GB2388457A

GB2388457A - Fingerprint identification system

Info

Publication number: GB2388457A
Application number: GB0210610A
Authority: GB
Inventors: Graham Leslie Wright
Original assignee: Central Research Laboratories Ltd
Current assignee: Central Research Laboratories Ltd
Priority date: 2002-05-09
Filing date: 2002-05-09
Publication date: 2003-11-12
Also published as: GB0210610D0

Abstract

A fingerprint classification and recognition system processes a subject fingerprint to be identified to create respective groups of data representing ridge lines in the fingerprint and analyses the grouped data at the scale of the ridge lines. The subject fingerprint is surrounded by a clear field, and the system locates the subject fingerprint with respect to the field, to permit ridge lines which end naturally at the perimeter of the fingerprint as captured to be differentiated from ridge lines which have true endings in the fingerprint area, and which thus are to be identified as minutiae. Such a system is less susceptible to the detection of false bifurcations in the ridge lines due to fingerprint image defects.

Description

1, 2388457

FINGERPRINT IDENTIFICATION SYSTEM

This invention relates to systems for identifying fingerprints, and it addresses particularly certain 5 problems that arise in the automatic identification and classification of minutiae; these comprising specific marks, such as ridge endings, bifurcations and short ridges (or so-called "islands") whose relative placement is known to accurately characterize fingerprints.

Procedures are known for identifying the minutiae and for creating a map of vectors indicative of their placement, relative to one another, in the fingerprint. In order to identify a subject fingerprint amongst a large number of 15 candidate (or reference) fingerprints, it is known, therefore, to compare a vector map derived from the subject fingerprint with each of many candidate vector maps that are held in a reference store; these having been respectively derived from known fingerprints.

Various procedures are known for reducing the number of candidates with which subject fingerprint data needs to be compared, in order to reduce overall processing time and complexity, and typical procedures in this respect include 25 identification of macroscopic features such as whorls, loops and arches contained in the fingerprints. Only candidates having the requisite distribution of such macroscopic features (as dictated by the features of the subject fingerprint) are utilized in the comparison of 30 minutiae based data. It will be appreciated that any convenient known procedure, or a combination of such procedures, may be used in conjunction with the present invention to effect initial data reduction.

35 In some circumstances, however, such as where a fingerprint-based recognition device is used to determine

2 l ( whether individuals may be afforded access to an area, or to a bank or other monetary account or be cleared to participate in some financial transaction, for example, and where the individual claims a particular identity, as by 5 entering PIN data, there is no need for an initial data reduction procedure, since the subject fingerprint needs only to be compared with the specific candidate mark associated with the PIN.

10 In any event, whatever the circumstances regarding the overall number of comparisons needed in any given set of circumstances, it is of the utmost importance that the minutiae are correctly identified and classified so that their relative positions can be accurately specified.

Ill existing systems, difficulties arise in this respect, principally in relation to the accurate recognition and classification of minutiae in view of false data which typically is incorporated into fingerprint data captured by 20 most practical fingerprint readers. Such false data can arise from such things as dirt or moisture (existing and/or transferred from a fingerprint being taken) on the reader surface. Such dirt and/or moisture can create anomalous bridges between minutiae such as ridges, thereby creating 2s false date.

If the fingerprint has been retrieved (e.g. in a forensic situation) from a surface other than that of a reader, the situation can be even worse, as fragments of the 30 fingerprint-bearing medium and/or other contaminants can confuse the image and of course the fingerprint itself may be incomplete and/or indistinct.

It is an object of this invention to improve the accuracy 35 with which the minutiae associated with fingerprints to be

( identified automatically may be recognized, classified and vectorially correlated.

According to the invention there is provided a fingerprint 5 classification and recognition system in which a subject fingerprint to be identified is processed to create respective groups of data representing ridge lines in the fingerprint and to analyse the grouped data at the scale of the ridge lines. Such a system is advantageously less 10 susceptible than prior systems to the detection of false bifurcations in the ridge lines due to fingerprint image defects. In order that the invention may be clearly understood and 15 readily carried into effect, one embodiment thereof will now be described by way of example only and with reference to the accompanying drawings of which: Figure l shows, in block schematic form, a system in 20 accordance with one example of the invention; Figure 2 shows an inner loop and a bifurcated region of a fingerprint; and 25 Figures 3(a) and 3(b) show noise regions falsely bridging ridge lines.

In this example of the invention, it is assumed that the subject fingerprint as captured is surrounded, wholly or 30 partially, by a notionally clear "outfield".

Identification of the placement of the subject fingerprint with respect to the clear outfield is an important

precursor to the further processing, since it permits ridges which end naturally at the perimeter of the 35 fingerprint as captured to be differentiated from ridges which have true endings in the fingerprint area, and which

( thus are to be identified as minutiae. Thus the first stage of processing of the system shown in Figure 1 is that, identified at step 101, of identifying the placement of a subject fingerprint, as captured, within the s notionally clear outfield.

This processing step can be achieved, for example, by comparing the contrast range in small blocks (of up to 8x8 pixels) throughout the image to that over a larger area, lo central of the image, which is assumed to contain fingerprint data. Those blocks whose contrast range lies below a certain proportion of the contrast in the central area are deemed "bad", and characterized as likely to lie in the clear outfield. Blocks that are not, in themselves,

IS deemed "bad", but are largely or wholly surrounded by bad" blocks are yerlerally indicative of image artifacts caused, for example, by surface contamination of the fingerprint capture device, and may be recategorized as "bad".

20 In the next stage (102) of the process currently being described, the image is binarised, using a locally variable threshold with median filtering if necessary, to generate a binary fingerprint image comprising ridges and spaces having approximately equal widths.

2s An optional phase (103), known as skeletonization, can be implemented at this stage if required in order to reduce the widths of all ridges to that of a single pixel without at any point losing the continuity of the ridge. A typical 30 skeletonization process' as described, for example, in Patent No. GB 2 278 945, is accomplished by repeatedly comparing the value of each pixel with that of each of its eight nearest neighbours; removing pixels as long as the condition of maintenance of continuity can be met.

In another optional step (104), usable as an alternative to skeletonization, isolated pixels, arising from high frequency noise, are eliminated using classical erosion and dilation techniques.

s In accordance with this example of the invention, the next step, 105, is effective to assign identify groups of contiguous (ridge) pixels as being part of a common group.

lo In effect, the image is analysed sequentially, pixel by pixel. At each stage, the pixels adjacent to the pixel selected for analysis are examined. If the selected pixel has a value indicating that it lies on a ridge, and one of the neighbouring pixels has been assigned to a group of 15 pixels representative of a ridge, the selected pixel is assigned to that group. It will be appreciated that, in this part of the process, pixels that have not already been identified by the binarisation process as being part of a ridge cannot, by definition, be grouped in this way.

This simple procedure will group together areas of contiguous pixels but will, in many circumstances, produce contiguous blocks with different group numbers, depending upon the edge shape of each block (or ridge line, in 25 practice). It is therefore necessary to recursively analyse (106) each pixel in the image, reassigning group numbers appropriately when two contiguous pixels are found to have differing group numbers, until no contiguous groups with different group numbers are found.

It will be understood that the definition of contiguousness may be amended so that a requirement is made to have anywhere between 1 and 7 adjacent pixels in contact.

3s In step 107, vectors proceeding in each of eight radial lines from a grouped pixel are examined. The length of

( each vector is chosen to be approximately 1.5 times the pitch of the ridge lines; the pitch being estimated, for example, by examining the density profile of a two dimensional Fourier transform of the fingerprint so as to S optimize the absolute vector length for each particular fingerprint. Accurately matching the vector length to the above value is not essential, however, and other values can be chosen if preferred.

10 It will be appreciated that each ridge pixel has, by operation of the preceding steps, already been assigned to a group tagged in association to that ridge, so that a starting pixel is assigned, for example, to group A. Is Proceeding along each vector, each pixel previously identified as lying on a ridge ls examined. If a pixel so examined is tagged as a member of another group (hence identified as lying on a different ridge to that of the starting pixel) the vector line is attributed to that other 20 group and no further pixels along it are examined.

If, on the other hand, an examined pixel is tagged as being a member of the same group A (i.e. lies on the same ridge) as the starting pixel) two counters are triggered; the 25 first to count ridge elements and the second to count non-

ridqe elements (i.e. gaps) encountered when pet ng along a vector. The process is conditioned to search for values of the first counter exceeding a threshold chosen to be greater than the average ridge width and for values of 30 the second counter corresponding to the average inter-width spacing. If the second counter provides a count corresponding to the average inter-ridge spacing, and the vector then encounters a pixel tagged as belonging to a different group than the starting pixel, the vector is 35 assigned to the new group and analysis along that vector line is terminated.

( If, on the other hand, the next pixel encountered along the vector line is tagged to the same group as the starting pixel, the vector is assigned to the starting group and it is assumed that a bifurcation of the line has been 5 encountered.

Where the count accumulating in the first counter exceeds its predetermined threshold value, which typically is less than or equal to the average width of the ridge in pixels, 10 the vector is assigned to group A and it is assumed that the vector lies substantially along the direction of the ridge. With regard to the choice of the threshold values mentioned 15 above, the following criteria are relevant.

Where the pixel under analysis lies on the edge of a ridge, at least one vector will progress through that ridge in such a manner that the first counter will be incremented to 20 a value indicative of at least the average ridge width.

The threshold for the first counter, above which the vector is assigned to the same group, should therefore at least equal that value. Such a situation will typically arise for vectors lying substantially along the length of a ridge.

The threshold for the second counter will typically correspond to the inter-ridge spacing. Proceeding along a vector, if the threshold of the first counter is not exceeded whereas that of the second counter is exceeded, 30 then the vector should be assigned to the group of the first pixel encountered after the second counter's threshold is exceeded.

Once each vector has been attributed to a group (or remains 35 unassigned since no attribution can take place) the pixel

! that is at the focus of the eight vectors is then assigned, at step 108, to one of three types: a) only one group, other than group A, attributed to any 5 of the vector lines. This type is characteristic of the inner ridge of a loop or the bifurcated branch of a line (see Figure 2); b) Two (or more) groups, other than group A, attributed 10 to the vector lines and more than T_END vectors attributed to group A, where T_END is typically 1. This type is characteristic of the main body of lines in a non-

bifurcated region; 15 c) T_END vectors or fewer attributed to group A. This type is characteristic of ridge endings.

It will be appreciated that the presence of noise in the fingerprint will cause some of the pixels in a given line region to be incorrectly typed. For this reason, regions 20 within a ridge line are identified through having a preponderance of one type of pixel, or a proportion of one type above a given threshold value.

In the case of a bifurcated line, the position of the 25 bifurcation is specified by the intersection of the regions typed in the above manner.

It is necessary in some circumstances to set a lower threshold on the allowable size of a type b) line so that 30 small areas where contamination has developed false bridging areas between lines, as shown in Figure 3 (a), are not confused with bifurcation areas. Large bridged areas, of the kind shown in Figure 3(b), which can arise due to wetness of the finger for example, are automatically 35 detected in the preprocessing stage and discarded on that basis.

In contrast to existing techniques, which depend on close examination of small local areas for shapes or features characteristic of a bifurcation or other mark type, it will s be appreciated that the present invention involves analysis of the fingerprint at the scale of the entire ridge line and is thus less susceptible to the detection of false bifurcations due to fingerprint image defects.

10 In the case of interior loops, the position of the peak of the loop can be determined from the termination position of the region, coupled with the fact that such a region terminates in an area other than the edge of the frame or the clear outfield'' referred to earlier, thereby

15 distinguishing from the situation in which a bifurcated ridge extends to the edge of the fingerprint image.

Ridge endings are associated with regions wherein the majority pixel type is c). No more than two such regions should be associated with a given ridge. The possibility 20 that, in some small areas, pixels might be incorrectly assigned to type c) due to severe noise in the image, can be addressed by selecting the two most likely areas for the ridge ending, either statistically (i.e. those with the largest number of pixels) or in the basis that they are at 25 the extremity of the ridge. The latter approach can exhibit some limitations for ridges which curl, generally spirally, in on themselves.

3s

Claims

lo ( CLAIMS:

1. A fingerprint classification and recognition system including means for processing a subject fingerprint to be identified to create respective groups of data representing ridge lines in the fingerprint and for analysing the grouped data at the scale of the ridge lines.

2. A system according to claim 1 wherein the subject lo fingerprint is at least partially surrounded by a substantially clear field and the system further includes

means for locating the subject fingerprint with respect to the said field, thereby to permit ridge lines which end

naturally at the perimeter of the fingerprint as captured 15 to be differentiated from ridge lines which have true endings in the fingerprint area, Did which thus are to be identified as minutiae.

3. A system according to claim 2 wherein said means for 20 locating comprises means for comparing the contrast range in blocks each of a relatively small area throughout the image to that over a relatively large area positioned to ensure at least some content of fingerprint data.

25 4. A system according to claim 3 wherein those blocks whose contrast range lies below a certain proportion of the contrast in the relatively large area are characterized as likely to lie in the clear outfield.

30 5. A system according to any preceding claim wherein the processing means includes means for creating a binary fingerprint image comprising ridges and spaces having approximately equal widths.

1 1 6. A system according to claim 5 including means conditioned to assign groups of contiguous pixels conforming to ridge lines as being part of a common group.

5 7. A system according to claim 6 wherein said means conditioned to assign is configured to analyse the fingerprint image sequentially, pixel by pixel.

8. A system according to claim 7 wherein the pixels 10 adjacent to each pixel selected for analysis are examined and the system further includes means, effective if the selected pixel has a value indicating that it lies on a ridge line, and one of the neighbouring pixels has been assigned to a group of pixels representative of a ridge 15 line, for assigning the selected pixel to that group; further means being included for recursively analysing each pixel in the image, and for reassigning group numbers appropriately when two contiguous pixels are found to have differing group numbers, and for repeating the recursion 20 until no contiguous groups with different group numbers are found. 9. system according to any of claims 6, 7 and 8 including means for examining a plurality of vector lines 25 of a given length proceeding in a regular pattern from each grouped pixel; the length of each vector line being approximately 1.5 times the pitch of the ridge lines; and means for assigning pixels identified as lying upon ridge lines to respective groups identified with the respective 30 ridge lines.

10. A system according to claim 9 including means for further examining the group identifiers associated with respective pixels in order to determine whether

( i) a respective pixel is part of a bifurcated branch and the vector emerges from the branch line containing the pixel and intercepts a pixel from another branch; or s ii) the vector lies substantially along the direction of the ridge; or iii) the ridge is curved so that the vector emerges from the ridge into a gap and then intersects the ridge again.

11. A system substantially as herein described and/or as shown in the accompanying drawings