KR20030038306A - Method for Identifying Biological Information Using Bipartite Matching - Google Patents

Abstract

PURPOSE: A biometric authentication method is provided to determine more rapid matching combinations and to reduce a matching time and calculation amount by avoiding a comparison of all the feature points of a target image with those of a reference image. CONSTITUTION: The method comprises several steps. Data on feature points are extracted, and graphically converted on a two dimensional plane(S102). One feature value is calculated for the data on feature points(S104). A similarity matrix is generated(S106), and a Bipartite matching algorithm is applied to the similarity matrix(S108). If matching combinations are determined, a similarity among the matching combinations are calculated, it is determined whether a target image is identical to a reference image, and a biometric authentication is made according to the determination result(S110).

Description

Method for Identifying Biological Information Using Bipartite Matching}

The present invention relates to a biometric information authentication method, and more particularly, to a method for authenticating biometric information by determining a matching combination faster by applying a bipartite matching.

Recently, as biometric technology is applied to various online and offline security technologies, the demand is rapidly increasing.

Biometric technology extracts biosignals such as fingerprint, iris, face, voice, vein, and palm print, stores them as biosignals of the reference image, and saves them when the biosignals of the target image requesting authentication are received. It is a method of authenticating whether or not it is the same person by comparing whether it matches a signal.

The most important point of the biometric technology is to arrange the feature point set of the target image as an input to match the feature point set of the reference image. For example, when 100 feature points exist in a reference image and 100 feature points exist in a target image, it is determined how to calculate the similarity by matching 100 feature points.

To this end, conventionally, one feature point of the target image is compared with all feature points of the reference image to find a combination that best matches each other to determine each matching combination. In the case of using the conventional method, the matching combination can be determined only after many calculations, N !, when there are N feature points.

In the past, many of these calculations required a lot of time for the authentication process, which has been a major factor in the rise in hardware prices.

In the related art, when a number of users increases to shorten the matching time, a method of assigning a separate user number has been used, but this is contrary to the basic purpose of biometrics not to use a user ID. There is a need for an algorithm to determine the matching combination of information.

In the present invention, to solve the problems of the prior art as described above, it is proposed a biometric information authentication method that can determine a matching combination of biometric information faster without a lot of calculation.

It is still another object of the present invention to propose a biometric information authentication method which can determine a matching combination of biometric information at high speed by applying viperite matching to a biometric information matching operation.

1 is an overall flowchart illustrating a process of recognizing biometric information using viperite matching according to a preferred embodiment of the present invention.

2 is a diagram illustrating a result of extracting feature points from a photographed fingerprint image;

3 illustrates an example of graphical representation of feature points using a minimum cost tree algorithm.

4 is a diagram illustrating an example of calculating the similarity between feature points of a reference image and feature points of a target image according to an exemplary embodiment of the present invention.

FIG. 5 is a flowchart illustrating a method of calculating similarity between feature points of a reference image and a target image input to a similarity matrix according to a preferred embodiment of the present invention. FIG.

FIG. 6A illustrates a matrix representing a result of calculating edge feature values for a reference image having the feature point tree structure as shown in FIG. 4, and FIG. 6B illustrates an edge for a target image having the feature point tree structure as shown in FIG. 4. A matrix showing the results of calculating feature values.

FIG. 7 illustrates an example of the similarity matrix implemented through the process of FIG. 5.

8 is a flow chart illustrating a method of applying a viperite match to a similarity matrix to determine an appropriate match combination.

9A-9F illustrate an example of applying a Hungarian method using a similarity matrix.

In order to achieve the above object, the biometric information authentication method using the viperite matching according to the present invention comprises the steps of: (a) detecting the feature points from the photographed biometric information image; (B) converting the feature points detected in the step (a) into a two-dimensional graph format; Calculating (c) feature information of each feature point detected in step (a) as one feature value; (D) calculating edge feature values between the feature points using the feature values of the feature points calculated in step (c); (E) generating a similarity matrix by calculating similarity between feature points of the pre-stored biometric information image and feature points of the photographed biometric information image using the edge feature value calculated in step (d); (F) determining a matching combination of feature points of the pre-stored biometric image and feature points of the photographed biometric information by applying viperite matching to the similarity matrix generated in step (e); And (g) determining whether the photographed biometric information image is the same as the previously stored biometric information image by calculating similarity between the matching combinations determined in the step (f).

In step (b), the feature values may be converted into a graph using a minimum cost tree algorithm.

In the step (c), the edge feature value of the specific feature point may be calculated using the difference between the feature value of the other feature points connected to the specific feature point and the feature value of the specific feature point in the graph of the step (b).

The feature point detected in step (a) may be the ending and viperation of the fingerprint or the tear in the iris.

The feature value including the feature information of the feature point may be a sum of values set for the type, angle, and distance of the fingerprint.

The feature value including the feature information of the feature point may be a sum of values set for the type, angle, and distance of the iris rupture.

The step (e) may include determining edge feature values of specific feature points of the photographed biometric information image; Determining edge feature values of a specific feature point of a reference image to calculate similarity with respect to the specific feature point; Determining whether edge feature values of the determined captured biometric information image exist at edge feature values of specific feature points of the reference image; Maintaining the edge feature value when the edge feature value of the photographed biometric information image is present in the edge feature value of the reference image, and resetting the edge feature value by multiplying the edge feature value by −1 if it does not exist. ; And calculating similarity by adding the reset edge feature values.

Step (f) may apply viperite matching using the Hungarian method.

The method of applying viperite matching using the Hungarian method includes: subtracting each row of the similarity matrix to a minimum value of each row (S1); Subtracting each column of the similarity matrix to a minimum value of each column (S2); Determining a minimum number of lines including all zeros present in the similarity matrix after subtracting the minimum of each row and each column (S3); If the minimum line number coincides with the number of feature points, determining a feature combination corresponding to each zero as a matching combination (S4); If the minimum number of lines does not match the number of feature points, determining a minimum value among values not included in the line including zero and subtracting values not included in the line as the minimum value; Increasing the value of the point at which the line intersects by the minimum value (S6); If the minimum number of lines is determined again so as to include all zeros, and if the determined minimum number of lines coincides with the number of the feature points, the combination of feature points corresponding to each zero is determined as a matching combination, and if not, the step (S5) to (S6) may include repeating the rule.

Hereinafter, exemplary embodiments of a biometric information authentication method using viperite matching according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a process of recognizing biometric information using viperite matching according to a preferred embodiment of the present invention.

As shown in FIG. 1, first, feature point information of biometric information is extracted (S100). Here, the biometric information may include fingerprint, iris, face, vein information, and the like. The feature point information refers to information on a part from which the feature can be extracted from the information such as the fingerprint iris. For example, in the case of the iris, information on the type and location of the tears scattered in the iris, and in the case of the fingerprint, may include information about the ending and the vibration of the fingerprint.

The tear information of the iris may extract the location of the feature point by analyzing the contrast information of the iris image information input to the camera. Contrast information can be extracted using a gray scale or the like.

In order to extract feature point information of a fingerprint, a preprocessing process of extracting a ridge input from a camera is first performed. To extract the ridge of the fingerprint, the fingerprint image is binarized using a gray scale or the like, and after the ridge portion is finely converted from the binarized image, feature points such as viperation and edging are extracted.

2 is a diagram illustrating a result of extracting feature points from a photographed fingerprint image.

After extracting the information on the feature point, the extracted feature point is converted into a graph form in a two-dimensional plane (S102).

According to a preferred embodiment of the present invention, a minimum cost tree algorithm is used to represent the feature points in graphical form. The minimum cost tree algorithm refers to an algorithm that sets each feature point as one node and connects nodes so that the distance between nodes is minimized.

FIG. 3 illustrates an example in which feature points are expressed in a graph form using a minimum cost tree algorithm.

FIG. 3 is a minimum cost algorithm applied to the feature points of the fingerprint shown in FIG. 2, and as shown in FIG. 3, each feature point is connected by a line of the minimum distance between the nodes.

As such, expressing feature points in graphical form is for calculating edge feature values between connected feature points.

After the feature points are converted into a graph format, the feature information is calculated as one feature value (S104).

In the case of a fingerprint, as characteristic information on a feature point, there is feature information on the type of the feature point (ending and biperation), an angle with a reference feature point, and a distance from the reference feature point. In the case of the iris may include information about the type of tear, the location of the tear. Conventionally, these feature information are individually compared with respect to the pre-stored reference image and the target image to be authenticated, but the present invention calculates one feature value that combines the individual feature information.

For example, if the type of the feature point is X, the angle from the reference feature point is Y, and the distance from the reference feature point is Z, the calculated value of X + Y + Z is calculated as the feature value.

In case of equity, there are two types of feature points, ending and perforation. Therefore, -1 for ending and -1 for ending, the angle can be set as it is, and the distance can be set by multiplying the real distance by a certain number. There will be.

According to another embodiment of the present invention, the feature value may be calculated by weighting each feature information when the feature value is calculated.

If it is determined that the type information of the feature point is the most important among the information on the type, angle, and distance of the feature point, the feature value is calculated by giving a relatively large weight to the feature point type and a relatively small weight to the distance or angle.

When the weight for the type is C1, the weight for the angle is C2, and the weight for the distance is C3, the feature value may be calculated by XC1 + YC2 + ZC3 in consideration of these weights.

After calculating each feature information as one feature value, a similarity matrix is generated (S106).

The similarity matrix is calculated by expressing each similarity between the feature points of the reference image and the feature points of the target image as a matrix.

For example, if there are five feature points of the reference image from M1 to M5 and four feature points of the target image are from N1 to N4. Similarly, the similarity between the combinations is calculated by considering all combinations that may occur between the feature points of the reference image and the feature points of the target image. A detailed method of calculating the similarity will be described later with a separate drawing.

After generating the similarity matrix, viperite matching is performed on the pseudo-earth matrix (S108).

Viperite matching includes a variety of methods, including the Ford-Fulkerson method, the Edmonds-Karp method, the Dinitz method, the Goldberg and Tarjan method, and the Hungarian method. It may be performed using the method. According to a preferred embodiment of the present invention, it is preferred to perform viperite matching using a Hungarian method in which the calculation can be performed quickly.

When viper type matching is performed, a matching combination may be determined by determining which feature point of the target image matches the feature point of the target image.

When the matching combination is determined, the degree of similarity between the matching combinations is calculated to determine whether the reference image and the target image match, and the biometric information is authenticated (S110).

4 is a diagram illustrating an example of calculating a similarity between feature points of a reference image and feature points of a target image according to an exemplary embodiment of the present invention, and FIG. 5 is a reference image input to a similarity matrix according to an exemplary embodiment of the present invention. And a method for calculating the similarity between the feature points of the target image.

As shown in FIG. 5, edge feature values are first calculated for the feature points of the reference image (S500). The edge feature value is a value representing the similarity with other feature points connected to one feature point.

In the case of FIG. 4, the reference image has five feature points of M1 to M5, and the target image has four feature points of N1 to N4. Since M2 is connected to M1 and M3, edge feature values with M1 and edge feature values with M3 can be calculated. In FIG. 4, the edge characteristic value with M1 of M2 is shown by a1, and the edge characteristic value with M3 of M2 is shown by a3.

According to a preferred embodiment of the present invention, the edge feature values may be calculated by the difference of the feature values calculated for each feature point. For example, when the feature value of M2 is W2 and the feature value of M1 is W1, the edge feature value a2 between M2 and M1 is Will be able to calculate by. It will be apparent to those skilled in the art that calculating the edge feature value as the difference between the feature values of the corresponding two feature points is only one embodiment of the invention and the edge feature value can be calculated in other ways.

As described above, other feature point information connected to all the feature points of the reference image is grasped, and edge feature values with the connected feature points are calculated.

After the edge feature values are calculated for the reference image, the edge feature values of the feature points of the target image are calculated (S502).

The method of calculating the edge feature of the target image is the same as the method of calculating the edge feature of the reference image.

In the case of Fig. 4, since feature point N2 is connected to N1 and N3, it has edge feature values between N1 and N3. As described above, the edge feature value b1 between N1 and N2 is calculated as the difference between the feature values of N1 and N2, and the edge feature value b2 between N2 and N3 is calculated as the difference between the feature values of N2 and N3.

FIG. 6A illustrates a matrix representing a result of calculating edge feature values for a reference image having the feature point tree structure as shown in FIG. 4, and FIG. 6B illustrates an edge for a target image having the feature point tree structure as shown in FIG. 4. A matrix showing the results of calculating the feature values is shown.

As illustrated in FIGS. 6A and 6B, when the edge feature values are calculated and stored as a matrix, edge feature value information related to a specific feature point may be quickly extracted. For example, if a line is drawn to M2 to extract an edge feature value related to M2, which is one of the feature points of the reference image, it can be seen that the edge feature values related to M2 are a1, a2, and a3.

In order to calculate the similarity between the specific feature point of the reference image and the specific feature point of the target image, a process of setting a variable such as setting the number of edge feature values related to the specific feature point of the reference image to n is performed (S504).

In FIG. 4, when the similarity between M2 and N2 is to be calculated, n = 3 is set since the edge feature values related to M2 are three of a1, a2, and a3.

When the parameter setting is completed, it is determined whether a first edge feature value with respect to a specific feature point of the reference image exists in the target image (S504). When calculating the similarity between M2 and N2 in FIG. 4, it is determined whether an edge feature value corresponding to a1 exists in the edge feature value of N2 when the first edge feature value is a1. According to an embodiment of the present invention, it is possible to determine whether an edge feature value having a difference between the value of a1 and a predetermined threshold or less exists in the edge feature value related to N2.

In FIG. 4, since a1 and b1 have similar angles and positions, it may be determined that there is an edge feature value b1 corresponding to a1.

If there is an edge feature value of the target image corresponding to the first edge feature value, the first edge feature value is maintained as it is (S508).

If there is no edge feature value of the target image corresponding to the first edge feature value, the value obtained by multiplying the first edge feature value by −1 is reset to the first edge feature value (S510).

After determining whether the first edge feature value exists in the target image, if the second edge feature value exists, the same process is repeated for the second edge feature value.

When a3, which is one of the edge feature values of M2 of FIG. 4, is set as the second edge feature value, it is determined whether an edge feature value having a difference between the value of a3 and a predetermined threshold or less exists in the edge feature value associated with N2. .

In FIG. 4, since the edge feature similar to the angle of a3 is not present at N2, it will be determined that a3 is not present at the edge feature of N2, and in this case, as described above, The value multiplied by 1 is reset to the second edge feature value.

After resetting the edge feature values for a specific feature point of the reference image by the method described above, all the reset feature values are added, and the value is set as the similarity between the specific feature point of the reference image and the specific feature point of the target image. (S512).

In the process of calculating the similarity between M2 and N2 in FIG. 4, edge feature values corresponding to a1 and a2 exist in N2 and edge feature values corresponding to a3 for edge features a1, a2, and a3 related to M2. If not present, the similarity may be calculated as a1 + a2-a3 as shown in FIG. 4.

The value set as the similarity is input to the similarity matrix, and the process illustrated in FIG. 5 is performed for all combinations that may occur between the feature point of the reference image and the feature point of the target image to implement the similarity matrix.

FIG. 7 illustrates an example of the similarity matrix implemented through the process of FIG. 5.

As shown in FIG. 7, the similarity matrix calculated for all combinations of the feature points of the reference image and the feature points of the target image is input to the similarity matrix.

8 is a flow chart illustrating a method of determining an appropriate match combination by applying viperite matching to a similarity matrix.

FIG. 8 illustrates an example of performing viperite matching using a Hungarian method among viperite matching, and it will be apparent to those skilled in the art that various other viperite matching methods may be used as described above.

As shown in FIG. 8, first, values in each row are subtracted from the similarity matrix as the minimum value of each row (S800).

FIG. 9A illustrates an example of subtracting the values in each row from the similarity matrix illustrated in FIG. 7 by the minimum value of each row.

After subtracting each row to the minimum value of each row, the values in each column are subtracted to the minimum value of each column (S802). FIG. 9B illustrates an example of subtracting the values in each column from the similarity matrix shown in FIG. 9A by the minimum value of each column.

The minimum number of lines that may include all zeros in the similarity matrix changed through the steps S800 to S802 is determined (S804). 9C shows an example of determining the minimum number of lines including zero in the changed similarity matrix.

If the minimum number of lines that may include zero matches the feature score, a combination corresponding to each zero existing in the matrix is determined as a matching combination (S808).

If the minimum number of lines that may include 0 does not match the number of feature points, in step S804, a minimum value among values not included in the line is determined (S810).

If the minimum value is determined, the values not included in the line are subtracted to the minimum value (S812). FIG. 9D illustrates a matrix of values not included in a line by 1, which is a minimum value not included in a line in the matrix of FIG. 9C.

After subtracting values not included in the line to the minimum value, the value at the point where the line crosses is increased by the minimum value (S814). FIG. 9E illustrates a matrix in which the value of the point where the lines intersect in the matrix of FIG. 9D is increased by a minimum value of 1. FIG.

Thereafter, step S806 is again performed to determine the minimum number of lines to include all zeros in the changed matrix.

If the minimum number of lines matches the number of feature points, a combination corresponding to each zero existing in the matrix is determined as a matching combination (S808). FIG. 9F shows an example of a line drawn to include all zeros in the matrix shown in FIG. 9E.

In FIG. 9F, since the number of feature points of the reference image and the target image is five and the minimum number of lines including all zeros is five, a matching combination may be determined. Since the combination corresponding to each zero is a matching combination, in Fig. 5, (M1, N1), (M2, N2), (M3, N4), (M4, N3), and (M5, M5) are selected as matching combinations.

When the matching combinations are determined, the similarity of each matching combination is calculated, and the added value is determined to authenticate the biometric information of the user corresponding to the target image.

As described above, according to the biometric information authentication method using the viperite matching according to the present invention, it is not necessary to determine the matching combination by comparing all combinations of the feature points of the reference image and the feature points of the target image, and thus the time required for matching. And there is an advantage that can greatly reduce the amount of calculation.

Claims (10)

In the method for authenticating biometric information, Detecting a feature point from the photographed biometric information image (a); (B) converting the feature points detected in the step (a) into a two-dimensional graph format; Calculating (c) feature information of each feature point detected in step (a) as one feature value; (D) calculating edge feature values between the feature points using the feature values of the feature points calculated in step (c); (E) generating a similarity matrix by calculating similarity between feature points of the pre-stored biometric information image and feature points of the photographed biometric information image using the edge feature value calculated in step (d); (F) determining a matching combination of feature points of the pre-stored biometric image and feature points of the photographed biometric information by applying viperite matching to the similarity matrix generated in step (e); And And (g) determining whether the photographed biometric information image is the same as the previously stored biometric information image by calculating similarity between the matching combinations determined in the step (f). Information authentication method. The method of claim 1, The step (b) is a biometric information authentication method using viperite matching, characterized in that for converting the feature values in a graph form using a minimum cost tree algorithm. The method of claim 1, In the step (c), the edge feature value of the specific feature point is calculated using the difference between the feature value of the feature value of the specific feature point and other feature points connected to the specific feature point in the graph of the step (b). Biometric information authentication method using the viperite matching. The method of claim 1, The method of detecting biometric information using viperite matching, characterized in that the feature points detected in step (a) are the ending and the vibration of the fingerprint. The method of claim 1, The feature point detected in step (a) is a biometric information authentication method using viperite matching, characterized in that the tear of the iris. The method of claim 1, The feature value including the feature information of the feature point is a sum of the values set for the type, angle, and distance of the fingerprint biometric information authentication method using viperite matching. The method of claim 1, The feature value including the feature information of the feature point is a biometric information authentication method using viperite matching, characterized in that the sum of the value set for the type, angle, distance of the iris rupture. The method of claim 1, Step (e), Determining edge feature values of specific feature points of the photographed biometric information image; Determining edge feature values of a specific feature point of a reference image to calculate similarity with respect to the specific feature point; Determining whether edge feature values of the determined captured biometric information image exist at edge feature values of specific feature points of the reference image; Maintaining the edge feature value when the edge feature value of the photographed biometric information image is present in the edge feature value of the reference image, and resetting the edge feature value by multiplying the edge feature value by −1 if it does not exist. ; And And calculating similarity by adding the reset edge feature values. The method of claim 1, Step (f) is a method of authenticating biometrics using viperite matching, characterized in that to apply the viperite matching using the Hungarian method. The method of claim 9, The method of applying viperitet matching using the Hungarian method, Subtracting each row of the similarity matrix to a minimum value of each row (S1); Subtracting each column of the similarity matrix to a minimum value of each column (S2); Determining a minimum number of lines including all zeros present in the similarity matrix after subtracting the minimum of each row and each column (S3); If the minimum line number coincides with the number of feature points, determining a feature combination corresponding to each zero as a matching combination (S4); If the minimum number of lines does not match the number of feature points, determining a minimum value among values not included in the line including zero and subtracting values not included in the line as the minimum value; Increasing the value of the point at which the line intersects by the minimum value (S6); If the minimum number of lines is determined again so as to include all zeros, and if the determined minimum number of lines coincides with the number of the feature points, the combination of feature points corresponding to each zero is determined as a matching combination, and if not, the step (S5) to (S6) rules of the biometric information applying the viperite matching, characterized in that it comprises the step of repeating.