ES2581593A1 - System and method for the comparison of fingerprints and fingerprints based on multiple deformable clusters of matching minutiae (Machine-translation by Google Translate, not legally binding) - Google Patents

Abstract

The present invention relates to a system and method for the comparison of fingerprints independent of the type of minutia descriptor. The system is effective in the verification of fingerprints and fingerprints, as well as in the identification of fingerprints and fingerprints. The method performs multiple alignments of the tracks and for each alignment finds a cluster of matches. The clusters are mixed from higher to lower weight to cover the entire footprint independently of their global deformations. Then a thin-plate spline model is constructed from the mixed clusters and with this new matching minutiae are found. The resulting similarity value is calculated from the weight of all matching minutiae found. (Machine-translation by Google Translate, not legally binding)

Description

image 1

DESCRIPTION

System and method for fingerprint and palm print comparison based on multiple deformable clusters of matching minutiae.

Field of the invention

The present invention belongs to the area of Computational Sciences and the specific field of Fingerprint and Palm Print Recognition.

State of the art

Fingerprints and palm prints are the marks formed by the papillary ridges and grooves on the fingers and palms respectively. These have the characteristics that they are formed in the sixth month of intrauterine life of the human being and do not disappear until time after the death of the people, there are no two people with the same footprints and they do not change naturally during the life of a person. person. This is why fingerprints and palm prints are important elements in people's identity.

Due to the complexity and number of footprints that exist, it is necessary to create systems that allow their processing efficiently. Although the first applications were in forensic sciences, the popularity of the subject is mainly due to applications in civil systems such as physical access control to facilities, logical access control to software and voter control in elections. One of the most important components in these systems is the fingerprint comparison method.

In the area of Computer Science, fingerprint comparison (or fingerprint matching) is carried out through a finite sequence of logical steps that returns a similarity value, which is known as the fingerprint comparison method. Depending on the application context, fingerprint comparison methods can be classified into: fingerprint verification methods and latent fingerprint identification methods. The objective of fingerprint verification applications is essentially to determine, given two fingerprints, if they come from the same finger. For its part, the objective of latent fingerprint identification applications basically consists of, given a latent fingerprint, searching for all the fingerprints from the same finger present in a database.

There are many methods for fingerprint verification that are currently marketed and the results in international competitions such as the FVC-onGoing (B. Dorizzi, R. Cappelli, M. Ferrara, D. Maio, D. Maltoni, N. Houmani , S. Garcia-Salicetti, and

A. Mayoue, "Fingerprint and on-line signature verification competitions at ICB 2009", in International Conference on Biometrics (ICB 2009), 2009, pp. 725-732) show the high efficiency achieved by this type of method. The good results achieved by these methods are due to the fact that they work with fingerprints captured with specialized scanners to take the fingerprints of people voluntarily and under controlled conditions, so these fingerprints (also known as impressions) have an excellent level of detail, a lot of information and little noise.

image2

On the other hand, latent prints are those formed on surfaces that are not necessarily uniform, involuntarily and not controlled by people when handling objects. This is why latent fingerprints have partial information, low details and a lot of noise compared to prints (AA Paulina, J. Feng, and AK Jain, "Latent fingerprint matching using descriptor-based hough transform", IEEE Transactions on Information Forensics and Security, vol. 8, no. 1, pp. 3145, 2013). Hence, the methods for latent fingerprint identification reported in the literature or available for sale are few and still low in efficacy (AA Paulino, J. Feng, and AK Jain, "Latent fingerprint matching using descriptor-based hough transform", IEEE Transactions on Information Forensics and Security, vol. 8, no. 1, pp. 3145, 2013).

Some authors (B. Decann and A. Ross, "Can a 'poor' verification system be a 'good' identification system? A preliminary study", in 2012 IEEE International Workshop on Information Forensics and Security, 2012, pp. 31-36) they argue that good methods for verification are not necessarily good for identification. Other authors

(E. Liu, AK Jain, and J. Tian, "A coarse to fine minute-based latent palmprint matching", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 10, pp. 2307-2322, 2013 ) are of the opinion that the methods created for the fingerprint comparison are not necessarily good for the palm print comparison.

Specification of the invention

The present invention consists of a system and method for comparing finger and palm prints based on multiple clusters of matching minutiae.

The method of the present invention, unlike other solutions proposed at an international level, makes it possible to compare fingerprints and palm prints with a high degree of efficiency / effectiveness for both latent fingerprint identification and fingerprint verification.

Another advantage of the present method over existing solutions is that it can be used with any minutiae descriptor (AK Jain, J. Feng, and K. Nandakumar, "Fingerprint Matching", Computer, vol. 43, no. 2, pp. 36 -44, 2010) which allows its application in any context. For example, for applications in lightweight hardware architectures, minutiae pairs or trios can be used while in applications with higher computational power, descriptors that combine the minutiae information and other features such as the orientation image and the image of minutiae can be used. frequency.

The locally coincident minutiae pairs have been found in this proposal from an avid strategy (TH Carmen, CE Leiserson, RL Rivest, and C. Stein, Introduction to Algorithms, 2nd edition. The MIT Press, 2001, p. 1184) that seeks to select the fewest number of matching pairs with the highest similarity values while trying to minimize repetition of minutiae in the selected pairs. To compare the footprints, the proposed methodology performs multiple alignments and for each alignment it finds a cluster of matches.

image3

For this, a new definition of cluster is proposed that does not depend on the connectivity between the minutiae descriptors that compose it to achieve tolerance to the absent minutiae and the errors of the feature extractor.

The method does not discretize the minutiae transformation space by aligning the footprints to avoid losing information with the discretization.

The method performs a dynamic number of alignments to find clusters of matches over most of the area of each of the footprints being compared.

To tolerate non-linear distortions at the global level in the footprints, a subset of the clusters found is mixed by checking certain geometric constraints between the cluster centroids and checking that they do not find different matching minutiae pairs sharing a common minutiae. This heuristic aims to get many truly matching minutiae pairs while keeping the number of false matches low.

Two pairs of minutiae with similar characteristics at the local level do not necessarily have the same weight because the proposed method also takes into account the global characteristics to calculate said weight. Furthermore, for the identification of latent prints, a different way of calculating weight is proposed than that used for fingerprint verification and that used for palm print verification.

Experiments carried out using fingerprint and palm print databases have shown that the proposed method is capable of increasing the efficiency of commercial systems by up to 19.5% when the quality of the images is poor. Because the method calculates multiple clusters of matches, the efficiency improvement is greatest when the quality of the footprints is worse.

Brief description of the figures

Figure 1 shows the general scheme of fingerprint recognition systems in which our invention is applied. These systems can be described by the following modules: fingerprint capture module (11, Fig. 1), fingerprint storage module (12, Fig. 1) and fingerprint processing module (13, Fig. 1a).

The fingerprint capture module (11, Fig. 1) is made up of electronic components that allow fingerprints to be digitized, for example: contact scanners, contactless scanners, digital cameras, etc.

The fingerprint storage module (12, Fig. 1) is composed of electronic components that allow the storage and persistence of the fingerprints over time, examples: flash memory, portable hard disk, computer, computer network, etc.

The fingerprint processing module (13, Fig. 1) accesses the fingerprints obtained by the capture module (11, Fig. 1) to execute methods such as: fingerprint improvement method, data encryption method, method compression, feature extraction method, etc.

image4

image5

image6

This module is in charge of determining the fingerprint information that will be saved in the storage module (12, Fig. 1). This module also contains the method that compares a fingerprint obtained from the capture module with one or more fingerprints stored in the storage module. Examples that make up this module are: CPU, microcontroller, FPGA, etc.

Figure 2 shows an enlarged fingerprint fragment with four examples of the landmarks (or minutiae) used by the invention to compare fingerprints. (21, Fig. 2) and (23, Fig. 2) are examples of minutiae known as bifurcations; while (22, Fig. 2) and (24, Fig. 2) are examples of minutiae known as terminations. The invention compares the footprints represented by bifurcations and terminations.

Figure 3 shows the minutiae of two prints of the same finger. The minutiae in the footprint (31, Fig. 3) are closer than the minutiae in the footprint (32, Fig. 3) so the present invention finds multiple clusters of coincident minutiae (33, Fig. 3), (34 , Fig. 3) and (35, Fig. 3). The double bolt segment (36, Fig. 3) is 42 pixels long and is an example of the deformation between the two tracks.

Figures 4a and 4b show a block diagram that constitutes an abstraction of the general steps of the proposed methodology.

Best method of carrying out the invention

The invention consists of a system and a method that by specifying one of three options to calculate the weight of the matching minutiae pairs in the module (13, Fig. 1) can be applied in three different contexts: a. identification of latent palm and fingerprints; b. fingerprint verification; c. verification of palm prints.

The invention compares the footprints represented by their minutiae. The minutiae are the points where the continuity of the ridges breaks and in the invention they are represented as (x, y, θ). (x, y) are the coordinates of the point in two dimensions where the ridge breaks; for example, the coordinates of the minutiae (11, Fig. 1) and (14, Fig. 1) coincide with the centers of the respective circles. θ represents the direction of the minutia and in examples (11, Fig. 1) and (14, Fig. 1) are the angles represented by image7 the arches. The set of all minutiae is defined as image7 U = {(x, y, θ): x, y, θ

image8 }, where

is the set of real numbers.

Different global distortions between the footprints being compared cause multiple clusters of matching minutiae to appear. The example in figure 3 graphically shows the difference in the distance between minutiae belonging to different clusters. The gray segments with double bolts show a distortion in the footprint (32, Fig. 3) of up to 42 pixels compared to the footprint (31, Fig. 3). Note that this distance is more than double needed (15 pixels) to match 97.5% of the minutiae according to (S. Pankanti, S. Prabhakar, and AK Jain, "On the individuality of fingerprints", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, no. 8, pp. 1010-1025, 2002). Hence, using a distance threshold of 42 pixels instead of 15 would increase the number of matching spurious minutiae that would be found when comparing different fingerprints. To tolerate these distortions, the invention uses small threshold values to find multiple clusters of matching minutiae and then relaxes these thresholds to determine which clusters to mix.

image9

Given two sets of minutiae L and I, the invention defines "coincident minutiae cluster" (or simply "cluster") as the set of minutiae pairs e = {(q1, p1), (q2, p2), ... , (ql, pl)} with centroid rC = (qh, ph)

image8 C that meets the following conditions for each pair (qk, pk)

image10 C:

image11

15o The function Ψ transforms a minutia a into another d using as a reference a pair of minutiae (b, c) and is defined as:

image12

o The function σe calculates the similarity between two minutiae according to the distance between them and is defined as follows:

image13

being the function that calculates the Euclidean distance between two minutiae and is defined as follows:

image14

o The function σθ calculates the similarity between two minutiae according to the difference between the directions of the minutiae and is defined as follows:

image15

image16

where dθ is the function that calculates the difference between the minutiae directions and is defined as follows:

image17

The best method of carrying out the invention is as follows:

// To find matches between the tracks locally the descriptors of both tracks are compared and an avid strategy is used to determine the locally matching minutiae pairs.

Step 1. Let (41, Fig. 4a) L and I be the respective sets of minutiae of the footprints to be compared, each minutia having an associated minutiae descriptor. Let A = {} be a set that will contain pairs of matching minutiae.

// Comparing the minutiae descriptors of both footprints (42, Fig. 4a).

Step 2. For every minutia q

image8 L and p

image8 I compare their respective descriptors for minutiae.

// Finding locally matching minutiae (43, Fig. 4a).

Step 3. Sort descendingly all pairs (q, p) according to the similarity value of their respective minutiae descriptors and store in A the pairs with similarity greater than zero.

Step 4. Let M = {} be a set in which, at most, max {│L│, │I│} pairs of matching minutiae will be stored, trying to minimize repetitions of minutiae within different pairs.

Step 5. For each (qh, ph)

image8 To do:

to. Yes

image18 (qk, pk) image19 M [(qh = qk) V (ph = pk)] then update M = MU {(qh, ph)}.

// Each pair of locally matched minutiae (44, Fig. 4a) is used to align the footprints (45, Fig. 4a) and find the cluster of matches (46, Fig. 4a). The minutiae pair on which the alignment is performed is considered the centroid of the found cluster. The weight of each centroid is calculated based on the position and relative direction between the minutiae of each pair in the cluster (47, Fig. 4a). A centroid has a high weight value when its cluster contains many matching minutiae pairs and low footprint deformation. This weight is calculated in three different ways, giving rise to a different application of the method for each of the following contexts: a. identification of latent palm and fingerprints; b. fingerprint verification; c. verification of palm prints.

Step 6. Let Cs = {} be the set to be filled in the next step with clusters of matching minutiae pairs obtained from multiple alignments.

image20

Step 7. For each (qh, ph)

image8 M perform an alignment as follows:

to.

Let B = {} be the set that will be filled in the next step with all matching minutiae pairs from the current iteration.

b.

For each (qg, pg)

image21

M, determine if qg matches pg by aligning by (qh, ph) as follows:

c.

Sort B in descending order according to similarity values

image22

image23 of its 20 elements.

d. Let Ch = {} be the cluster that will contain a subset of the matching minutiae pairs from the current iteration. Be

image24 = (qh, ph) the centroid of the Ch cluster.

25 e. For each (qg, pg)

image21 B achieve the fulfillment of condition (b) of the cluster definition in the invention as follows:

image25

image26

F. Calculate the weight of the centroid (qh, ph) as

image27 = p (qh, ph, L, I, Ch) according to:

o For the identification of fingerprints and palm prints:

o For the verification of palm prints:

o For fingerprint verification:

image28

image29

image30

15 where L 'is the minutiae subset of L that is intercepted with the area occupied by the minutiae of I when aligning the tracks by the pair (a, b); while I 'is the minutiae subset of I that is intercepted with the area occupied by the minutiae of L when aligning the tracks by the pair (a, b).

twenty

g. If │Ch│> 3 then update the cluster set Cs = Cs U {Ch}.

// The weight of each cluster is calculated by accumulating the weight of the minutiae pairs it contains (8, Fig. 4a).

25

Step 8. For each Ch

image10 Cs do:

to. Be

image31 = 0 the weight associated with the Ch cluster.

30 b. For each (qk, pk)

image10 Ch do:

image32

// Initialize the resulting set (9, Fig. 4a) and iterate through each cluster (50, Fig. 4a) from 35 higher to lower weight to try to mix with the resulting set (51, Fig. 4a).

Step 9. Sort Cs descendingly according to the wch weight of its elements.

Step 10. Let Cs' = {} be the set of clusters that will be selected to be mixed. Let M '= {} be the set of minutiae pairs resulting from the cluster mix in Cs'.

9

image33

image10 Cs verify that eh can be mixed with each Cg cluster in Cs' as follows:

image34

5

// A Thin Plate Spline (TPS) model (AM Bazen and SH Gerez, "Fingerprint matching by thin-plate spline modeling of elastic deformations", Pattern Recognition, vol. 36, no. 8, pp. 1859-1867, 2003.) is created from matching minutiae pairs found up to this point. The TPS model is used to find new pairs of

10 matching minutiae (61, Fig. 4b).

Step 12. Build a TPS model as in (AM Bazen and SH Gerez, "Fingerprint matching by thin-plate spline modeling of elastic deformations", Pattern Recognition, vol. 36, no. 8, pp. 1859-1867, 2003.) but, instead of using the

15 local coincidences as in (AM Bazen and SH Gerez, "Fingerprint matching by thin-plate spline modeling of elastic deformations", Pattern Recognition, vol. 36, no. 8, pp. 1859-1867, 2003.), in this invention the matches (M ') found when merging the clusters are used.

20 Step 13. Let M "be the set of new coincident minutiae pairs found between L and I through the TPS model. This step is performed as in

(A. M. Bazen and S. H. Gerez, "Fingerprint matching by thin-plate spline modeling of elastic deformations", Pattern Recognition, vol. 36, no. 8, pp. 18591867, 2003.) but using a single iteration instead of multiple.

25

// Calculating the weight of the new matches (62 al65, Fig. 4b)

Step 14. For each (qh, ph)

image8 M "-M ', repeat steps 7.a to 7.f to calculate the weight

of the matching minutiae pair (qh, ph).

30 // The final similarity value (66, Fig. 4b) is the sum of the weights of the matching minutiae pairs found as a result of the cluster mix and the use of the constructed TPS model.

35 Step 15. Let score = 0 be the resulting similarity value.

image35

image36

image37

image8 M "UM 'do:

to. Update score = score +

image38

Step 17. Return score.

The terms in which this specification has been described should always be taken with a broad and non-limiting nature.

Claims (17)

image 1 image2 image3

1.

A system for comparing fingerprints and palm prints based on multiple deformable clusters of matching minutiae characterized by being made up of various electronic components such as a fingerprint scanner to digitize fingerprints or palm prints; a memory or storage device for storing the information related to the digitized fingerprints; and a data processing module that is preferably a CPU, although an FPGA, microcontroller, or another that performs the same functions can be used, which, through at least one control method, analyzes and processes the digitized fingerprints with the objective to improve them or extract from them information that is used to compare the digitized fingerprints with others previously stored through the use of multiple clusters of minutiae to solve the following problems: fingerprint verification, palmprint verification, and identification latent finger and palm prints; returning in all cases the similarity value between two fingerprints represented in terms of minutiae descriptors.

2.

A method for fingerprint and palm print comparison based on multiple deformable clusters of matching minutiae, characterized in that it consists of the following general steps:

to.

compare minutiae descriptors,

b.

select locally matching minutiae,

c.

find locally matching minutiae clusters and weights,

d.

calculate the weight of the clusters,

and.

mix the clusters according to their weight to find globally matched minutiae pairs,

F.

build a TPS model and find new global matches,

g.

calculate the weight of global match pairs in found with the TPS model,

h.

calculate and return the similarity value.

3.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of matching minutiae, according to claim 2, characterized in that the step relative to comparing minutiae descriptors consists of finding the similarity value of comparing each minutiae descriptor of one fingerprint with each minutiae descriptor of the other fingerprint.

Four.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of matching minutiae, according to claim 2, characterized in that the step relating to selecting the locally matched minutiae consists of selecting pairs of minutiae if any of the minutiae of each pair that is selected is not involved in another selected pair of higher similarity value of their respective minutiae descriptors.

5.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the step related to finding the locally coincident minutiae clusters and weights consists of performing an alignment for each pair of minutiae matches locally and determine a cluster of matches for each alignment. The definition of cluster in the invention does not depend on the connectivity between the minutiae descriptors that make up the cluster, this achieves a greater tolerance for missing minutiae and feature extractor errors. Coincident minutiae within a single cluster meet certain geometric constraints with respect to the centroid of their cluster. Each cluster of matches is used to calculate the weight of the minutiae pair that generated it through an alignment. Two pairs of minutiae with similar characteristics at the local level do not necessarily have the same weight because the invention also takes into account the global characteristics to calculate said weight. Furthermore, for the identification of latent fingerprints and palm prints, a different way of calculating the weight is proposed than that used for fingerprint verification and that used for palmprint verification.

6.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to the previous claim, characterized in that the step related to finding the clusters and weights of locally coincident minutiae varies the way of calculating the function p (weight locally matched minutiae) differently for three different applications: a. identification of latent palm and fingerprints; b. fingerprint verification; c. verification of palm prints.

7.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the step related to calculating the weight of the clusters consists of calculating the weight of the cluster by adding the weight of the pairs minutiae contained in that cluster.

8.

The method for comparing fingerprints and palms based on multiple deformable clusters of matching minutiae, according to claim 2, characterized in that the step related to mixing the clusters according to their weight to find the pairs of matching minutiae globally consists of mixing the highest to lowest weight match clusters by checking for certain geometric constraints between the cluster centroids and checking that there are no different matching minutiae pairs sharing a common minutiae.

9.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the step related to building a TPS model and finding new global matches consists of finding new pairs of matching minutiae from of the Thin-Plate Spline model that is constructed from the matching minutiae pairs found by mixing the clusters.

10.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the relative step to calculate the weight of the pairs of global coincidences found with the TPS model consists of finding the clusters and weights of

12 image4 13 image5 image6 image7 matching minutiae according to claim 5 but this time from new global matches calculated according to claim 9.

eleven.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the step related to calculating and returning the similarity value consists of returning as the similarity value of the fingerprints that are compare, the sum of the weights of the matching minutiae pairs found according to claims 8 and 9.

12.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, according to claim 2, characterized in that the relative step to calculate and return the similarity value returns a real number greater than or equal to zero than the more The larger it is, the greater the similarity between the footprints being compared.

13.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, characterized in that it is capable of using minutiae of the bifurcation type where the minutiae are located at the origin of the bifurcation and the direction coincides with the direction of the groove that it is located in the middle of the ridges that run parallel from the bifurcation.

14.

The method for comparing fingerprints and palms based on multiple deformable clusters of coincident minutiae, characterized in that it is capable of using minutiae of the termination type where the minutiae are located at the origin or termination of the ridges and the direction of each minutia coincides with the direction of the ridge where the minutia resides.

fifteen.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, characterized in that it is capable of using minutiae whose horizontal and vertical coordinates are represented from the origin of coordinates being in the upper left corner of the Image of footprints where the horizontal axis increases from left to right and the vertical axis increases from top to bottom.

16.

The method for comparing fingerprints and palm prints based on multiple deformable clusters of coincident minutiae, characterized in that it is capable of using minutiae whose directions are represented by angles in radians that increase according to the clockwise direction and whose origin of coordinates is the upper left corner of the image of the footprints, starting to measure the angle from the horizontal axis which increases from left to right.

17.

The method for comparing fingerprints and palms based on multiple deformable clusters of coincident minutiae, characterized in that it is capable of using any type of minutiae descriptor such as minutiae pairs, minutiae trios, Minutia Cylinder Codes. ), descriptors based on the orientation image, etc.

14