WO2008155313A2

WO2008155313A2 - Biometric print enrolment and authentication

Info

Publication number: WO2008155313A2
Application number: PCT/EP2008/057570
Authority: WO
Inventors: Peter Corcoran
Original assignee: National University Of Ireland, Galway
Priority date: 2007-06-18
Filing date: 2008-06-16
Publication date: 2008-12-24
Also published as: WO2008155313A3; IE20070437A1

Abstract

Multiple images of a biometric print are acquired and processed to acquire multiple sets of data about features in the print. Each set of data comprises, for each of a plurality of features in the print, relational information about the relationship between a feature and at least one neighbouring feature in the print. The sets of acquired data are compared and the quality of the relational information is classified according to results of the comparison between the sets of data. An enrolment profile is created for the print which includes a plurality of classified items of relational information. Features which fall below a certain threshold of reliability or repeatability may be omitted entirely from the enrolment profile, or from a subsequent authentication process where a claimant's print is compared against enrolment profiles. The classification of relational information for a feature can be based on whether the relational information for that feature matches exactly in the sets of data, or matches within adjacent ranges (e.g. of angle or distance). The biometric print can be a fingerprint.

Description

BIOMETRIC PRINT ENROLMENT AND AUTHENTICATION

Field of the Invention

The present invention relates to acquiring data about a biometric print during an enrolment process and using acquired data about a biometric print during an authentication process.

Background to the Invention

Biometric data is increasingly being used as a way of verifying the identity of a person. Biometric data is used in applications such as identity cards and passports, criminal records and security systems. In a typical biometric system, biometric data is first acquired about a person's fingerprint during an enrolment process and this data is stored in a database of enrolled prints or in a storage device contained within an identity card. Subsequently, biometric data is acquired from a person seeking access to a secure entity and this newly acquired biometric data is compared in a matching process with the biometric data stored in the database or data stored on the identity card. The quality of this match determines whether or not the person is allowed access to the secure entity.

It is desirable that a biometric system is as reliable as possible, with a low rate of false negatives (low false reject rate, FRR) - i.e. people who should be allowed access are erroneously prevented from gaining access - and a low rate of false positives - i.e. people who should not be allowed access are erroneously allowed to gain access.

Fingerprints are widely regarded as the most reliable biometric. Various papers have been published on the subject of fingerprint analysis and matching. Most techniques rely on identifying characteristic points in a fingerprint, which are known as minutiae. A minutia is typically a point where a ridge in a fingerprint ends, or a point where a ridge splits into two, known as a bifurcation. Some known techniques rely on finding an accurate match between a set of minutiae, which are defined in terms of their absolute position within the print. Other known techniques propose finding a triplet of minutiae, where the triplet is defined by connecting lines, the length of the connecting lines, and the angle formed between the ridge at a minutia and the connecting line between a pair of minutiae. This last technique is more robust and is described, for example, in the papers: "Non-Alignment Fingerprint Matching Based on Local and Global Information", Zhao et al., Proceedings of the First International Conference on Innovative Computing, Information and Control (ICICIC '06); "Fingerprint Minutiae Matching Based on the Local and Global Structures", Jiang et al., Proc. 15^th Int'l Conference Pattern Recognition, 2, 2000 1038-1041, "Fingerprint Recognition by Combining Global Structure and Local Cues", Gu et al., IEEE Transactions on Image Processing, Vol.15, No.7, July 2006.

The matching process requires significant processing resources, particularly when the match is performed against a large database of enrolled prints. It has also been found that some features on a fingerprint are not as 'repeatable' as others. The fingerprint is essentially a rubber-like surface which is imposed on a flat scanning surface when a print is scanned and some features of the print are more likely to be distorted than other features. This distortion and low repeatability can lead to a high rate of false negatives.

Biometric systems are now being used in Consumer Electronics (CE) applications, such as fingerprint recognition to acquire access to a Personal Digital Assistant (PDA) or a secure digital storage device. Consumer Electronics devices generally have additional constraints compared to high-end biometric devices, such as a limited amount of processing resources, a limited power supply and a lower quality scanning device which is prone to pixellation. Although CE devices have such constraints, it is still desirable that the number of false negatives is low, as it can be frustrating when a legitimate user is prevented from accessing a device.

US 2006/0117188 describes several techniques for enhancing the quality of biometric prints. In one technique, acquired prints are discarded if they do not have a high enough quality, such as a high enough number of minutiae. In another technique, pairs of biometric prints are compared and an overall score is calculated for a particular print. This gives a limited indication of the reliability (repeatability) of a print.

The present invention seeks to provide an improved method of acquiring data about a biometric print. One aspect of the invention seeks to provide a method of acquiring and authenticating biometric print data which is more suited to Consumer Electronics applications.

Summary of the Invention A first aspect of the present invention provides a method of acquiring data about a biometric print comprising: acquiring a first image of the print and processing the image to acquire a first set of data about features in the print; acquiring at least a second image of the print and processing the second image to acquire a second set of data about features in the print; wherein each set of data comprises, for each of a plurality of features in the print, relational information about the relationship between a feature and at least one neighbouring feature in the print, the method further comprising: comparing the sets of acquired data and classifying the quality of the relational information according to results of the comparison between the sets of data; and, creating an enrolment profile for the print which includes a plurality of classified items of relational information.

It has been found advantageous to compare a number of prints acquired during an enrolment process as some of the relational information about features has a low repeatability between scans. Comparing data from multiple images and classifying the quality of the relational information in the data provides useful information which can be used in a subsequent authentication process. In effect, this classification gives an authentication process information about the quality of features within a print. Relational information which falls below a certain threshold of reliability or repeatability may be omitted entirely from the enrolment profile, or from the authentication process. Alternatively, or additionally, the classification of information can allow an authentication process to apply a weighting to matches found between certain items of relational information of an enrolment profile and relational information of a claimant's print. For example, a higher weighting can be given towards particular items of relational information which are known to reliably characterise the print of a particular enrollee, and less weighting can be given towards items of relational information which are known to be unreliable. This method has advantages over prior art methods which rely on each feature (minutia) being accurately acquired and compared during authentication. Such prior art methods tend to experience problems when stored data about an enrolled print expects a feature to be present and data about a newly acquired print lacks the feature (or vice versa). This method is particularly suitable for consumer electronics applications where a key goal is to achieve first-time matches for the right fingerprint (i.e. a low rate of false negatives) and it is possible to accept a degree of false positives.

As described above, the enrolment profile may include all of the classified relational information, or just a subset of the relational information, such as only the most reliable/repeatable relational information. Reducing the amount of information has an advantage in significantly reducing the amount of processing during a subsequent authentication process, since a reduced amount of data needs to be compared. Although the amount of data is reduced, the classification of relational information ensures that the authentication process concentrates on using the most reliable information which characterises a print. This method also has an advantage of reducing the amount of data which needs to be stored in a database of enrollees.

Advantageously, the method uses a limited number of ranges or 'bins' in the classification process. Rather than looking for an absolute match between features the method can allocate an item of relational information to one of N range bins. As an example, angular measurements can be allocated to one of sixteen possible angular ranges, rather than an exact measurement (possibly to one or more decimal places) and distances can similarly be allocated to one of M range bins (where N and M can be equal, or different). Rather than simply discarding certain information as not matching, the classifying process can look for relational information which matches in adjacent range bins. When relational information is close to a boundary of a range bin, a particular angle can appear on one side of the boundary in one image and the other side of the boundary in another image. Thus, some features will be subject to slight deformation between images and because of their proximity to range boundaries are therefore less useful for classification. The same is true for the distance measurement between minutiae. For example, consider that the angle 'bins' are from 0-9 degrees, 10- 19 degrees, etc. Each the angle at a particular minutia point is measured it will vary, i.e. statistically it has a mean and variance. Some points will have a mean which is very close to the limits of a 'bin'; thus a point that has a mean value of, say, 11 degrees would be more likely to occasionally be incorrectly categorized as belonging to the 0-9 degree bin than a point that has a mean value of, say, 14 degrees which would almost always be categorized as belonging to the 10-19 degree bin. Similarly, certain points will have very wide variance - this is particularly true of points close to regions of the image where the flow direction of the fingerprint ridges is changing rapidly; these points are also "poor" classifiers, even if they have a mean which is centred in the middle of an ' 'angle bin' ' .

The concept of 'range bins' can also be used when storing data about features of a print. This can significantly reduce the storage requirements for the enrolment profile. Alternatively, it may be preferred to only use the range bins to help classify the quality of data between images, and to store data in a more accurate form.

A further aspect of the invention provides a method of authenticating a biometric print comprising: acquiring an image of the print and processing the image to acquire a set of data about features in the print, wherein the set of data comprises, for each of a plurality of features in the print, relational information about the relationship between a feature and at least one neighbouring feature in the print; comparing the set of acquired data and relational information in an enrolment profile for a print which has been acquired using the method described above; and, determining an authentication result based on similarity of the comparison.

The embodiments described in this application use fingerprints as an example of a biometric print, but the invention is not limited to fingerprints, and can be applied to any kind of biometric print.

The functionality described here can be implemented in software, hardware or a combination of these. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. Accordingly, another aspect of the invention provides software for implementing the method. The software may be stored on an electronic memory device, hard disk, optical disc or other machine-readable storage medium. The software may be delivered as a computer program product comprising code tangibly embodied on a machine-readable carrier or it may be downloaded to the processing device via a network connection.

Brief Description of the Drawings

Embodiments of the invention will be described, by way of example only, with reference to the accompanying drawings in which: Figure 1 shows a minutia point in a fingerprint and a triplet of neighbouring minutiae points;

Figure 2 shows an image sensing array and sectors defining a range of angular values;

Figure 3 shows a method of enrolling a print; Figures 4A and 4B show sample data acquired during the method of Figure 3;

Figure 5 shows a method of authenticating a print;

Figure 6 schematically shows apparatus for implementing the invention.

Detailed Description of the Drawings Figure 1 shows a small part of a fingerprint and a set of minutiae in the print. As described above, each of the minutiae can be an end of a ridge in the print, a point where a ridge splits into two (bifurcation) or some other feature of interest in the print. When raw image data is acquired from a scan, a decision can be taken when analysing the image data as to what features in the print are to be treated as minutiae.

Figure 1 shows a minutia 10 (which will be called a key point) located at the end of a ridge 11 and three neighbouring minutiae A, B, C. The relationship between key point 10 and its neighbouring minutiae is characterised by the distance between them, and the angle formed between the connecting line and the fingerprint ridge at the neighbour point (in this example all angles are calculated in the clockwise direction). This information acts as a reference which is independent of the rotation of the fingerprint. It is also possible to use the angle formed between the connecting line and the ridge at the key point 10 itself as a third local feature. Table 1 shows this data for the key point 10. Thus, the distance between the key point 10 and neighbour A is A | , the angle between the connecting line and the ridge at point A is 284.1° and the angle formed between the connecting line and the ridge at the key point 10 is 50.1°.

Table 1

The group can comprise more than three neighbouring minutia or, less preferably, fewer minutiae. A full set of data to characterise a fingerprint will comprise a large number of tables of the type shown in Table 1, each characterising a point with respect to it's neighbours.

It can be seen that Table 1 shows the exact value of the neighbour angle and self angle, to one decimal place. However, there are reasons why these angles may not be correct, or may not be repeatable. As noted above, a fingerprint is essentially a rubber-like surface which is imposed on a flat scanning surface and some features are more likely to be distorted than others as the print is pressed against the scanning surface. There are also limitations in the image sensing array which acquires an image of the print, particularly in consumer electronics devices. An actual image sensing array has a limited resolution, and this results in a pixelated image of the fingerprint. Figure 2 shows an array 20 of pixels 21 which each pixel 21 representing the smallest discrete point that the array 20 can resolve. When the array 20 is used to scan a fingerprint, and the scanned image is analysed, the position of a minutia may not be resolved accurately. Instead, the position of a minutia will be resolved to the position of the nearest pixel, or may not be resolved at all. The measured distance between points, and the calculations of the neighbour angle and self angle will deviate from their true values. When the same fingerprint is subsequently scanned and analysed by the same scanner, different values of the distance, neighbour angle and self angle can result due to the effects of the pixelated scan array 20. Similarly, if the same fingerprint is subsequently scanned and analysed by a scanner with a different resolution, different values of the distance, neighbour angle and self angle can result due to the effects of the pixelated scan array 20. Some differences also arise from image enhancement algorithms which can interpret very "thin" regions of the ridge patterns differently on subsequent scans; the pixelated nature of the sensor is often the root cause of such variations. Another cause of problems with acquiring a reliable print is that a person's fingerprint can be affected by environmental conditions; for example, long-term exposure to water can cause minutiae to temporarily disappear.

Although the pixelation will cause the calculated distance or angle to deviate from the actual value, it has been found useful to classify the angle at minutiae points within a range of angular values. As an example, Figure 2 shows angular segments 22 centred on a point of interest in the array. Sixteen different range bins are shown, each covering 22.5°. A point falling within a range bin will be resolved to within +/-11.25 degrees of arc of the true value. This should be sufficient for Consumer Electronic applications where it is more important that a correct fingerprint is always recognized (low false reject rate, FRR) and the occasional false authentication is acceptable. The selection of the bin size can be modified depending on the accuracy that is required. Sixteen range bins equates to 4 bits of data, which is a convenient size of data for a processor to manipulate. It will be appreciated that the number of range bins can be selected according to a required level of accuracy and/or performance and that the number of range bins for the angular data need not be the same as the number of range bins for the distance data. In a preferred embodiment of the invention, data for a scanned print is stored at a high level of accuracy (e.g. angles are stored to one or more decimal places) and the range 'bins' are used as part of a classification process to classify the repeatability of data.

Figure 3 shows an overview of the process of acquiring and using prints. Firstly, there is an enrolment process 31-33 which determines an enrolment profile. At step 31 multiple scanned images of a person's fingerprint are acquired. As described below, it is preferable that a user repositions their finger between each scan. Each scanned image is typically a greyscale image. At this point the scanned greyscale image can be enhanced ('cleaned up') using known image processing techniques. The image is then converted to a binary (1 bit) black and white image. Each scanned image is analysed. Minutiae in the image are identified. All possible minutiae may be identified or only certain types of minutiae may be identified, e.g. only ridge ends. For each identified minutia, a calculation is made of data which characterises that minutia. As described above, this can include distance, neighbour angle and self angle. A preferred method has the following steps for each determined minutia point: (i) find all neighbouring minutiae beyond a predetermined minimum distance and within a predetermined maximum distance; there should be a minimum number of such neighbour minutiae - in our preferred embodiment this value is 3 - in order to successfully characterize a minutia point.

(ii) For each neighbour determine (a) the distance between the two points, D, (b) the angle between the main ridge (or averaged ridge field) at the determined minutia and the line connecting the two minutia Al and (c) the angle between the connecting line and the main ridge (or averaged ridge field) at the current neighbour minutia, A2; all angles are measured in a clockwise direction;

(iii) These three properties - D, Al and A2 are recorded in a table associated with each determined minutia point and provide a set of local feature triplets which uniquely characterize that minutia point;

At the end of this classification process there is a set of tables, one for each determined minutia point. Each table has at least N feature triplets (N=3 in the preferred embodiment).

Each scanned and analysed print has a corresponding set of data described above. At step 32, the sets of data - one set per print - are compared. The data is classified based on how closely properties match between sets of data. The classification scheme is described more fully below. The classified set of data can optionally be sorted and reduced in size, e.g. removing data for minutiae with low repeatability. The resulting set of data forms an enrolment profile which is stored at step 33.

The enrolment profile is subsequently used to perform a comparison with a newly acquired fingerprint during an authentication process.

Enrolment Process

The goal of the enrolment process is to selectively compare the extracted data sets to determine if certain data about properties of features in the print can be produced more consistently and reliably than others. To understand the problem we seek to address, consider a property of a minutia where the angle Al is very close to 45 degree; due to pixellation this property may appear to be slightly larger than 45 degree in certain acquisition conditions and slightly less than 45 degree in other cases. This will lead to a different set of triplet properties, and thus a potential misclassifϊcation of the associated minutia point. Where certain angles do not match reliably within the +/-11.25 degree bands defined in Figure 2, but are shown to match within two such bands - in this example the minutia matches within either the band from 22.5 degree to 45 degree, or else to the band from 45 degree to 67.5 degrees - then they are said to match adjacently.

To this end we take our first classified set of minutiae points and for each minutia point we compare it with all minutia points in each subsequently acquired image as follows:

(i) when all triplet properties match exactly between two minutia points they are marked as a definite match; (ii) when two or more properties match exactly between two minutia points they are marked as a probable match;

(iii) when at least one triplet property exactly matches between two minutia points they are marked as a potential match.

The correspondences between minutia points are next classified as follows:

(i) where all triplet properties (in the preferred embodiment this is at least 3) match exactly across all enrolment images for a particular minutia this is marked as a Class Al minutia in the enrolment profile; (ii) where at least two triplet properties match exactly across all enrolment images this is a Class A2 minutia;

(iii) where only one triplet property matches exactly across all enrolment images this is a Class A3 minutia

(iv)where additional properties can be shown to match adjacently then that minutia point is described as an A2+ (or A2++ where there are two adjacent features, etc) and is considered to provide a more accurate matching capability;

(v) where a minutia point matches exactly to more than one minutia in at least one of the acquired images it is considered to have a higher probability of creating false acceptances and is marked as a grade B minutia according to the same criteria for grade A minutia, i.e. Grade Bl matches three triplet properties;

B2 matches only two and B3 matches only one.

Ideally an enrolment profile will have several Al features as these are the most desirable for obtaining fast and accurate matching of fingerprints. A profile with poor quality minutiae would require the user to repeat the enrolment process.

The minutiae feature tables are next sorted and merged according to the different feature classes. Triplet properties which are not reliably matched, either exactly or adjacently, are optionally removed from the merged table. This merged table, together with the classification details of each minutia point are known as the enrolment profile.

At the end of this process the enrolment profile will comprise a reduced amount of data about features in the print. As an example, an enrolment profile may include between 4 and 8 properties with high repeatability (occurring identically in 3 out of 4 enrolment images). Each of these properties is characterized by (i) one of 16 separating distance ranges; (ii) one of 16 angular ranges on the originating minutia and (iii) one of 16 angular ranges on the terminating minutia.

Given that the local characteristics require both a distance and one (possibly two) angle(s) to be matched, it is possible to be fairly confident that there will not be too many false positives. Preferably, minutiae that are too close (within a minimum radius) should be discarded as a slight variation in distance in either axis (dx, dy) will produce very large deviations in the angles. For example, one possible limit could be to ensure minutiae are separated by at least 5% of the size of the fingerprint, e.g. for a 240 x 240 pixel scan minutiae would need to be separated by at least 12 pixels. Also it is possible to set a maximum distance beyond which minutiae are discarded. For example, one possible limit would be to discard minutiae that are separated by more than 50% of the size of the scan. A decision can also be taken to discard minutiae that are positioned within 5% of the boundary of the scan area. This could be used as a way of eliminating points on the edge of the fingerprint which may prove unreliable.

Figures 4A and 4B show two example scanned prints (after conversion to binary form) and their respective tables of minutiae and features for each minutia. Figures 4A and 4B represent two separate scans of the same print. As part of the process of comparing data sets, each feature table must be sorted according to the distance between minutiae pairs and the associated end angles must then be compared to determine matching minutiae pairs; the degree of matching will depend on each of the three measures: (i) the separating distance, (ii) the angle between local ridges and the starting minutia and (iii) the angle between local ridges and the ending minutiae. The best matches are when all three properties lie within a certain range of values for all of the enrolled fingerprints.

Authentication Process Figure 5 shows steps of a method for authenticating the print of a claimant. At step 35 an image of the claimant's print is acquired, enhanced and converted to binary form in the same manner as in step 31. Minutiae in the print are identified and a set of data is created which lists the minutiae and the relationship of each minutia to neighbouring minutiae.

The data set for the claimant's print is then compared, at step 36, with the stored enrolment profiles. For successful authentication the print should match a minimum number of minutia points from the enrolment profile. The match criteria and the number of matches will depend on the level of authentication accuracy required for a particular application. Step 37 can use authentication criteria which specify the accuracy required for the particular application. The criteria may be based on a required number of matches (in terms of classification category), an overall score or some other criteria. Step 37 issues an authentication result which can be used, for example, to permit or deny access to a secure entity, such as a building or consumer electronics device, to permit an e-commerce transaction to occur, or the result may be displayed to an operator (e.g. security officer).

In one embodiment of the authentication process, a single fingerprint of the claimant is firstly acquired. Every potential feature from this single image is matched with the classified (and reduced) set in the enrolment profile. Because the enrolment profile contains data about the most repeatable features of a print, the matching process should find at least some of those most repeatable features from the single acquired authentication image. It can be seen that this reduces the amount of processing as a relatively large set of data about features in the new image are compared with a much smaller (reliable) set of features which have previously been enrolled. The data about features of the single acquired image of the claimant do not need to match all of the data in an enrolment profile for a match to be declared. This has a further advantage where someone (being careless, or lazy) did not scan exactly the same part of their fingerprint as was used for enrolment. Further images of the claimant's print can be acquired, and processed, if necessary.

In many CE applications it is desirable to have a high authentication reliability and occasional false acceptances are quite tolerable. The penalty, in such applications, is that the CE device works unreliably for an unauthorized user. For such applications the authentication criteria can be any match between at least one triplet property of the new fingerprint and a triplet property of any Class A minutia of the enrolment profile would be sufficient for those minutiae to match. An even less stringent criterion might include Class B minutiae as well. A more stringent criterion might require that at least one triplet property matches exactly and all properties match at least adjacently. During the enrolment process it would be expected to find 4-5 minutiae which each had at least two high quality properties associated with them (three is even better) and one "confirmation quality" property. During authentication it is possible to cycle through all the minutiae of the single acquired image of the claimant's print, calculating all the properties and looking for any matches. Although there may be some random matches, there are rarely more than 2-3 such random matches between any two fingerprints, whereas it would be expected to find matches for most of the 8-10 high quality properties and also for some of the "confirmation quality" properties that were determined during the enrolment process.

Considering a particular minutia point and considering the properties created with its 4-

5 nearest neighbours, a criterion can be set which specifies that at least two of those 4-5 properties must match accurately. For some applications (e.g. e-commerce transactions) a stricter criterion can be applied, such that at least 3 of 4 properties matched (or 4 of 5).

In another embodiment of the invention, multiple images of a claimant's print are acquired and data for the print is classified in the same, or a similar, manner as described for the enrolment process to identify the most repeatable properties of the claimant's print. The authentication at step 55 then compares enrolment profiles 54 of enrollees in database 53 with the set of properties derived from the images of the claimant's print.

Figure 6 shows apparatus for performing the method described above. An image scanner 20 comprises an array of image sensing elements. The term "scanner" is intended to encompass any type of equipment capable of sensing an image of a print, and includes flat-bed image sensing arrays which remain fixed during the 'scan' and a movable sensing array which is scanned across the scanning surface. The scanner can use an optical, capacitive or ultrasonic technology, as is well known. Image data acquired by the scanner 20 is applied to an image processing function 51. This performs image enhancement (if required) and conversion from greyscale to binary. Image processing function 51 also performs feature extraction (identifying minutiae and relationship between minutiae). For the process of enrolment of a new enrollee, multiple data sets are acquired and forwarded to a classification function 52. The relational information is classified and reduced (if necessary) to form an enrolment profile for new enrollee, which is then forwarded to store 53. If the apparatus is also required to authenticate prints for a user (called a claimant), the apparatus includes the features shown on the right-hand side of Figure 6. Image scanner 20 acquires an image of a claimant's print. Image processing function 51 enhances the print, converts to binary data and extracts features and relational information to form a data set for the claimant. This data set is then compared, by authentication function 55, against a set of stored enrolment profiles 54 in store 53. Alternatively, the enrolment profile may be stored on a portable storage medium in a secure form, such as a passport, identity card etc. Authentication function 55 can also receive authentication criteria which specify the accuracy required for the particular application. The criteria may be based on a required number of matches (in terms of classification category), an overall score or some other criteria. Authentication function 55 issues an authentication result 56.

The words "comprises/comprising" and the words "having/including" when used herein with reference to the present invention are used to specify the presence of stated features, integers, steps or components but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.

It is appreciated that certain features of the invention, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable sub-combination.

Claims

1. A method of acquiring data about a biometric print comprising: acquiring a first image of the print and processing the image to acquire a first set of data about features in the print; acquiring at least a second image of the print and processing the second image to acquire a second set of data about features in the print; wherein each set of data comprises, for each of a plurality of features in the print, relational information about the relationship between a feature and at least one neighbouring feature in the print, the method further comprising: comparing the sets of acquired data and classifying the quality of the relational information according to results of the comparison between the sets of data; and, creating an enrolment profile for the print which includes a plurality of classified items of relational information.

2. A method according to claim 1 wherein the classifying comprises classifying the quality of each item of relational information into one of a plurality of categories according to results of the comparison between the sets of data.

3. A method according to claim 1 or 2 wherein the relational information in each set of data is resolved to within a range of values.

4. A method according to claim 3 wherein classification of relational information for a feature is based on whether the relational information for that feature matches exactly in the sets of data, or matches within adjacent ranges.

5. A method according to claim 4 wherein classification of relational information for a feature is also based on the proportion of the total number of sets of data that the relational information matches exactly, or adjacently.

6. A method according to claim 4 or 5 wherein classification of relational information for a feature is downgraded if the relational information for a feature in one set of data matches with multiple items of relational information within the sets of data.

7. A method according to any one of the preceding claims wherein the relational information comprises the length of a line connecting a first feature to a second feature.

8. A method according to any one of the preceding claims wherein the relational information comprises an angle formed between a feature at a first feature and a line connecting the first feature to a second feature.

9. A method according to claim 8 wherein the relational information comprises an angle formed between a feature at the feature and a line connecting the second feature to the first feature.

10. A method according to any one of the preceding claims wherein the relational information comprises information about the relationship between a feature and at least N neighbouring features in the print, where N>=2.

11. A method according to claim 10 where N=3.

12. A method according to any one of the preceding claims wherein the step of creating an enrolment profile comprises storing only some of the classified items of relational information.

13. A method according to claim 12 wherein the step of creating an enrolment profile comprises storing only the classified items of relational information having the highest classification.

14. A method according to any one of the preceding claims further comprising prompting a user to reposition between acquiring images of the print.

15. A method of authenticating a biometric print comprising: acquiring an image of the print and processing the image to acquire a set of data about features in the print, wherein the set of data comprises, for each of a plurality of features in the print, relational information about the relationship between a feature and at least one neighbouring feature in the print; comparing the set of acquired data and relational information in an enrolment profile for a print which has been acquired using the method according to any one of the preceding claims; and, determining an authentication result based on similarity of the comparison.

16. A method according to claim 15 wherein the step of determining an authentication result comprises using at least one similarity criterion which specifies the quality of match required.

17. A method according to any one of the preceding claims wherein the biometric print is a fingerprint and the feature is a minutia in the fingerprint.

18. A processing apparatus for performing the method according to any one of the preceding claims.

19. Software comprising instructions which, when executed by a processor, perform the method according to any one of claims 1 to 17.

20. Apparatus for enrolling biometric prints, comprising: a scanner for acquiring biometric prints of an enrollee; a processor arranged to perform the method according to any one of claims 1 to

14.

21. A data structure comprising at least one enrolment profile for a biometric print which comprises a plurality of features, the enrolment profile comprising a plurality of classified items of relational information for each of a plurality of features in the print, the relational information specifying the relationship between a feature and at least one neighbouring feature in the print.