A peak in the correlation output (see Figure 8B) indicates a match
between
the input image 800 related to the contents of the receptacle 104 and the
template (or
filter) 804' for the particular image 804. The result of the processing is
then conveyed
to the user by output module 108.
The content and format of the database 110 may be further varied from one
implementation to the next, and the skilled person in the art will readily
appreciate in
light of the present description that other approaches to generating templates
(or
filters) may be used without detracting from the spirit of the invention.
Many methods for generating filters are known in the art and a few such
methods will
be described later on in the specification. The reader is invited to refer to
the
following document for additional information regarding phase only filters
(POF):
Phase-Only Matched Filtering, Joseph L. Homer and Peter D. Gianino, Appl. Opt.
Vol. 23 no. 6, 15 March 1994, pp.812-816.
In a non-limiting example of implementation, the generation of the template
(or filter)
is performed in a few steps. First, the background is removed from the
particular
image of the particular target object. In other words, the particular image is
extracted
from the background and the background is replaced by a black background. The
resulting image is then processed through a Fourier transform function. The
result of
CA 02546296 2008-10-31
this transform is a complex image. A phase only filter (POF) for example will
only
contain the complex conjugate of the phase information (between zero and 2 pi)
which is mapped to a 0 to 255 range values. These 256 values correspond in
fact to
the 2561evels of gray of an image.
As a variant, in order to reduce the amount of data needed to represent the
whole
range of 3D orientations that a single target object can take, a MACE (Minimum
Average Correlation Energy) filter is used to generate a template (or filter)
for a given
target object. Typically, the MACE filter combines several different 2D
projections of
a given object and encodes them in a single MACE filter instead of having one
2D
projection per filter. One of the benefits of using MACE filters is that the
resulting
database 110 would take less space since it would include fewer items. Also,
since the
number of correlation operations needed to identify a single target object
would be
reduced, the total processing time to determine whether a given object is
present
would also be reduced. The reader is invited to refer to the following
document for
additional information regarding MACE filters: Mahalanobis, A., B.V.K. Vijaya
Kumar, and D. Casasent (1987); Minimum Average Correlation Energy Filters,
Appl.
Opt. 26 no. 17, 3633-3640.
Another way of reducing the processing time of the correlation operation is to
take
advantage of the linear properties of the Fourier transform. By dividing the
particular
image into several sub-images, a composite image can be formed, herein
referred to as
a mosaic. When a mosaic is displayed at the input of the correlator, the
correlation is
computed simultaneously on all the sub-images without incurring any
substantial time
penalty. A mosaic may contain several different target objects or several
different
orientations of the same target object or a combination of both. Figure 7
depicts a
mosaic including a target object in various orientations and scales. The
parallel
processing capabilities of a mosaic effectively increase the throughput of an
optical
correlator. The reader is invited to refer to the following document for
additional
information regarding the use of a mosaic in an optical correlator: Method and
apparatus for evaluating a scale factor and a rotation angle in image
processing,
26
CA 02546296 2008-10-31
Alain Bergeron et al., United States Patent, no. 6,549,683, April 15, 2003.
The inner workings of the aforementioned non-limiting example optical
correlator are
illustrated in Figure 9. On the left hand side appears a laser source 900 that
generates
a collimated coherent light beam used to project images across the correlator.
The
light beam is directed first through a small set of lenses 902 used to expand
its
diameter in order to illuminate, in parallel, the whole surface of a first
liquid crystal
display (LCD) screen 904. The input image 800 related to the contents of the
receptacle 104 is displayed on the first LCD screen 904 either through a
direct camera
interface or provided as a VGA image by a computing device. The first LCD
screen
904 is illuminated by the light beam and the image is propagated through the
correlator. In the illustrated example, the input image 800 related to the
contents of
the receptacle 104, which is captured by the camera is that of a gun on a
conveyor
belt.
The light beam modulated by the first image on the first LCD screen 904 is
then
propagated through a second set of lenses 906, referred to as a Fourier lens
since it
performs the equivalent of the Fourier transform mathematical operation. The
inherent properties of light are used to physically perform the appropriate
calculations. Specifically, the propagation of light is a function which
corresponds to
the kernel of the Fourier transform operation, thus the propagation of light
along the
axis of a Fourier lens represents a sufficiently strong approximation of this
natural
phenomenon to assert that the light beam undergoes a Fourier transform.
Otherwise
stated, a lens has the inherent property of performing a Fourier transform on
images
observed at its front focal plane, provided that this image is displayed at
its back focal
plane. The Fourier transform, which can normally be rather computation-
intensive
when calculated by a digital computer, is performed in the optical correlator
simply
by the propagation of the light. The mathematics behind this optical
realization is
equivalent to the exact Fourier transform function and can be modeled with
standard
fast Fourier algorithms. For more information regarding Fourier transforms,
the
reader is invited to consider B.V.K. Vijaya Kumar, Marios Savvides, Krithika
Venkataramani,and Chunyan Xie , "Spatial frequency domain image processing for
27
CA 02546296 2008-10-31
biometric recognition", Biometrics ICIP Conference 2002 or alternatively J. W.
Goodman, Introduction to Fourier Optics, 2nd Edition, McGraw-Hill, 1996.
After going through the Fourier lens 906, the signal is projected on a second
LCD
screen 908 on which is displayed the target template (or filter) 804' (i.e.,
Fourier
transform) for the particular image 804. When the Fourier transform of the
image
associated with the receptacle 104 goes through the second LCD screen 908 on
which
the target template (or filter) 804' is displayed, the light beam crosses a
second
Fourier lens 910 which, again, optically computes the equivalent of a Fourier
transform multiplication. This operation corresponds to a correlation in the
spatial
domain. The template (or filter) 804' displayed on the second LCD screen 908
in fact
induces a phase variation on the incoming light beam. Each pixel can
potentially
induce a phase change whose magnitude is equivalent to its gray level. As such
the
Fourier transform displayed on the first LCD screen 904 is multiplied with the
Fourier
transform of the template (or filter) 804' for the particular image 804, which
is
equivalent to performing a correlation.
The second Fourier lens 910 finally concentrates the light beam on a small
area
camera or camera 912 where the result of the correlation is measured, so to
speak.
The camera 912 in fact measures energy peaks on the correlation plane. The
position
of a correlation peak corresponds in fact to the location of the target object
center in
the input image 800.
Referring back to Figure 8C, the camera 912 communicates the signal from the
optical correlator to the detection signal generator module 306. In this
specific
implementation, the detection signal generator module 306 is a computing unit
including a frame grabber and software. The software is adapted to processing
the
signal received from the camera 912 to detect energy peaks as gray level video
signals
varying between 0 and 255. A strong intensity peak on the correlation plane
indicates
a match between the input image 800 related to the contents of the receptacle
104 and
the particular image 804 on which the particular template 804' is based. The
location
28
CA 02546296 2006-05-11
of the energy peak also indicates the location of the center of the particular
image 804
in the input image 800 related to the contents of the receptacle 104.
Fourier Transform and Spatial Frequencies
The Fourier transform as applied to images will now be described in general
terms.
The Fourier transform is a mathematical tool used to convert the information
present
within an object's image into its frequency representation. In short, an image
can be
seen as a superposition of various spatial frequencies and the Fourier
transform is a
mathematical operation used to compute the intensity of each of these
frequencies
within the image. The spatial frequencies represent the rate of variation of
image
intensity in space. Consequently, a smooth or uniform pattern mainly contains
low
frequencies. Sharply contoured patterns, by contrast, exhibit a higher
frequency
content.
The Fourier transform of an image f(x,y) is given by:
F(u,v)- fff (x,Y)e .i2R(ux+vy)dXdJ' (1)
where u, v are the coordinates in the frequency domain. Thus, the Fourier
transform is
a global operator: changing a single frequency of the Fourier transform
affects the
whole object in the spatial domain.
A correlation operation can be mathematically described by:
~00
C(, ~) - f J f(x,Y)h'" (x - s, Y-~)dxdY (2)
_00_00
where s and ~ represent the pixel coordinates in the correlation plane, C(s,
~) stands for
the correlation, x and y identify the pixel coordinates of the input image,
f(x, y) is the
original input image and h*(E, ~) is the complex conjugate of the correlation
filter.
In the frequency domain the same expression takes a slightly different form:
C(s, ~) = s -1 (F(u, v)H' (3)
29
CA 02546296 2006-05-11
where 3 is the Fourier transform operator, u and v are the pixel coordinates
in the
Fourier plane, F(u,v) is the Fourier transform complex conjugate of the image
acquired with the camera f(x,y) and H*(u,v) is the Fourier transform of the
template
(or filter). Thus, the correlation between an input image and a template (or
filter) is
equivalent, in mathematical terms, to the multiplication of their respective
Fourier
transforms, provided that the complex conjugate of the template (or filter) is
used.
Consequently, the correlation can be defined in the spatial domain as the
search for a
given pattern (template/filter), or in the frequency domain, as filtering
operation with
a specially designed matched filter.
Advantageously, the use of optics for computing a correlation operation allows
the
computation to be performed in a shorter time than by using a digital
implementation
of the correlation. It turns out that an optical lens properly positioned
(i.e. input and
output images are located on the lens's focal planes) automatically computes
the
Fourier transform of the input image. In order to speed up the computation of
the
correlation, the Fourier transform of a particular image can be computed
beforehand
and submitted to the correlator as a template (or filter). This type of filter
is called a
matched filter.
Figure 10 depicts the Fourier transform of the spatial domain image of a
number `2'.
It can be seen that most of the energy (bright areas) is contained in the
central portion
of the Fourier transform image which correspond to low spatial frequencies
(the
images are centred on the origin of the Fourier plane). The energy is somewhat
more
dispersed in the medium frequencies and is concentrated in orientations
representative
of the shape of the input image. Finally, little energy is contained in the
upper
frequencies. The right-hand-side image shows the phase content of the Fourier
transform. The phase is coded from black (0 ) to white (360 ).
Generation of Templates (or Filters)
Matched filters, as their name implies, are specifically adapted to respond to
one
image in particular: they are optimized to respond to an object with respect
to its
energy content. Generally, the contour of an object corresponds to its high
frequency
CA 02546296 2008-10-31
content. This can be easily understood as the contour represent areas where
the
intensity varies rapidly (hence a high frequency).
In order to emphasize the contour of an object, the matched filter can be
divided by its
module (the image is normalized), over the whole Fourier transform image. The
resulting filter is called a Phase-Only Filter (POF) and is defined by:
POF(u,v)= H*(u'v) (4)
IH"(u,v)
The reader is invited to refer to the following document for additional
information
regarding phase only filters (POF): "Phase-Only Matched Filtering", Joseph L.
Homer and Peter D. Gianino, Appl. Opt. Vol. 23 no. 6, 15 March 1994, pp.812-
816.
Because these filters are defined in the frequency domain, normalizing over
the whole
spectrum of frequencies implies that each of the frequency components is
considered
with the same weight. In the spatial domain (e.g. usual real-world domain),
this means
that the emphasis is given to the contours (or edges) of the object. As such,
the POF
filter provides a higher degree of discrimination, sharper correlation peaks
and higher
energy efficiency.
The discrimination provided by the POF filter, however, has some
disadvantages. It
turns out that, although the optical correlator is somewhat insensitive to the
size of the
objects to be recognized, the images are expected to be properly sized,
otherwise the
features might not be registered properly. To understand this requirement,
imagine a
filter defined out of a given instance of a2'. If that filter is applied to a
second
instance of a`2' whose contour is slightly different, the correlation peak
will be
significantly reduced as a result of the great sensitivity of the filter to
the original
shape. A different type of filter, termed a composite filter, was introduced
to
overcome these limitations. The reader is invited to refer to the following
document
for additional information regarding this different type of composite filter:
H.J.
Caufield and W. T. Maloney, Improved Discrimination in Optical Character
Recognition, Appl. Opt., 8, 2354, 1969.
31
CA 02546296 2008-10-31
In accordance with specific implementations, filters can be designed by:
- Appropriately choosing one specific instance (because it represents
characteristics
which are, on average, common to all symbols of a given class) of a symbol and
calculating from that image the filter against which all instances of that
class of
symbols will be compared; or
- Averaging many instances of a given to create a generic or `template' image
from
which the filter is calculated. The computed filter is then called a composite
filter
since it incorporates the properties of many images (note that it is
irrelevant
whether the images are averaged before or after the Fourier transform operator
is
applied, provided that in the latter case, the additions are performed taking
the
Fourier domain phase into account).
The latter procedure forms the basis for the generation of composite filters.
Thus
composite filters are composed of the response of individual POF filters to
the same
symbol. Mathematically, this can be expressed by:
hcomp(x,Y)=aohp(x,Y)+aehn(x,Y)+...+azhx(x,Y) (5)
A filter generated in this fashion is likely to be more robust to minor
signature
variations as the irrelevant high frequency features will be averaged out. In
short, the
net effect is an equalization of the response of the filter to the different
instances of a
given symbol.
Composite filters can also be used to reduce the response of the filter to the
other
classes of symbols. In equation (5) above, if the coefficient b, for example,
is set to a
negative value, then the filter response to a symbol of class b will be
significantly
reduced. In other words, the correlation peak will be high if ha(x,y) is at
the input
image, and low if hb(x,y) is present at the input. A typical implementation of
composite filters is described in: Optical Character Recognition (OCR) in
Uncontrolled Environments Using Optical Correlators, Andre Morin, Alain
Bergeron, Donald Prevost and Ernst A. Radloff, Proc. SPIE Int. Soc. Opt. Eng.
3715,
346 (1999).
32
CA 02546296 2008-10-31
Those skilled in the art will appreciate that the concepts described above can
also be
readily applied to the screening of people. For example, in an alternative
embodiment, a system for screening people is provided. The system includes
components similar to those described in connection with the system depicted
in
Figure 1. In a specific example of implementation, the image generation device
102
is configured to scan a person and possibly to scan the person along various
axes to
generate multiple images associated with the person. The image(s) associated
with
the person convey information related to the objects carried by the person.
Figure 11
depicts two images associated with a person suitable for use in connection
with a
specific implementation of the system. Each image is then processed in
accordance
with the method described in the present specification to detect the presence
of target
objects on the person.
Those skilled in the art will appreciate that certain portions of the
apparatus 106 can
be implemented on a general purpose digital computer 1300, of the type
depicted in
Figure 12, including a processing unit 1302 and a memory 1304 connected by a
communication bus. The memory includes data 1308 and program instructions
1306.
The processing unit 1302 is adapted to process the data 1308 and the program
instructions 1306 in order to implement the functional blocks described in the
specification and depicted in the drawings.. The digital computer 1300 may
also
comprise an I/O interface 1310 for receiving or sending data elements to
external
devices.
Alternatively, the above-described apparatus 106 can be implemented on a
dedicated
hardware platform where electrical/optical components implement the functional
blocks described in the specification and depicted in the drawings. Specific
implementations may be realized using ICs, ASICs, DSPs, FPGAs, an optical
correlator, digital correlator or other suitable hardware platform.
33
CA 02546296 2006-05-11
In a specific example of implementation, the optical correlator suitable for
use in the
system described herein includes a video input and a digital input. The video
input is
suitable for receiving a signal derived from an image generation device and
the digital
input is suitable for receiving a signal derived from images in a database. In
a specific
implementation, the video input is suitable for receiving a signal in an NTSC
compatible format and the digital input suitable for receiving a signal in a
VGA
compatible format. It will be appreciated that the digital input suitable for
receiving a
signal in a VGA compatible format may be replaced by any other suitable
digital
input interface adapted for receiving signals of lower or higher resolution
that the
VGA compatible format signal. Similarly, it will also be appreciated that the
video
input suitable for receiving a signal in an NTSC compatible format may be
replaced
by any other suitable analog or digital video signal interface suitable for
receiving
signals in suitable formats such as, but not limited to, PAL and SECAM. In a
non-
limiting implementation, the optical correlator is adapted to process an image
received
at the video input having an area of 640x480 pixels. However, it will be
readily
apparent that, by providing suitable interfaces, larger or smaller images can
be
handled since the optical correlator's processing capability is independent of
the size
of the image, as opposed to digital systems that require more processing time
and
power as images get larger. In yet another alternative implementation, the
video input
is replaced by a second digital input adapted for receiving an image signal in
any
suitable digital image format. In such an implementation, the image generation
device 102 (shown in figure 1) generates a digital format image and
communicates the
latter to the apparatus 106. Advantageously, by providing a digital machine-to-
machine exchange of images between the image generation device 102 and the
apparatus 106, the digital image generated by the image generation device 102
can be
processed directly without the requirement of effecting an analog-to-digital
conversion at the apparatus 106.
In a variant, a single optical correlator can be shared by multiple image
generation
devices. In such a variant, conventional parallel processing techniques can be
used
for sharing a common hardware resource.
34
CA 02546296 2006-05-11
Although the present invention has been described in considerable detail with
reference to certain preferred embodiments thereof, variations and refinements
are
possible without departing from the spirit of the invention. Therefore, the
scope of
the invention should be limited only by the appended claims and their
equivalents.
