Method for comparing features extracted from images of fingerprints

• US 7,986,820 B2
• Filed: 10/19/2001
• Issued: 07/26/2011
• Est. Priority Date: 10/19/2001
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Inventors: Moghaddam, Baback
• Primary Examiner(s): Chawan; Sheela C
• Application Number:
  US10/087,409
• Publication Number:
  US 20030076985A1
• Time in Patent Office: 3,567 Days
• Field of Search: 382/115, 382/124, 382/125, 382/173, 382/181, 382/190, 382/192, 382/266, 382/254, 382/272, 382/270, 382/289, 382/276, 382/286, 382/195, 382/201, 382/228, 382/206, 382274-275, 382/296, 382/295, 382/224, 382/260, 382/301, 382/395, 340/5.82, 340/5.83, 356/71, 705/53
• US Class Current: 382/124
• CPC Class Codes: G06V 40/1365 Matching; Classification

Abstract

Features are extracted from a test and reference image to generate a test and reference record. Each feature has a location, and orientation, and furthermore, the features of the reference records also have associated weights. The features of the test record are approximately aligned with the features of the reference record. Then, differences between the locations and orientations of the features of the reference record and the features of the test record are measured, and the weights of all features of the reference record that are less than a predetermined difference when compared with the features of the test record are summed to determine a similarity score that the test record matches the reference record.

Claims

1. A method for comparing features of a test record with features of a reference record, each feature having a location and orientation, comprising;

determining a weight for each feature of the reference record;

aligning the features of the test record with the features of the reference record;

measuring differences between the locations and orientations of the features of the reference record and the features of the test record; and

summing the weights of all features of the reference record that are less than a predetermined difference when compared with the features of the test record to determine a similarity score that the test record matches the reference record.

2. The method of claim 1 further comprising:

comparing the features of the test record with a plurality of reference records to determine a plurality of similarity scores; and

selecting a particular reference record as a candidate reference record if the corresponding similarity score is greater than a maximum threshold.

3. The method of claim 1 further comprising:

extracting the features from images of fingerprints.

4. The method of claim 1 wherein the alignment is a rigid transformation including global translation and rotation.

5. The method of claim 1 wherein the alignment is a rigid transformation including only global translation to approximately align the features of the test record with the features of the reference record.

6. The method of claim 5 wherein a probability of matching features is represented by a zero-mean Gaussian function ƒ

(0;

σ

²), where σ

² is a variance of the function.

7. The method of claim 6 wherein the similarity score S for a test record T of M features and a reference record R of N features is determined by

8. The method of claim 1 further comprising:

measuring differences only between pairs of features that are approximately aligned.

9. The method of claim 1 wherein the determining step further comprises;

identifying, for each feature in the reference record, a local neighborhood of features; and

setting the weight of each feature proportional to a function of distance between each feature and the local neighborhood of features.

10. The method of claim 9 wherein the function for weight w_i and k features in the local neighborhood is

11. The method of claim 9 wherein the distance function is an arithmetic mean.

12. The method claim 9 wherein the distance function is a geometric mean.

13. The method of claim 9 wherein the distance function is maximum distance.

14. The method of claim 1 wherein the sum of the weights of the features of the reference record is normalized to one.

15. A method for comparing features of a test record with features of a reference record, each feature having a location and orientation, comprising;

determining a weight for each feature of the reference record, the determining further comprising;

identifying, for each feature in the reference record, a local neighborhood of features; and

setting the weight of each feature proportional to a function of distances between each feature and the local neighborhood of features;

aligning the features of the test record with the features of the reference record;

measuring differences between the locations and orientations of the features of the reference record and the features of the test record; and

summing the weights of all features of the reference record that are less than a predetermined difference when compared with the features of the test record to determine a similarity score that the test record matches the reference record, and wherein the similarity score S for a test record T of M features and a reference record R of N features is determined by