KR20110107660A - Vlsi architecture for the fuzzy fingerprint vault system and matching method thereof - Google Patents

Abstract

A VLSI structure of a fingerprint purge bolt system and a fingerprint authentication matching method are disclosed. The VLSI structure of the fingerprint fuzzy bolt system according to the present invention extracts a registration feature point from a registration fingerprint image for registration, generates a fingerprint template by inserting a false feature point into the extracted registration feature point, and geometric hashing of the generated fingerprint template. A registration processing module for generating a registration hash table for performing automatic sorting based on the method and storing the registration hash table in a database; An authentication processing module for extracting an authentication feature point from an authentication fingerprint image for authentication and generating an authentication hash table for performing automatic alignment on the extracted authentication feature point based on a geometric hashing method; And calculating a similarity level by comparing the conversion feature point of the registration hash table with the conversion feature point of the authentication hash table, and determining whether to authenticate the authentication fingerprint image according to the result.

Description

VLSSI structure of fingerprint fuzzy bolt system and fingerprint authentication matching method {VLSI Architecture for the fuzzy fingerprint vault system and matching method}

The present invention relates to a VLSI (Very Large Scale Integration) structure of a fingerprint purge bolt system and a fingerprint authentication matching method thereof, and more particularly, a fingerprint purge bolt that performs fingerprint authentication while securely protecting fingerprint information using a purge bolt. A VLSI structure of a system and a fingerprint authentication matching method therefor.

The rapid development of information and the development of the Internet has led to an increase in the exchange of information through networks and the growing scale of industries related to electronic commerce such as online banking, which has increased the demand for accurate personal authentication. In this environment, a PIN (Personal Identification Number) or password method is used as the most common means of authentication. However, since there is a risk of leakage and forgetting, a security problem has recently emerged. The introduction of biometric information recognition technology such as fingerprint information is spreading as a personal authentication technology that can solve the disadvantage of this method.

The fingerprint recognition system has a very high reliability of economic installation cost and security, and is a fingerprint recognition system verified with unique characteristics of unique people based on hundreds of years of global application cases. In particular, since the system can be miniaturized, mobility and space utilization are very high. In addition, with the development of the network, with increasing interest in security and privacy, fingerprint recognition technology as a personal authentication method is developing into the most prominent technology field among the image recognition technology. Such fingerprint recognition technology is developing from a simple access control system to a remote application system such as Internet banking and e-government.

As such, user authentication using fingerprint information provides convenience and strong security. However, if the fingerprint information stored for user authentication is stolen by others, unlike a password or PIN, it cannot be changed or is limited, causing serious problems. Therefore, researches are being actively conducted to securely transmit / store important fingerprint information of a user by using technologies such as encryption, watermarking, and steganography.

Among them, the fingerprint fuzzy vault which protects fingerprint information by applying the fuzzy vault theory to fingerprint recognition is a cryptographic method that only a legitimate user can obtain a secret key by integrating the user's secret key and fingerprint information. Way. That is, the fingerprint fuzzy bolt generates a polynomial using the user's secret key, randomly generates a chaff minutiae, and then configures a fingerprint template together with the user's fingerprint minutiae. To protect.

However, when the fuzzy bolt theory is applied to the fingerprint recognition system, two fingerprints must be aligned first to measure the similarity between the stored fingerprint information and the input fingerprint information, and hundreds of false information have been added to the fingerprint template. There is a problem that it is not easy to align the two fingerprints in a state where it is not possible to distinguish which is the fingerprint feature point information of the user and which is added false information.

The present invention was devised to solve the above problems, and even if hundreds of false information is added to the fingerprint template, it is possible to easily perform fingerprint authentication by solving the problem of alignment process, as well as such a fuzzy bolt system. It is an object of the present invention to provide a VLSI structure of a fingerprint fuzzy bolt system and a fingerprint authentication matching method thereof, which can be implemented in hardware to reduce the execution time of fingerprint authentication.

According to an embodiment of the present invention, a VLSI structure of a fingerprint purge bolt system extracts a registration feature point from a registration fingerprint image for registration and inserts a false feature point into the extracted registration feature point to generate a fingerprint template. A registration processing module for generating a registration hash table for performing automatic alignment on the generated fingerprint template and storing the generated fingerprint template in a database; An authentication processing module for extracting an authentication feature point from an authentication fingerprint image for authentication and generating an authentication hash table for performing automatic alignment on the extracted authentication feature point based on a geometric hashing method; And an authentication module that compares the conversion feature point of the registration hash table with the conversion feature point of the authentication hash table, calculates similarity, and determines whether to authenticate the authentication fingerprint image according to the result.

Here, the registration processing module and the authentication processing module is preferably implemented as a software module.

In addition, the authentication module is preferably implemented as a hardware module.

Preferably, the authentication module, the memory for storing the registration hash table and the authentication hash table; A comparison unit for comparing coordinates, angles, and types of the conversion feature points of the registration hash table and the authentication hash table; And a calculator configured to calculate the number of matching conversion feature points of the registration hash table and the authentication hash table. In this case, the authentication module preferably authenticates the authentication fingerprint image when the number of matching conversion feature points is greater than or equal to a set value.

The registration hash table may be expressed as follows.

Here, r is the number of enrolled fingerprint templates, and tr _i is a hash table generated based on m _i among enrolled fingerprint templates and generated through geometric hashing, and may include r-1 transform feature points.

In addition, the authentication hash table may be expressed as follows.

Here, s is a number of authentication of the fingerprint template, tr _j will be formed as a hash table, created by the selection based on the m _j of the authentication fingerprint template and geometric hashing to s-1 of converting a feature point Issa.

The size of the memory is preferably equal to or greater than the sum of the number p of conversion feature points of the registration hash table and the number q of conversion feature points of the authentication hash table.

Here, the number p of conversion feature points in the registration hash table is [r x (r-1)], and r is preferably the number of registration fingerprint templates.

Further, the number q of conversion feature points of the authentication hash table is [s x (s-1)], and s is preferably the number of authentication fingerprint templates.

Preferably, the j th authentication fingerprint conversion feature point tr _j is input to the comparator in the authentication hash table, and the i th registration fingerprint conversion feature point tr _i is input to the comparator in the registration hash table, and the input two transformations are input. Compute the number of matched transformed feature points by comparing the coordinates, angles, and types of the feature points, input tr _{i + 1} into the comparison unit in the registered hash table, and compare tr _j and tr _{i + 1} to match. calculates the number of transformation feature point, to enter the tr _{j + 1} in the authentication hash table in the comparison unit tr _i And iterative comparison with tr _{i + 1} .

On the other hand, the fingerprint purge vault system, the step of inputting the j th authentication fingerprint conversion feature point tr _j in the comparison hash table; Inputting an i th registration fingerprint conversion feature point tr _i in the registration hash table into the comparison unit; Calculating the number of matched transform feature points by comparing coordinates, angles, and types of the input transform feature points; Inputting tr _{i + 1} into the comparison unit in the registration hash table and comparing tr _j and tr _{i + 1} to calculate the number of matched transform feature points; And the input to the comparison unit the tr _{j + 1} in the authentication hash table tr _i And performing an iterative comparison with tr _{i + 1} .

According to the present invention, by implementing the hardware structure of the fingerprint fuzzy bolt by integrating the software module and the hardware module, it is possible to protect the fingerprint information by using the geometric hashing-based fingerprint fuzzy bolt and shorten the time for performing fingerprint authentication.

1 is a block diagram schematically illustrating a fingerprint purge bolt system according to an embodiment of the present invention.
FIG. 2 is a diagram schematically illustrating an authentication module of the fingerprint purge bolt system of FIG. 1.
3 shows an example of a fingerprint image, where (a) shows an image of a registered fingerprint and (b) shows an image of an authentication fingerprint.
4 is a diagram illustrating an integrated implementation of a software module and a hardware module of the fingerprint purge bolt system of FIG. 1.
FIG. 5 is a diagram illustrating fingerprint matching by the authentication module of FIG. 1.
6 is a flowchart illustrating a fingerprint authentication matching method by the authentication module of FIG. 1.

Hereinafter, with reference to the accompanying drawings will be described in detail the VLSI structure and fingerprint authentication matching method of the fingerprint purge bolt system according to an embodiment of the present invention.

1 is a block diagram schematically illustrating a fingerprint purge bolt system according to an embodiment of the present invention.

Referring to the drawings, the fingerprint purge bolt system 100 may include a registration processing module 110, an authentication processing module 120, and an authentication module 130.

The registration processing module 110 extracts a registration feature point from a registration fingerprint image for registration, generates a fingerprint template by inserting a false feature point into the extracted registration feature point, and automatically aligns the generated fingerprint template based on a geometric hashing method. Create a hash table to register and store it in the database.

The authentication processing module 120 extracts an authentication feature point from an authentication fingerprint image for authentication, and generates an authentication hash table for performing automatic alignment on the extracted authentication feature point based on a geometric hashing method.

The authentication module 130 compares the conversion feature points of the registration hash table with the conversion feature points of the authentication hash table, calculates similarity, and determines whether to authenticate the authentication fingerprint image according to the result. To this end, the authentication module 130 may include a memory 132, a comparator 134, and a calculator 136, as shown in FIG. 2.

The memory 132 stores the registration hash table generated by the registration processing module 110 and the authentication hash table generated by the authentication processing module 120 as a database. Here, as shown, the memory 132 may independently store the registration hash table and the authentication hash table, or may be stored together in one memory.

The comparison unit 134 compares the coordinates, angles, and types (types) of the conversion feature points of the registered hash table and the authentication hash table.

The calculation unit 136 counts and calculates the number of matching conversion feature points based on a comparison result of coordinates, angles, and types of each conversion feature point of the registration hash table and the authentication hash table by the comparison unit 134.

In general, fingerprint feature information used in the fingerprint recognition system, as shown in Figure 3a, the end point (ending point) 310 through which the ridge passes in the fingerprint image and the branching point (bufurcation point) 320 is divided into one ridge Use One feature point extracted from the fingerprint image may have a coordinate of the feature point and type information of the feature point and may be represented by (x, y, θ, type).

Recently, Juels and Sudan proposed a cryptographic method called fuzzy vault. In detail the operation process of the fuzzy bolt, user A first generates a polynomial p using the secret key k, and then calculates p (A) using its own A set that can be generated from the fingerprint feature points. Then, randomly adding false information (false feature points in case of fingerprint) to A set, R set (fingerprint template stored in case of fingerprint) is generated. User B uses his B set to obtain secret key k. If B set and A set are very similar (in case of fingerprint, if stored fingerprint feature set A and input fingerprint feature set B are very similar), B can obtain a sufficient number of p (A) information from the R set, remove false information through an error correction process, and obtain a secret key by solving the polynomial p. In particular, since the fuzzy bolt operates to solve the polynomial according to the similarity of the two sets, the feature points of the fingerprint information input from the same finger are suitable for the fingerprint recognition in which the order is changed, missing or added.

The registration processing module 110 extracts a registration feature point from a registration fingerprint image for registration, and generates a fingerprint template by inserting a false feature point into the extracted registration feature point. In other words, by applying a fingerprint fuzzy vault to generate a polynomial using the user's private key for registration, and randomly generates a false feature point to protect the registration feature by configuring the fingerprint template with a real minutiae. In this case, the registration processing module 110 generates a registration hash table for performing automatic alignment on the generated fingerprint template based on the geometric hashing method, and stores the registration hash table in the memory 132 as a database.

Similarly, the authentication processing module 120 extracts an authentication feature point from an authentication fingerprint image for authentication. In this case, even when the fingerprint of the same person is input, the coordinate value of the feature point may be translated and the direction may be rotated as shown in FIG. 3B whenever the fingerprint is input to the fingerprint input device. For the same person, FIG. 3A is a fingerprint image input when registering, and FIG. 3B is a fingerprint image input when authenticating. The end point 310 of FIG. 3A and the end point 330 of FIG. 3B are the same feature point pairs, and similarly, the branch point 320 of FIG. 3A and the branch point 340 of FIG. 3B are the same feature point pairs. In this way, all the feature points extracted from the two fingerprints can be identified based on the difference between the position and the direction, and all pairs of feature points determined to be similar can be found, or scores can be obtained according to the similarity to determine whether the threshold is greater than or equal to. This is a method of comparing absolute phases on the assumption that fingerprints are input in the same location and in the same direction. A process of correcting two fingerprints to have the same reference before checking similarity of location and direction may be added. have. That is, even in the same pair of feature points as shown in FIGS. 3A and 3B, an absolute coordinate value is different in direction according to an input time, and thus an alignment process of moving and rotating two fingerprints by a changed amount may be added. The authentication processing module 120 generates an authentication hash table for performing automatic alignment based on the geometric hashing method for the extracted and corrected authentication feature points. The registration processing module 110 and the authentication processing module 120 may be implemented as a software module.

Here, the reason for generating a hash table for performing the automatic alignment based on the geometric hashing method is that in order to measure the similarity between the stored fingerprint information and the input fingerprint information, the two fingerprints must first be aligned. This is because the information has been added and it is not easy to align the two fingerprints without being able to distinguish which of the user's fingerprint feature point (registration feature point) information and which is added false information.

As one of such geometric hashing methods, a method of inserting false feature points using a three-dimensional hash table and automatically performing fingerprint alignment on a transformed region where inverse transformation is impossible without additional information may be used. In this case, a model-based object recognition method may be used. Model-based object recognition is a method of finding a scene that contains the same object as an object from a set of objects. In this case, the object may be represented by a point or a line and the information of the object transformed in a geometric manner is called a feature, and the object information is configured by minimizing the feature. The object information to be compared with the object information thus made is AND or OR operation and the similarity can be confirmed according to the result.

Many object recognition methods use a method of predicting or verifying a feature between a scene and an object. Unlike the conventional method, the geometric hashing method processes the preprocessing step and the recognition step in parallel, so that the execution time of the recognition process is efficient. In particular, the feature of geometric hashing capable of recognizing some occluded objects is well suited to the situation of fingerprint fuzzy vaults, where a fingerprint feature point of a real user must be distinguished from a vault containing a large number of false information. The geometric hashing algorithm is divided into the preprocessing process (registration process) and the recognition process (identification process) as follows.

The pretreatment process is performed only once offline. First, feature points expressed as points of an object are stored in the form of a hash table containing a lot of information by geometric transformation in order to be stored in the database. For example, if n feature points of an object are stored in a database, one feature point is selected as a reference point and moved to the center of the coordinate axis by using a geometric transformation of rotation or movement along with other pairs of feature points. The remaining feature points are converted in the same way, and the transformed coordinates are marked by the bin defined in the hash table. The reference point is selected for all n feature points and the remaining feature points are converted in the same way.

In the recognition process performed online, the feature point S of the object to be recognized is extracted, and an arbitrary feature point is selected as a reference point and moved to the center of the coordinate axis by using a geometric transformation of rotation or movement together with other feature points that are paired. Then convert the remaining feature points in the same way. The generated table is compared with the hash table created during the preprocessing and increases the number of bins. Again, after selecting the reference point for all n features and converting the remaining feature points in the same way as above, the bin with the highest score is finally recognized as the most similar object. In particular, selecting a reference point may select one feature point and may be two or more depending on circumstances. In addition, the feature of the object may be represented by a line rather than a point.

The geometric hashing method described herein is merely an example, and the geometric hashing method applicable to the present invention is not limited to the above-described method.

The authentication module 130 authenticates an authentication fingerprint image when the number of matching conversion feature points is greater than or equal to a set value according to a comparison result of coordinates, angles, and types of each conversion feature point of the registration hash table and the authentication hash table, or the registration hash table and The authentication fingerprint image may be authenticated when the similarity of the conversion feature points of the authentication hash table is greater than or equal to a set value.

4 is a diagram illustrating an integrated implementation of a software module and a hardware module of the fingerprint purge bolt system of FIG. 1. That is, FIG. 4 illustrates an example of an integrated implementation system in which the registration processing module 110 and the authentication processing module 120 are implemented as software modules, and the authentication module 130 is implemented as hardware modules.

The registration feature registration process of the fingerprint fuzzy bolt system may be performed by extracting a fingerprint feature point, generating a false feature point, generating a fingerprint template, generating a registration fingerprint hash table, and storing a fingerprint database. Also, the fingerprint recognition process may be performed by extracting feature points of an input fingerprint and generating an authentication fingerprint hash table. At this time, the authentication fingerprint hash table generated in the fingerprint recognition process and the registration fingerprint hash table generated in the registration process are matched with each other to calculate the similarity of the transformed feature points and perform authentication on the input fingerprint.

FIG. 5 is a diagram illustrating fingerprint matching by the authentication module of FIG. 1. The authentication module 130 includes a memory 132 for storing a registration hash table and an authentication hash table, 2 ⁿ comparison units 134 (hereinafter also referred to as a matching module), and a registration fingerprint conversion feature point and an authentication fingerprint conversion. And a calculation unit 136 for calculating the number of matched transform feature points for the feature points. Here, each hash table consists of transform feature points. The registration fingerprint conversion feature point is a feature point generated by geometrical transformation of a fingerprint feature point of a user and an inserted false feature point in order to protect it. In addition, the authentication fingerprint conversion feature point is a feature point generated by geometric transformation of the feature point of the fingerprint for authentication. In this case, the registration hash table may be expressed as in Equation 1.

Where r is the number of enrolled fingerprint templates. In addition, tr _i is composed of a hash table, a selection based on the m _i of the registered fingerprint and the template generated by the geometric hashing to r-1 of converting a feature point.

In addition, the authentication hash table may be expressed as in Equation 2.

Here, s is the number of authentication fingerprint template, tr _j is composed of a hash table is selected based on the authentication of the fingerprint template and m _j generated by the geometric hashing to s-1 of converting a feature point.

In addition, when the memory 132 storing the registration hash table and the authentication hash table is implemented as one memory, the size of the memory is the sum of the number of conversion points p of the registration hash table and the number q of conversion feature points of the authentication hash table. It is preferable to have the above size. At this time, the number p of conversion feature points in the registration hash table is [r x (r-1)], and r is the number of registration fingerprint templates. The number q of conversion feature points in the authentication hash table is [s x (s-1)], and s is the number of authentication fingerprint templates.

6 is a flowchart illustrating a fingerprint authentication matching method by the authentication module of FIG. 1. First, the j-th authentication fingerprint conversion feature point tr _j from the authentication hash table V is input to the matching module 134 (S501 to S505). Further, the i-th registered fingerprint conversion feature point tr _i from the registration hash table E is input to the matching module 134 (S509, S511).

Next, the coordinates, angles, and types of the two converted feature points are compared (S513). The number of matched transform feature points is calculated (S515, 517), and tr _{i + 1} is input into the matching module 134 at E until i is equal to or greater than r-1, and then tr _j and tr _{i + 1} are compared. (S519). After comparing all the transform feature points of E, iterative comparison is performed by inputting tr _{j + 1} into the matching module 134 at V until j becomes greater than or equal to s-1.

For example, the hardware structure consisting of two matching modules 134 is as follows. First, tr ₀ and tr ₁ of the authentication hash table are input to the matching module 1 and the matching module 2, respectively, and tr ₀ of the registered hash table is input to the matching module 1 (matching module 1 and The matching module 2 is shown as Matching 1 and Matching 2 in FIG. 4). Matching of feature point conversion module 1, the two tr (tr ₀ tr ₀ and the registered hash table of authentication hash table) in the input are compared. The comparator 134 calculates the number of pairs of matched transform feature points. After that, tr ₀ of the registration hash table is transmitted from the matching module 1 to the matching module 2, and tr ₁ of the registration hash table is transmitted from the registration hash table to the matching module 1. Matching module 1 compares tr tr ₀ of ₁ and the authentication hash table, the hash table of the registration, the matching module (2) is the _one of tr tr ₀ and authenticated hash table, the hash table at the same time of the registration. After determining whether the transform feature points match, the of the registered hash table is transmitted to the next matching module. In this same manner, all trs of the registered hash table are executed until the matching module 2 compares with tr ₁ of the authentication hash table, and then tr ₂ and tr ₃ of the authentication hash table are matched with the matching module 1. It is input to the matching module 2. The tr ₀ of the registration hash table is transmitted to the matching module 1 to perform the above process. It repeats until tr _{s -1} of the authentication hash table is input to the matching module 2 and compared with tr _{r -1} of the registration hash table.

Experimental results show that the hardware architecture with two matching modules is 0.20 seconds for 36 fingerprint features and 200 false features for 0.43 seconds for 400 false features. On the other hand, when applied only in software under the same conditions, respectively, it was 1.13 seconds and 2.20 seconds, respectively.

100: fingerprint purge bolt system 110: registration processing module
120: authentication processing module 130: authentication module

Claims (11)

Extracts a registration feature point from a registration fingerprint image for registration, generates a fingerprint template by inserting a false feature point into the extracted registration feature point, and registers to perform the automatic alignment on the generated fingerprint template based on a geometric hashing method. A registration processing module for generating a hash table and storing the hash table in a database;
An authentication processing module for extracting an authentication feature point from an authentication fingerprint image for authentication and generating an authentication hash table for performing automatic alignment on the extracted authentication feature point based on a geometric hashing method; And
An authentication module for comparing similarity between the conversion feature point of the registration hash table and the conversion feature point of the authentication hash table and calculating similarity, and determining whether to authenticate the authentication fingerprint image according to the result.
VLSI structure of the fingerprint purge bolt system comprising a. The method of claim 1,
The registration processing module and the authentication processing module VLSI structure of the fingerprint fuzzy bolt system, characterized in that implemented as a software module. The method of claim 1,
The authentication module is a VLSI structure of the fingerprint fuzzy bolt system, characterized in that implemented as a hardware module. The method of claim 1,
The authentication module,
A memory for storing the registration hash table and the authentication hash table;
A comparison unit for comparing coordinates, angles, and types of the conversion feature points of the registration hash table and the authentication hash table; And
A calculation unit for calculating a number of matching conversion feature points of the registration hash table and the authentication hash table,
And wherein the authentication fingerprint image is authenticated when the number of matching conversion feature points is greater than or equal to a set value. The method of claim 4, wherein
The registration hash table is a VLSI structure of a fingerprint fuzzy bolt system, characterized in that:

Here, r is the number of enrolled fingerprint templates, and tr _i is a hash table generated based on m _i among enrolled fingerprint templates and generated through geometric hashing, consisting of r-1 transform feature points. The method of claim 4, wherein
The authentication hash table is a VLSI structure of a fingerprint fuzzy bolt system, characterized in that:

Here, s is a number of authentication of the fingerprint template, tr _j is selected based on the m _j of the authentication fingerprint template and a hash table created by the geometric hashing constituted by any s-1 of converting a feature point. The method of claim 4, wherein
And the size of the memory is equal to or greater than the sum of the number p of conversion feature points of the registration hash table and the number q of conversion feature points of the authentication hash table. The method of claim 7, wherein
The number p of conversion feature points of the registration hash table is [rx (r-1)], and r is the number of registration fingerprint templates. The method of claim 7, wherein
The number q of conversion feature points of the authentication hash table is [sx (s-1)], and s is the number of authentication fingerprint templates. The method of claim 4, wherein
Input the j th authentication fingerprint conversion feature point tr _j in the authentication hash table into the comparison unit, the i th registration fingerprint conversion feature point tr _i in the registration hash table, and input the coordinates of the two conversion feature points, Compute the number of matched feature points by comparing the angles and types, input tr _{i + 1} to the comparison unit in the registered hash table, and compare the number of matched feature points by comparing tr _j and tr _{i + 1} . And tr _{i + 1} in the authentication hash table by entering the comparison unit tr _i And repeating the comparison with tr _{i + 1} . Inputting the j th authentication fingerprint conversion feature point tr _j in the authentication hash table to the comparison unit;
Inputting an i th registration fingerprint conversion feature point tr _i in the registration hash table into the comparison unit;
Calculating the number of matched transform feature points by comparing coordinates, angles, and types of the input transform feature points;
Inputting tr _{i + 1} into the comparison unit in the registration hash table and comparing tr _j and tr _{i + 1} to calculate the number of matched transform feature points; And
In the authentication hash table tr _{j + 1} input to the comparison unit tr _i And performing an iterative comparison with tr _{i + 1}
Fingerprint authentication matching method by a fingerprint fuzzy bolt system comprising a.

Images