Apparatus and computer-implemented method for fingerprint based authentication

Patent metadata

Field	Value
Publication number	US-9672409-B2
Application number	US-201615189055-A
Country	US
Kind code	B2
Filing date	Jun 22, 2016
Priority date	Jul 3, 2015
Publication date	Jun 6, 2017
Grant date	Jun 6, 2017

Abstract

A computer-implemented method of performing fingerprint based authentication from matching local features represented by binary features which can be matched in an efficient implementation in one or both of software and hardware by computing Hamming distances between the binary features. A local feature in a verification image is said to be matching with a local feature in an enrolment image if the Hamming distance between the binary features falls below a pre-determined threshold. The computer-implemented method retains information about the similarity of local features in the two images and utilities it in an efficient way with the objective of improving fingerprint recognition rates and enabling finger liveness detection. In an aspect a normalized feature similarity distribution is generated as part of the representation in recognition and liveness detection.

First claim

The invention claimed is:

1. A computer-implemented method of processing a fingerprint image, comprising:
acquiring a present fingerprint image from a fingerprint sensor and computing a plurality of first feature vectors of the present fingerprint image;
retrieving a plurality of second feature vectors of an enrolled fingerprint image from a storage;
applying a similarity measure to compute a plurality of similarity values that represents a degree of similarity between a set of first feature vectors and second feature vectors;
performing first classification of at least a portion of the plurality of similarity values to generate a first signal indicative of whether the present fingerprint image falls within a class defined from enrolled images;
performing second classification of at least a portion of the plurality of similarity values to generate a second signal indicative of whether the present fingerprint image falls within a class of fingerprints from live fingers or a class of fingerprints from imitated fingers;
generating an authentication signal from the first signal and the second signal;
characterized in:
matching first feature vectors and second feature vectors to form a first set of matching pairs;
reducing the first set of matching pairs to a second set of matching pairs that comprises those matching pairs that agree to a uniform geometrical transformation;

wherein computation of the plurality of values of the similarity measure is restrained to be computed from the pairs of first feature vectors and second feature vectors that falls into the second set of matching pairs;
wherein the first feature vectors and the second feature vectors that fall outside one or both of the first set of matching pairs and the second set of matching pairs are collected in a first set of outlier features; and
wherein one or both of the first classification and the second classification includes all or at least a portion of the outlier features in generating the first signal and the second signal, respectively.

2. A computer-implemented method according to claim 1 , wherein the plurality of values of the similarity measure is computed by firstly generating a vector representation of the similarity between a first feature vector and a second feature vector; and secondly computing an aggregated similarity value across the vector representation of the similarity.

3. A computer-implemented method according to claim 1 , comprising:
organising the plurality of the similarity values in a histogram representation;
wherein one or both of the first classification and second classification is performed to classify the plurality of similarity values from the histogram representation.

4. A computer-implemented method according to claim 1 , wherein the first feature vectors are computed from image data in regions about respective positions in the present fingerprint image; and wherein the second feature vectors are computed from image data in regions about respective positions in a previously enrolled fingerprint image.

5. A computer-implemented method according to claim 1 , wherein one or both of the first classification and second classification is performed to classify those similarity values which represent coincidence of the present fingerprint image with a group of enrolled images from a compound feature vector that comprises similarity values arranged in a histogram representation and additionally one or more features selected from the group of:
a count of matching pairs in the second set of matching pairs;
a sum of matching pairs in the second set of matching pairs,
a sum of similarity values, such as a sum of Hamming distances;
a sum of similarity values in the second set of matching pairs;
the mean of similarity values, such as the mean of Hamming distances;
the mean of similarity values in the second set of matching pairs, such as the mean of Hamming distances;
the standard deviation of similarity values, such as the standard deviation of Hamming distances;
the standard deviation of similarity values in the second set of matching pairs, such as the standard deviation of Hamming distances;
a ratio of the number of pairs in the second set of matching pairs to the number of pairs in the first set of matching pairs.

6. A computer-implemented method according to claim 1 , wherein one or both of the class of fingerprints from live fingers and the class of fingerprints from imitated fingers is/are defined from a collection of fingerprints from live fingers and a collection of fingerprints from imitated fingers.

7. A computer-implemented method according to claim 1 , wherein the second classification classifies values of the similarity measure from supervised training; comprising the steps of:
a) acquiring multiple sets of fingerprint images from multiple live individuals, and generating pairs of images thereof that matches; wherein for each pair of fingerprint images, first feature vectors are computed from a first fingerprint image and second feature vectors are computed from a second fingerprint image;
b) for each pair of fingerprint images, computing a plurality of values of the similarity measure that measures the similarity between the images in the pair of fingerprint images;
wherein step a) is performed on fingerprint images acquired from live fingers, and wherein, in step b), the plurality of values of the similarity measure are collectively labelled by a first label; and
wherein step a) is performed on fingerprint images acquired from imitated fingers, and wherein, in step b), the plurality of values of the similarity measure are collectively labelled by a second label; and
wherein a classifier is trained from one or both of a first training set comprising the plurality of values of the similarity measure obtained from live fingers with the first label as a supervisory signal and a second training set comprising the plurality of values of the similarity measure obtained from imitated fingers with the second label as a supervisory signal.

8. A computer-implemented method according to claim 7 , wherein the plurality of values of the similarity measure in the training sets are organised in histogram representations; and wherein one or both of the first classification and second classification is performed to classify the plurality of similarity values from the histogram representation.

9. A computer-implemented method of configuring a first classifier by supervised training to distinguish live fingerprint images from imitated fingerprint images; comprising the steps of:
a) acquiring multiple sets of fingerprint images, and generating pairs of images thereof that matches; wherein for each pair of fingerprint images, first feature vectors are computed from a first fingerprint image and second feature vectors are computed from a second fingerprint image;
b) for each pair of fingerprint images, computing a plurality of values of the similarity measure that measures the similarity between the images in the pair of fingerprint images;
wherein step a) is performed on fingerprint images acquired from live fingers, and wherein, in step b), the plurality of values of the similarity measure are collectively labelled by a first label; and
wherein step a) is performed on fingerprint images acquired from imitated fingers, and wherein, in step b), the plurality of values of the similarity measure are collectively labelled by a second label; and
wherein the first classifier is trained from one or both of a first training set comprising the plurality of values of the similarity measure obtained from live fingers with the first label as a supervisory signal and a second training set comprising the plurality of values of the similarity

Assignees

• Fingerprint Cards Ab

Inventors

• Jonsson Kenneth
• Setterberg Eric
• Tingdahl David

Classifications

• G06V40/1376 Primary

  Matching features related to ridge properties or fingerprint texture · CPC title

• G06V40/1388

  using image processing · CPC title

• G06K9/001

  Physics · mapped topic

• G06K9/00013

  Physics · mapped topic

• G06K9/0008

  Physics · mapped topic