WO2004084120A1

WO2004084120A1 - Biometric verification apparatus and method

Info

Publication number: WO2004084120A1
Application number: PCT/GB2004/001154
Authority: WO
Inventors: Tej Paul Kaushal; David Thomas Sheerin; Richard Home
Original assignee: Qinetiq Limited
Priority date: 2003-03-20
Filing date: 2004-03-18
Publication date: 2004-09-30
Also published as: GB0306356D0

Abstract

A method for checking the identity of an individual comprising the steps of acquiring a thermal image of the individual (4), processing the image to extract low spatial frequency information therefrom, converting the extracted low spatial frequency information into a numerical data-string characteristic of said individual, and comparing the numerical data-string with at least one reference numerical data-string in order to verify or otherwise the identity of the individual. In particular, the processing step uses a plurality of frequency domain filters applied (8) to a Fourier transform (6) of the thermal image to select spatial frequencies of interest corresponding with preferred orientations in the thermal image. The selected spatial frequencies for each preferred orientation are summed (10) and the results for all preferred orientations collated (12) to produce the numerical data-string having the form of a sequence of single integer digits. An apparatus (30) for checking the identity of an individual using the above method.

Description

Biometric Verification Apparatus and Method

The present invention relates to an apparatus and method for verifying the identity of an individual using biometric data in the form of a thermal image of said individual.

Identification, using thermal imaging, of individuals is an established biometric technique, see for example US Patent 5,163,094. The identification method described in US 5,163,094 consists of generating a thermal image of a portion of an individual's body (typically part of the individual's face), producing a digital image (a facial thermogram) comprising thermal contours corresponding with unique structural features of the individual, and comparing the digital image with a stored reference image.

The thermal patterns discernible in the facial thermogram derive primarily from the pattern of superficial blood vessels under the individual's skin. The identification technique described in US Patent 5,163,094 relies on the fact that the complexity and vastness of the layout of blood vessels in the head and face assures that each person's vascular arrangement is unique.

Elemental shapes can be identified in the thermal contours of the facial thermogram (e.g. nestings of thermal contours) which are characteristic of a particular individual and which are unique to that individual. Such elemental shapes may be used to identify an individual. This is analogous to identifying a person from a fingerprint, a process in which two sets of elemental shapes may be considered to identify a person if a significant number of the elemental shapes in the two sets have corresponding positions and characteristics. Indeed, it has been suggested that, in the fullness of time, thermal facial imagery may replace conventional fingerprint identification techniques.

Alternative thermal identification techniques utilise conventional pattern processing methods to recognise an individual from a thermal facial image. Conventional pattern processing techniques include neural networks and eigenvector analysis (see for example Turk and Pentland, Eigenfaces for recognition, J. Cognitive Science, vol. 3, no. 1 , 1991). Conventional thermal identification techniques, such as those described above, are capable of identifying individuals reliably and with low error rates. However, such systems require thermal images having high spatial resolution and high thermal sensitivity in order to provide reliable identification. This has serious implications for the technical specification and cost of the identification apparatus. For example, a thermal imager capable of acquiring images having high spatial resolution and high thermal sensitivity is complex and expensive. Furthermore, processing of said thermal images is computationally intensive in terms of computer memory and computer processor time / specification, leading to sophisticated and expensive hardware and software.

Furthermore, traditional methods of facial recognition which utilise an eigenvector approach or neural networks require prior knowledge of the face to be identified in the form of an extensive set of reference and training images.

It is an object of the present invention to provide an alternative apparatus and method for verifying the identity of an individual.

According to a first aspect of the present invention, a method for checking the identity of an individual comprises the steps of

(i) acquiring a thermal image of a portion of the individual's body,

(ii) processing the thermal image to extract only low spatial frequency information therefrom,

(iii) converting the extracted low spatial frequency information into a numerical data- string characteristic of said individual, and

(iv) comparing the numerical data-string with at least one reference numerical data- string in order to verify or otherwise the identity of the individual.

The numerical data-string or bioPIN may comprise a numerical sequence of integer digits similar to a conventional Personal Identification Number (PIN). Alternatively, the numerical data-string may comprise a sequence of floating point numbers, alpha- numerical characters, decimal numbers, hexadecimal-numbers, binary numbers, graphical symbols or a barcode.

Unlike conventional thermal identification techniques, the method of the present invention utilises the low spatial frequency information conveyed in a thermal image to verify the identity of the individual portrayed in the thermal image. Hitherto, the low spatial frequency component of thermal images has been considered to convey little or no useful information and hence has been largely disregarded in terms of personal identification applications. Indeed, conventional thermal identification techniques increasingly rely on high spatial frequency information conveyed in an image to provide reliable discrimination.

In the interests of clarity, it should be noted that the low spatial frequency information in the thermal image is associated with coarse features in the thermal image. For example, where the portion of the individual's body depicted in the thermal image comprises the individual's head, the low spatial frequency information in the thermal image would relate to the approximate shape of the individual's head. Any high spatial frequency information in the thermal image, associated with fine details in the thermal image, is substantially ignored by the method of the present invention. Hence, where the portion of the individual's body depicted in the thermal image comprises the individual's head, fine details, such as the shape and layout of facial blood vessels, are ignored by the present method.

Preferably, the method comprises the steps of determining the collective power in a plurality of spatial frequencies of interest within the thermal image, in at least two orientations within the thermal image, converting the collective power for each of the at least two orientations into a numerical character, and arranging the numerical characters into a sequence to form the numerical data-string characteristic of the individual.

For the purposes of this specification, the term "numerical character" shall be defined as including the following; a decimal number, a cardinal number, a floating point number, an integer, an alpha-numerical character, a hexadecimal-number, a binary number, a graphical symbol and a barcode.

In a preferred embodiment, the step of processing the thermal image comprises the steps of

(i) applying a Fourier transformation to the thermal image to produce a symmetrical Fourier transform of the thermal image,

(ii) determining the modulus of the Fourier transform,

(iii) multiplying the modulus of the Fourier transform by a plurality of filters, each filter arranged to select spatial frequencies in a spatial frequency range within the Fourier transform corresponding with a different orientation in the thermal image,

(iv) summing the products from the multiplication step for each filter to produce a resultant sum for each filter.

The transformation of the thermal image from the spatial domain to the frequency domain, using a Fourier transformation, facilitates the extraction of the low spatial frequency information from the thermal image. The modulus of the symmetrical Fourier transform of the thermal image may simply be multiplied by each of the plurality of filters to extract the low spatial frequency information occurring within a plurality of orientations within the thermal image.

The filters may comprise a plurality of elements, each element in the filter corresponding with a pixel or a group of pixels in the Fourier transform. Advantageously, the step of converting the extracted low spatial frequency information into a numerical data-string comprises the steps of

(i) normalising the sum for each filter,

(ii) re-scaling the normalised sum for each filter with respect to a predefined range to form a numerical output for each filter comprising a numerical character,

(iii) arranging the numerical outputs from the filters into a sequence to form the numerical data-string characteristic of the individual.

The predefined range for a particular filter may be determined by analysing the normal variation of the sum outputs from that particular filter for a given sample population of individuals.

The filters may comprise a first filter arranged to select spatial frequencies in the Fourier transform in a first orientation and a second filter arranged to select spatial frequencies in the Fourier transform in a second orientation, the first and second orientations being substantially orthogonal to each other.

The orientation of the filters may be arranged to select spatial frequencies occurring in substantially horizontal and vertical directions in the original thermal image. This is advantageous in that the relative width and height of the portion of the individual's body depicted in the thermal image may be determined. Accordingly, the approximate shape of the portion of the individual's body may be determined. Alternatively, the filters may be arranged to select spatial frequencies occurring in substantially diagonal directions in the original thermal image. The approximate shape of the portion of the individual's body may be similarly determined.

Advantageously, the filters comprise a third filter arranged to select spatial frequencies in the Fourier transform in a third orientation aligned at an angle of substantially 45 degrees to the first orientation and a fourth filter arranged to select spatial frequencies in the Fourier transform in a fourth orientation aligned at an angle of substantially 45 degrees to the second orientation, the third and fourth orientations being substantially orthogonal to each other.

The orientation of the filters may be arranged to select spatial frequencies occurring in substantially horizontal, vertical and diagonal directions in the original thermal image. Again, this is advantageous in that the approximate shape of the portion of the individual's body depicted in the thermal image may be determined. The use of four filters provides an improved approximation of the shape of the portion of the individual's body depicted in the thermal image.

Conveniently, the portion of the individual's body comprises the individual's head.

Preferably, the numerical data-string comprises at least two numerical characters.

Advantageously, the numerical data-string comprises at least four numerical characters.

Preferably, each numerical character comprises one of a decimal number, a floating point number, an integer, an alpha-numerical character, a hexadecimal-number, a binary number, a graphical symbol and a barcode.

Conveniently, the reference numerical data string is input by the individual to be identified as part of the comparison step.

The method may further comprise the step of providing an output indicative of the result of the comparison of the numerical data-string with the reference numerical data-string. Advantageously the thermal image contains not more than 4096 pixels. The thermal image may be configured as an array of 64 x 64 pixels.

Preferably, the smallest temperature difference resolvable in the thermal image is 200 x10^"3 Kelvin or less.

According to a second aspect of the present invention, there is now proposed an automated teller machine (ATM) having a thermal imaging camera and a processor, the processor configured to check the identity of an individual using the method according to the first aspect of the present invention.

According to a third aspect of the present invention, there is now proposed an electronic point-of-sale (EPOS) terminal having a thermal imaging camera and a processor, the processor configured to check the identity of an individual using the method according to the first aspect of the present invention.

The use of the foregoing method within an ATM or EPOS terminal has potential advantages in terms of reduced fraudulent transactions, whilst adding minimal overheads in terms of hardware, software and network communications.

According to a fourth aspect of the present invention, an apparatus for checking the identity of an individual from a thermal image of a portion of the individual's body, comprises,

an image processor arranged to process the thermal image to extract low spatial frequency information therefrom,

means for converting the extracted low spatial frequency information into a numerical data-string characteristic of said individual, and

means for comparing the numeπcal data-string with at least one reference numerical data-string in order to verify or otherwise the identity of the individual. In a preferred embodiment, the image processing means comprises

means for applying a Fourier transformation to the thermal image, said means being adapted to produce a symmetrical Fourier transform of the thermal image and to determine the modulus of said Fourier transform,

multiplication means arranged to multiply the modulus of the Fourier transform by a plurality of filters, each filter being arranged to select spatial frequencies in a spatial frequency range within the Fourier transform corresponding with a different orientation in the thermal image, and

means configured to sum the products from the multiplication means for each filter, said means being arranged to produce a resultant sum for each filter.

Advantageously, the means for converting the extracted low spatial frequency information into a numerical data-string comprises

means for normalising the sum for each filter,

means for re-scaling the normalised sum for each filter with respect to a predefined range configured to calculate a numerical output for each filter comprising a numerical character, and

means for arranging the numerical outputs from the filters into a sequence comprising the numerical data-string characteristic of the individual.

In a preferred embodiment, the apparatus further comprises output means arranged to provide an output indicative of a result of the comparison of the numerical data- string with the reference numerical data-string.

The apparatus may further comprise a reference repository suitable for at least one reference numeπcal data-string.

Preferably, the apparatus comprises an input means for input, in use, of a reference data string by the individual to be identified.

Conveniently, the apparatus further comprises means adapted to read the reference data string from a removable storage means. The removable storage means may comprise one of an identification card, a credit card and a debit card.

According to a fifth aspect of the present invention there is now proposed an automated teller machine (ATM) having a thermal imaging camera and a processor, the processor comprising an apparatus according to the fourth aspect of the present invention. According to a sixth aspect of the present invention there is now proposed an electronic point-of-sale (EPOS) terminal comprising an apparatus according to the fourth aspect of the present invention.

According to another aspect of the present invention, a method for generating a biometric personal identification number characteristic of an individual, from a thermal image of a portion of the individual's body comprises the steps of

(i) processing the thermal image to extract only low spatial frequency information therefrom,

(ii) converting the extracted low spatial frequency information into a numerical data-string.

(ii) determining the modulus of the Fourier transform,

(iv) summing the products from the multiplication step for each filter to produce a resultant sum for each filter. Advantageously, the step of converting the extracted low spatial frequency information into a numerical data-string comprises the steps of

(i) normalising the sum for each filter,

According to a further aspect of the present invention, there is now proposed a computer program for checking the identity of an individual comprising,

(i) an input module arranged to receive a digitised thermal image of a portion of the individuals body and to capture a reference numerical data-string supplied by the individual, said digitised thermal image and numerical data-string being stored in an input memory store,

(ii) an image processing module arranged in communication with the input memory store so as to receive the digitised thermal image therefrom, the image processing module adapted to determine the collective power in a plurality of spatial frequencies of interest within the digitised thermal image, in

^• at least two orientations within the digitised thermal image, and to store the collective power for each of the at least two orientations in an image processing memory store,

(iii) a conversion module, arranged in communication with the image processing memory store and configured to retrieve data therefrom, the conversion module adapted to normalise and re-scale the collective power with respect to a predefined range for each of the at least two orientations so as to form a numerical output comprising a numerical character corresponding with each orientation,

(iv) a data processing module having an input arranged to receive the at least two numerical characters from the conversion module and configured to arrange the at least two outputs into a sequence to form a computed numerical data- string characteristic of the individual, and

(v) a comparison module, arranged in communication with the input memory store and the data processing module, adapted to compare the computed numeπcal data-string with the reference numerical data-string supplied by the individual and to provide an output indicative of the correlation between the computed numeπcal data-string with the reference numerical data-string.

The above mentioned computer program provides a means for automatically checking the identity of an individual in real-time. The computer program is advantageous in that it reduces the technical specification, and hence cost, of the identification apparatus. The computer programme enables a thermal imager having a low spatial and thermal sensitivity to be used. Moreover, the above computer programme is less intensive in terms of computer memory and computer processor time / specification than conventional thermal identification techniques. Consequently, the speed of the identification process is increased over conventional methods.

As well as adding an extra level of security to financial transactions (e.g. automated teller machines (ATM), point-of-sale applications), the computer program of the present invention is applicable advantageously in other access control applications. For example, the computer program can be used to control physical access to a facility, for example a room or vehicle, and to manage access to electronic facilities, for example a computer network.

Since the original image of the individual is not merely compressed, and cannot be reconstructed from the numerical data-string, the present computer programme has important data protection implications. Preferably, the image processing module comprises,

(i) a Fourier transform module adapted to apply a Fourier transformation to the digitised thermal image and to determine the modulus of the Fourier transform,

(ii) a filter module comprising a plurality of filters, each filter arranged to determine the collective power in the plurality of spatial frequencies of interest within the digitised thermal image, in the at least two orientations within the digitised thermal image.

According to another aspect of the present invention, there is now proposed a storage medium containing a computer program as described above.

The invention will now be described, by example only, with reference to the accompanying drawings in which;

Figure 1 shows a flow diagram of the method according to the present invention for verifying the identity of an individual from a facial thermal image of the individual,

Figure 2 illustrates a typical Fourier transform of a low spatial resolution and low thermal resolution facial thermal image. The figure represents the modulus of the Fourier transform of the thermal image,

Figure 3 shows schematic illustrations of typical pre-computed filter elements which are applied to the Fourier transformed facial thermal image to extract spatial frequency information therefrom. Each of the filter elements shown in figures 3a, 3b, 3c, and 3d determines the power in the spatial frequencies in a given range of orientations in the original image. The filter elements shown in figures 3a and 3b analyse the power in the spatial frequencies in first and second substantially diagonal orientations respectively in the original image. The filter elements shown in figures 3c and 3d analyse the power in the spatial frequencies in substantially vertical and a substantially horizontal orientations in the original image,

Figure 4 shows schematic illustrations of the outputs of the pre-computed filter elements illustrated in figure 3 when said filter elements are applied to the Fourier transform illustrated in Figure 1. Specifically, the filter output shown in figure 4a is the result of multiplying the Fourier transform illustrated in figure 1 by the filter element illustrated in figure 3a. The filter outputs shown in figures 4b, 4c and 4d show the results of multiplying the Fourier transform illustrated in figure 1 by the filter elements of figures 3b, 3c, and 3d respectively.

Figure 5 shows a schematic illustration of the apparatus of an embodiment of the present invention for verifying the identity of an individual from a facial thermal image of the individual. The method according to the present invention for verifying the identity of an individual involves generating a numerical data-string from a thermal image of an individual which is characteristic of said individual.

Referring to figure 1, the method according to the present invention consists of calculating a numerical data-string, referred to herein after as a biometric Personal Identification Number (bioPIN), from the spatial frequency characteristics of a person's head in a thermal image. In particular, the bioPIN is calculated exclusively from the low spatial frequency characteristics of the person's head in the thermal image.

Whereas conventional thermal identification techniques analyse fine details within a thermal image, for example the shape and position of small features within the image; the method of the present invention analyses low spatial frequency characteristics within a thermal image which correspond with coarse details within the image, for example the approximate shape of the. person's head.

The first step in the method comprises acquiring an image of the individual to be identified (4) using a thermal imager. The image will typically comprise a thermal image of the head and shoulders of the individual to be identified (i.e. a facial thermal image), but could comprise a thermal image of an alternative portion of the individual's body.

Mindful that the method of the present invention analyses low spatial frequency characteristics within the thermal image, a thermal imager having a low spatial resolution and a low thermal sensitivity is used. A suitable thermal imager comprises a bolometer having 4096 picture elements (pixels) arranged in a 64 x 64 array. The smallest temperature difference that the thermal imager is capable of detecting need only be 100 - 200 x10^"3 K (100 - 200 milli-Kelvin). The smallest temperature difference that the thermal imager is capable of detecting is known as the thermal sensitivity, thermal resolution or temperature resolution of the imager.

In this case, the thermal image comprises a small grey level image derived from the output of the thermal imager. The image has a spatial resolution of 64x64 pixels, and a thermal resolution of 8 bits per pixel (providing 256 grey levels). A thermal imager responsive to wavelengths in the range 3-5μm or 8-12μm is suitable.

A Fourier transform is applied to the thermal image to transform the image from the spatial domain to the frequency domain. The modulus of the Fourier transform is subsequently calculated. Fourier transformation is an established technique for analysing frequency components of analogue signals, and has been applied to visible-spectrum imagery (H. Spies and I. Rickets, "Face Recognition in Fourier Space", conference proceedings of Vision Interface 2000, Montreal, pages 38 - 44, 2000).

Referring to Figure 2, a typical Fourier transform (14) of a 64x64 pixel facial thermal image provides an indication of the spatial frequencies within the original thermal image. Figure 2 indicates the spatial frequency content of the original image, as a function of the orientation of features within the original thermal image. The spatial frequency increases with distance from the centre of figure 2. The brightness of points within the Fourier transform (14) corresponds to the magnitude of a given frequency within the original image. Note, the centre of figure 2 is merely zero frequency energy which conveys no useful information and may therefore be ignored

The modulus of the Fourier transform (14) is subsequently multiplied by pre- computed, oriented, filters (step 8 in figure 1).

Figure 3 shows schematic illustrations of typical pre-computed filter elements which are applied to the Fourier transformed facial thermal image to extract spatial frequency information therefrom.

Each of the filter elements shown in figures 3a, 3b, 3c, and 3d extracts the power in the spatial frequencies in a given orientation in the original image. The filter elements shown in figures 3a and 3b (16, 18) analyse the power in the spatial frequencies .in a first and second substantially diagonal orientation respectively in the original image. The filter elements shown in figures 3c and 3d (20, 22) analyse the power in the spatial frequencies in a substantially vertical and a substantially horizontal orientation in the original image. ln the interests of clarity, the power referred to above is a measurement of the amount of energy within the given band of spatial frequencies defined by each filter element.

The filter elements shown in figure 3 have a "bow tie" shape which defines a small band of spatial frequencies of interest in the Fourier space. In practice, the "bow tie" shape of each filter element defines a small range of orientations disposed generally in each of the aforementioned orientations in the Fourier image (i.e. the first substantially diagonal orientation, the second substantially diagonal orientation, the substantially vertical orientation and the substantially horizontal orientation. Other filter configurations may be used depending upon the system requirements.

The pre-computed filter array elements are set to one of two real floating point numbers, either 1 or 0, according to which frequency components of the Fourier transform are to be processed. The light areas shown schematically in figure 3 represent those elements of the array that are set to 1. This is a simple binary filter mask. More tailored masks, using floating point numbers in the range 0 to 1 , could be used to improve performance.

The next step in the method is to normalise the outputs from the filter arrays (resulting from the multiplication of the modulus of the Fourier transformed image with each oriented filter). For example, the values in each array are summed and divided by the number of pixels in the Fourier transformed image to give a sequence of real floating point numbers. Each of these numbers is re-scaled with respect to a predefined range and converted (12) into integers from 0 to 9.

By way of example, the predefined range for a particular filter array may be determined by analysing the normal variation of outputs from that particular filter array for a given sample population of individuals. Each filter array may therefore have a different predefined range.

For instance, if a predefined range for one particular filter output is 14.2 to 19.7 and the calculated output for that particular filter is 14.3 then the digit '0' is associated to that output, similarly an output of 17.7 would be assigned a digit of '6'. The sequence of digits assigned to the filter outputs form the numerical data-string or bioPIN which is characteristic of the individual portrayed in the original thermal image.

Each digit in the numerical data-string or bioPIN represents information in a particular orientation in the original facial thermal image. The complete numerical data-string or bioPIN is therefore capable of characterising the relative width and height of an individual's head within the original facial thermal image. The ratio of the width and height provides an indication of the approximate shape of the head.

Figure 3 illustrates four filter elements, producing a four digit bioPIN. However, the number of filter elements may be reduced or increased depending upon the particular application. For example, the number of filter elements may be reduced to two for low security applications, in which case a two digit bioPIN would be produced. Similarly, six or eight filter elements may be used for high security applications, in which case a six or eight digit bioPIN would respectively be produced.

In the foregoing description, the data-string or bioPIN comprises a numerical sequence of integer digits similar to a conventional Personal Identification Number (PIN). Alternatively, depending on the particular application, the bioPIN comprises any sequence of symbols capable of characterising the individual portrayed in the original thermal image, for example alpha-numerical characters, decimal numbers, hexadecimal-numbers, binary numbers, a barcode etc. Similarly, the step of re- scaling the normalised outputs from the filter arrays may be omitted, in which case the bioPIN comprises a sequence of floating point numbers. Furthermore, the bioPIN may be encrypted.

The foregoing method for producing a bioPIN, in real-time, which is characteristic of an individual may be used in access control applications to verify the identity of said individual. Such access control applications include financial transactions (automated teller machines (ATM), point-of-sale applications), controlling physical access to a facility, for example a room or vehicle, and managing access to electronic facilities, for example a computer network.

A practical example of the use of the method of present invention for the authorisation of credit card and debit card transactions comprises the following steps; The individual applies for and receives a credit card in the normal manner. A sample thermal image of the individual is acquired each time the individual uses the credit card during a given period (an enrolment period) at appropriately equipped electronic terminals. A bioPIN data-string is generated from each sample thermal image using the method of the present invention. Each bioPIN data-string is recorded in a reference database.

At the end of the enrolment period the bioPIN data-strings recorded during the enrolment period are validated and may be used to form the basis of future verification decisions for the individual associated with that credit or debit card.

For example, the credit card statement is issued to the individual in the normal manner and if the credit card bill is paid unchallenged, then the bioPIN data-strings recorded during the enrolment period are validated.. In this example, payment of the bill validates the bioPIN data-strings for the individual stored in the database.

Thereafter, each time the individual uses the credit card, a thermal image of the individual is acquired and a bioPIN calculated therefrom in real time. The bioPIN calculated in real time is compared with the validated bioPIN data-strings to verify the identity of the individual using the credit or debit card. In this example, verification will only be possible at ATMs equipped with a thermal imager.

As the individual continues to use the credit card, the reference bioPIN data-strings held in the database may be periodically updated to track changes in the individual's appearance.

If anyone other than the individual attempts to use the credit card, it is likely that there will be a significant mismatch between the new bioPIN and the reference bioPIN data-strings for the individual stored in the reference database. The ATM or electronic terminal will therefore indicate that the identity if the individual has not been confirmed and can refuse to authorise the transaction if the credit card company decides to do so.

In the above example, a reference database of bioPIN data-strings is used to verify the identity of the individual. The reference database of bioPIN data-strings may be held remotely to the ATM or electronic terminal, but be accessible by the ATM or electronic terminal. Alternatively, the validated reference bioPIN data-string(s) may be stored on the credit card. In this case, the bioPIN could be stored in an encrypted form.

In a further example of the use of the verification method according to the present invention, the reference database of validated bioPIN data-strings may be dispensed with entirely. In this case, the card holder is requested to enter a pre-issued reference bioPIN via some form of data entry means, for example a keypad or keyboard. As above, a thermal image of the individual is acquired and a bioPIN calculated therefrom in real time. The bioPIN calculated in real time is compared with the reference bioPIN data-string entered by the individual using the credit or debit card to verify their identity. This configuration provides several advantages. There is no central database of bioPINs, no additional ATM / EPOS traffic, and no data protection issues associated with storing biometric data.

In the above examples, the verification method of the present invention provides a secondary means for verifying the identity of the individual in combination with a primary identification means. The primary identification means is used in relatively high security applications to verify the identity of the individual, in which cases the verification method of the present invention provides a confidence check.

In the above examples the primary identification means comprises a credit or debit card, but could comprise a smart card, a conventional key, a conventional Personal Identification Number (PIN), or a password depending upon the particular application.

When used in conjunction with primary identification means, the method of the present invention provides reliable verification of an individual's identity since the primary identification means restricts the reference data-set against which the newly calculated bioPIN is compared. In this manner, the method of the present invention is merely comparing the newly calculated bioPIN against a few reference bioPINs in a database or a against a single bioPIN entered by the individual to be identified. The objective of the foregoing method is to indicate whether the individual attempting to gain access is more likely than not to be the authorised person. The objective of the method is not to recognise the individual from a database of thousands or even hundreds of a potentially authorised individuals.

Notwithstanding the foregoing, the method of the present invention may be used without primary identification means to verify the identity of an individual purely from a bioPIN data-string for low security applications.

A card-less version of the present invention could be used to control access to a restricted facility (whether physically, for example in the case of access to a building / vehicle, or electronically, for example in the case of access to a computer network).

For example, when checking into a hotel, the reception desk could measure your bioPIN and give you an encrypted four digit bioPIN number that only you can use to gain access to your room.

This encrypted bioPIN has the advantage over an ordinary PIN number in that it need not be kept secret. Instead of receiving a key, a list of bioPIN numbers could be provided to permit access to areas such as the leisure suite, or to the minibar, or to the television.

The equipment used to verify the identity of individuals would not need to be networked, but could be battery operated and remotely located, requiring no communication between units whatsoever. However, such systems would require certain data to be set periodically, for example, the date and randomisation seed, and a frequency of changing the encryption code (daily/weekly). All these functions could be set conveniently with a remote controller, or an infrared equipped personal digital assistant (PDA). This would save the expense of wiring and the cost of lost keys and keycards, etc.

When configured as discussed above, independent verification apparatus set with the same configuration (date, bioPIN algorithm etc.) would generate the same bioPIN for an individual. A unit set with a different configuration would generate a different bioPIN. An individual can therefore be given a printed bioPIN for each controlled area that he/she is allowed to enter, e.g. fitness suite, pool, or office areas and labs, and the numbers can be reset as frequently as required. When an individual wishes to gain access to an area, they simply enter their printed bioPIN number and the verification equipment measures and encodes their bioPIN. The verification equipment makes a local decision on whether or not to grant access, based on the similarity of the encoded bioPIN to the bioPIN entered by the individual.

The advantages of this system over existing access control systems is that no key or card is required (and therefore no possibility of illegal copies being made), the bioPIN number provided does not have to be kept secret and can be carried by the individual without compromising security. There is no need to memorise the bioPIN number. The access control system needs no networking, saving on installation and cabling costs.

The foregoing verification method is substantially insensitive to the orientation of the person's head in the thermal image. The orientation of the head in the thermal image may cause the bioPIN to change slightly, however the reference data-set of validated bioPINs is adaptable to comprise a range of bioPINs. Accordingly, the method still provides useful information upon which a verification decision can be made.

Although the verification system is substantially insensitive to the orientation of the person's head in the thermal image, the thermal image may be adjusted prior to applying the Fourier transformation to improve repeatability of results and enhance reliability. For example, the position and size of the person's head within the thermal image may be normalised to allow for differences in alignment and distance between the thermal imager and the person to be identified.

Since the method of the present invention utilises low spatial frequency characteristics of an individual's head in a thermal image, it is less sensitive to minor variations in the appearance of the individual than conventional facial identification techniques. For example, the present method is substantially unaffected by small features on the face.

It should be noted that the original thermal image of the individual to be identified is not being compressed in the present method. Accordingly, the original image cannot be reconstructed from the bioPIN data-string. This has important data protection implications. The error rate computed for the verification method of the present invention using an initial data set is 7% from a test sample of 25 people. This is comparable with the best conventional facial recognition and identification techniques but provides numerous advantages in terms of reduced computational requirements, reduced storage (memory) requirements, privacy, covertness, and difficulty in defeating.

Moreover, in embodiments of the invention where the individual to be identified supplies a pre-advised reference bioPIN, there is no requirement for stored reference images, stored data-strings or pre-calculated facial data.

In an alternative embodiment of the present invention, the low spatial frequency characteristics within the thermal image are deduced directly from the original grey level thermal image rather than from a Fourier transform of the image. In this case, pre-computed, oriented, filters are applied the grey level thermal image in the spatial domain rather than in the frequency domain (Fourier space). The pre-computed, oriented filters are a simplified representation of the low spatial frequency simple cells in the human visual cortex.

Claims

1. A method for checking the identity of an individual comprising the steps of

(i) acquiring a thermal image of a portion of the individual's body,

(iii) converting the extracted low spatial frequency information into a numerical data-string characteristic of said individual, and

(iv) comparing the numerical data-string with at least one reference numerical data-string in order to verify or otherwise the identity of the individual.

2. A method according to claim 1 comprising the steps of determining the collective power in a plurality of spatial frequencies of interest within the thermal image, in at least two orientations within the thermal image, converting the collective power for each of the at least two orientations into a numerical character, and arranging the numerical characters into a sequence to form the numerical data-string characteristic of the individual.

3. A method according to claim 1 or 2 wherein the step of processing the thermal image comprises the steps of

(ii) determining the modulus of the Fourier transform,

(iii) multiplying the modulus of the Fourier transform by a plurality of filters, each filter arranged to select spatial frequencies in a spatial frequency range within the Fourier transform corresponding with a different orientation in the thermal image, (iv) summing the products from the multiplication step for each filter to produce a resultant sum for each filter.

4. A method according to claim 3 wherein the step of converting the extracted low spatial frequency information into a numerical data-string comprises the steps of

(i) normalising the sum for each filter,

5. A method according to claim 3 or 4 wherein the filters comprise a first filter arranged to select spatial frequencies in the Fourier transform in a first orientation and a second filter arranged to select spatial frequencies in the Fourier transform in a second orientation, the first and second orientations being substantially orthogonal to each other.

6. A method according to claim 5 wherein the filters comprise a third filter arranged to select spatial frequencies in the Fourier transform in a third orientation aligned at an angle of substantially 45 degrees to the first orientation and a fourth filter arranged to select spatial frequencies in the Fourier transform in a fourth orientation aligned at an angle of substantially 45 degrees to the second orientation, the third and fourth orientations being substantially orthogonal to each other.

7. A method according to any of the preceding claims wherein the portion of the individual's body comprises the individual's head.

8. A method according to any of the preceding claims wherein the numerical data- string comprises at least two numerical characters.

9. A method according to any of the preceding claims wherein the numerical data- string comprises at least four numerical characters.

10. A method according to any of claims 2 - 9 wherein each numerical character comprises one of a decimal number, a floating point number, an integer, an alpha- numerical character, a hexadecimal-number, a binary number, a graphical symbol and a barcode.

11. A method according to any of the preceding claims wherein the reference numerical data string is input by the individual to be identified as part of the comparison step.

12. A method according to claim any of the preceding claims and further comprising the step of providing an output indicative of the result of the comparison of the numerical data-string with the reference numerical data-string.

13. A method according to any of the preceding claims wherein the thermal image contains not more than 4096 pixels.

14. A method according to claim 13 wherein the thermal image is configured as an array of 64 x 64 pixels.

15. A method according to claim 13 or 14 wherein the smallest temperature difference resolvable in the thermal image is 200 x10^"3 Kelvin or less.

16. A method for checking the identity of an individual substantially as herein before described with reference to figures 1-4 of the accompanying drawings.

17. An automated teller machine (ATM) having a thermal imaging camera and a processor, the processor configured to check the identity of an individual using the method according to any of claims 1-15.

18. An electronic point-of-sale (EPOS) terminal having a thermal imaging camera and a processor, the processor configured to check the identity of an individual using the method according to any of claims 1-15.

19. An apparatus for checking the identity of an individual from a thermal image of a portion of the individual's body, comprising,

means for comparing the numerical data-string with at least one reference numerical data-string in order to verify or otherwise the identity of the individual.

20. An apparatus according to claim 19 wherein the image processing means comprises

21. An apparatus according to claim 20 wherein the means for converting the extracted low spatial frequency information into a numerical data-string comprises

means for normalising the sum for each filter,

means for re-scaling the normalised sum for each filter with respect to a predefined range configured to calculate a numerical output for each filter comprising a numerical character, and means for arranging the numerical outputs from the filters into a sequence comprising the numerical data-string characteristic of the individual.

22. An apparatus according to any of claims 19 - 21 wherein the filters comprise a first filter arranged to select spatial frequencies in the Fourier transform in a first orientation and a second filter arranged to select spatial frequencies in the Fourier transform in a second orientation, the first and second orientations being substantially orthogonal to each other.

23. An apparatus according to claim 22 wherein the filters comprise a third filter arranged to select spatial frequencies in the Fourier transform in a third orientation aligned at an angle of substantially 45 degrees to the first orientation and a fourth filter arranged to select spatial frequencies in the Fourier transform in a fourth orientation aligned at an angle of substantially 45 degrees to the second orientation, the third and fourth orientations being substantially orthogonal to each other.

24. An apparatus according to any of claims 19 - 23 wherein the portion of the individual's body comprises the individual's head.

25. An apparatus according to any of claims 19 - 24 wherein the numerical data-string comprises at least two numerical characters.

26. An apparatus according to any of the claims 19 - 25 wherein the numerical data- string comprises at least four numerical characters.

27. An apparatus according to any of claims 21 - 26 wherein each numerical character comprises one of a decimal number, a floating point number, an integer, an alpha-numerical character, a hexadecimal-number, a binary number, a graphical symbol and a barcode.

28. An apparatus according to any of claims 19 - 27 and further comprising output means arranged to provide an output indicative of a result of the comparison of the numerical data-string with the reference numerical data-string.

29. An apparatus according to any of claims 19 - 28 and further comprising a reference repository suitable for at least one reference numerical data-string.

30. An apparatus according to claim 29 comprising an input means for input, in use, of a reference numerical data string by the individual to be identified.

31. An apparatus according to claim 29 and further comprising means adapted to read the reference numerical data string from a removable storage means.

32. An apparatus according to claim 31 wherein the removable storage means comprises one of an identification card, a credit card and a debit card.

33. An apparatus for checking the identity of an individual substantially as herein before described with reference to figure 5 of the accompanying drawings.

34. An automated teller machine (ATM) having a thermal imaging camera and a processor, the processor comprising an apparatus according to any of claims 19 - 32.

35. An electronic point-of-sale (EPOS) terminal comprising an apparatus according to any of claims 19 - 32.

36. A method for generating a biometric personal identification number characteristic of an individual, from a thermal image of a portion of the individual's body comprising the steps of

37. A method according to claim 36 comprising the steps of determining the collective power in a plurality of spatial frequencies of interest within the thermal image, in at least two orientations within the thermal image, converting the collective power for each of the at least two orientations into a numerical character, and arranging the numerical characters into a sequence to form the numerical data-string characteristic of the individual.

38. A method according to claim 36 or 37 wherein the step of processing the thermal image comprises the steps of

(ii) determining the modulus of the Fourier transform,

39. A method according to claim 38 wherein the step of converting the extracted low spatial frequency information into a numerical data-string comprises the steps of

(i) normalising the sum for each filter,

40. A method according to claim 39 wherein each numerical character comprises one of a decimal number, a floating point number, an integer, an alpha-numerical character, a hexadecimal-number, a binary number, a graphical symbol and a barcode.

41. A computer program for checking the identity of an individual comprising

(ii) an image processing module arranged in communication with the input memory store so as to receive the digitised thermal image therefrom, the image processing module adapted to determine the collective power in a plurality of spatial frequencies of interest within the digitised thermal image, in at least two orientations within the digitised thermal image, and to store the collective power for each of the at least two orientations in an image processing memory store,

(v) a comparison module, arranged in communication with the input memory store and the data processing module, adapted to compare the computed numerical data-string with the reference numerical data-string supplied by the individual and to provide an output indicative of the correlation between the computed numerical data-string with the reference numerical data-string.

42. A computer program according to claim 41 wherein the image processing module comprises,

43. A storage medium containing a computer program according to claim 41 or 42.