Biometric recognition using a classification neural network

• US 5,892,838 A
• Filed: 06/11/1996
• Issued: 04/06/1999
• Est. Priority Date: 06/11/1996
• Status: Expired due to Fees

• Current Assignee: Minnesota Mining and Manufacturing Co (3M Company)
• Original Assignee: Minnesota Mining and Manufacturing Co (3M Company)
• Inventors: Brady, Mark J.
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/664,215
• Time in Patent Office: 1,029 Days
• Field of Search: 382/224, 382/124, 382/125, 382/209, 382/218, 382/115, 382/190, 382/156, 382/159, 382/157, 382/278
• US Class Current: 382/115
• CPC Class Codes: G06V 40/1347 Preprocessing; Feature extraction

Abstract

A biometric recognition system involves two phases: creation of a master pattern set of authorized users biometric indicia and authentication using a classification neural network. To create the master pattern set, an image of an authorized biometric user'"'"'s indicia is divided into a plurality of regions of interest or "features". The system determines which features are the most useful for identification purposes. Master patterns are then created from these master features, thus creating a master pattern set. During authentication, a sample pattern set of a user to be authenticated is similarly created. A neural network is used to compare the sample pattern set with the master pattern set to determine whether the user should be authenticated.

Claims

1. A biometric recognition system for authenticating biometric indicia of a user, the system comprising:

data storage means for storing a plurality of master pattern sets, each master pattern set corresponding one of a plurality of authorized users, each of the master pattern sets defined by a plurality of master features and master orientation data of the plurality of master features;

vector generation means for producing a comparison vector representing the level of similarity between a master pattern set and the biometric indicia; and

a neural network for producing classification designators based on the comparison vector, wherein the classification designators are indicative of whether the user'"'"'s biometric indicia should be authenticated;

wherein the vector generation means comprises;

identification means for identifying sample features in the biometric indicia that best match each of the master features;

pattern generation means for generating sample orientation data based on the sample features that best match each of the master features; and

means for comparing the master orientation data and the sample orientation data to produce comparison orientation data;

wherein the comparison vector is based on the comparison orientation data and is also based on the similarity of the master features and their corresponding sample features; and

the master pattern sets are derived using the following equation;

##EQU5## where S is the set of all m-by-m features in an image, with the exception of R, which is a reference feature;

R_ij is an (i, j)th pixel gray level in feature R;

I_ij i s an (i, j)th pixel gray level in I; and

R and I are mean gray levels within the respective features.

2. The biometric recognition system according to claim 1, wherein the identification means uses a correlation function to determine the similarity of the master features and their corresponding sample features;

wherein the correlation function is ##EQU6## where S is the set of all m-by-m features in the image or subimage;

wherein R is a feature from the master pattern image and is compared with each candidate feature, I, from the sample image;

R_ij is the (i, j)th pixel gray level in master feature R and I_ij is the (i, j)th pixel gray level in sample feature I and R and I are the mean gray levels within the respective features.

3. The biometric recognition system according to claim 1, wherein the master orientation data and the sample orientation data include line lengths between the master features and between the sample features and slope data of the line lengths.

4. The biometric recognition system according to claim 3, wherein the comparison orientation data is produced by taking the difference between line lengths of the master features and corresponding line lengths of the sample features and between corresponding slope data of the corresponding line lengths.

5. The biometric recognition system according to claim 4, wherein the comparison orientation data is produced by taking a ratio between line lengths of the master features and corresponding line lengths of the sample features and a difference between corresponding slope data of the corresponding line lengths.

6. The biometric recognition system according to claim 1, further comprising biometric pattern acquisition means for acquiring an image of the biometric indicia and creating therefrom a sample image.

7. The biometric recognition system according to claim 6, wherein the data storage means further stores data relating to the biometric indicia.

8. The biometric recognition system according to claim 7, wherein the data includes expected locations of the master features in the sample image.

9. The biometric recognition system according to claim 8, wherein the identification means searches the expected locations of the sample image when identifying sample features that best match each of the master features using the equation of claim 1.

10. The biometric recognition system according to claim 1, wherein the vector generation means produces a plurality of comparison vectors, the system further comprising threshold means for receiving classification designators from the neural network, summing the number of classification designators indicating a match to produce a sum and comparing the sum with a threshold value to output a match signal when the sum exceeds the threshold value.

11. The biometric recognition system according to claim 10, wherein the threshold value is based on frequency density functions.

12. The biometric recognition system according to claim 1, further comprising interface means for providing an interface with the user.

13. A method for generating a master pattern set of a biometric indicia of a user, comprising the steps of:

(a) acquiring an image of the biometric indicia to produce a biometric pattern image, wherein the biometric pattern image includes a plurality of features;

(b) comparing each feature in the biometric pattern image with all other features in the biometric pattern image;

(c) assigning a uniqueness value to each feature in the biometric pattern image based on the results of comparing step (b);

(d) choosing, based on the uniqueness values, from the features in the biometric pattern image a plurality of master features;

(e) defining master patterns based on the master features;

(f) storing the master patterns and master features as the master pattern set,

wherein assigning step (c) uses the following equation;

##EQU7## where S is the set of all m-by-m features in an image, with the exception of R, which is a reference feature;

R_ij is an (i, j)th pixel gray in feature R;

I_ij is an (i, j)th pixel gray level in I; and

R and I are mean gray levels within the respective features.

14. The method according to claim 13, wherein the master patterns are defined based on the master features and on orientation data describing location relationships between the master features using the equation of claim 13.

15. The method according to claim 14, wherein the orientation data includes line lengths between the plurality of master features and slope data of the line lengths.

16. The method according to claim 13 wherein the uniqueness value for a feature is defined based on the uniqueness of the feature with respect to the other features and on variance within the feature.

17. The method according to claim 13 wherein the master features are chosen based on the uniqueness values and is also based on which features have the highest uniqueness values in an area of the biometric pattern image.

18. A method for authenticating a biometric pattern of a user, in which authorized biometric patterns are stored and represented by corresponding master pattern sets comprised of a plurality of master patterns based on a plurality of master features in the corresponding biometric pattern, comprising the steps of:

(a) acquiring an image of a biometric pattern to be authenticated and producing therefrom a sample pattern image, wherein the sample pattern image includes a plurality of sample features;

(b) retrieving a master pattern set;

(c) comparing each master feature in the retrieved master pattern set with the sample features and determining therefrom which sample feature best matches each master feature, and producing therefrom a set of matched sample features;

(d) defining sample patterns based on the set of matched sample features;

(e) comparing the sample patterns with the master patterns;

(f) authenticating the sample pattern image upon a favorable comparison step (e),

wherein comparing step (c) uses the following equation;

##EQU8## where S is the set of all m-by-m features in an image, with the exception of R, which is a reference feature;

R_ij is an (i, j)th pixel gray level in feature R;

I_ij is an (i, j)th pixel gray level in I; and

R and I are mean gray levels within the respective features.

19. The method according to claim 18 wherein comparing step (e) further comprises the steps of:

(e1) comparing the sample patterns with the master patterns to generate pattern difference vectors;

(e2) analyzing the pattern difference vectors and producing therefrom classification designators indicative of whether the sample and master patterns match;

(e3) determining that the sample patterns match the master pattern set if the classification designators exceed a threshold.

20. The method according to claim 19 wherein a neural network analyzes the pattern difference vectors.

21. The method according to claim 18 further including the step of receiving user identification information.

22. The method according to claim 21 wherein the master pattern set retrieved in step (b) is chosen based on the user identification information.

23. A method for authenticating a biometric pattern of a user, comprising the steps of:

storing a master pattern set from the user, the master pattern set defined by a plurality of master features and master orientation data of the plurality of master features;

acquiring a sample biometric pattern of a user to be authenticated;

identifying sample features in the sample biometric pattern that best match each of the plurality of master features;

generating sample orientation data based on a pattern generated by the identified sample features;

comparing the master orientation data and the sample orientation data to produce comparison orientation data;

producing a comparison vector based on the similarity of the plurality of master features and their corresponding identified sample features and based on the comparison orientation data;

producing classification designators based on the comparison vector; and

authenticating the biometric pattern if the classification designators indicate a match,

wherein the comparing step uses the following equation;

##EQU9## where S is the set of all m-by-m features in an image, with the exception of R, which is a reference feature;

R_ij an (i, j)th pixel gray level in feature R;

I_ij is an (i, j)th pixel gray level in I; and

R and I are mean gray levels within the respective features.