Fingerprint recognition method and apparatus

Patent metadata

Field	Value
Publication number	US-9904840-B2
Application number	US-201615222107-A
Country	US
Kind code	B2
Filing date	Jul 28, 2016
Priority date	Oct 28, 2015
Publication date	Feb 27, 2018
Grant date	Feb 27, 2018

Abstract

A method for fingerprint recognition utilizes an auto encode decode network to perform feature extraction on a first fingerprint image acquired by a fingerprint sensor and a second fingerprint image retrieved from a database. First and second fingerprint features corresponding to the first and second fingerprint images are thus obtained. The first and second fingerprint features may have equal dimensionality. Dimensionality reduction may be performed on the first and second fingerprint features to respectively obtain third and fourth fingerprint features. The third and fourth fingerprint features may have equal dimensionality. A cosine distance between the third and fourth finger print features may determine whether the first fingerprint image and the second fingerprint image belong to a same fingerprint.

First claim

We claim:

1. A fingerprint recognition method, the method comprising:
retrieving, by a fingerprint recognition device comprising a memory and a processor in communication with the memory, a first fingerprint image acquired by a fingerprint sensor and a second fingerprint image stored in a database;
performing, by the fingerprint recognition device, feature extraction on the first fingerprint image and the second fingerprint image with an auto encode decode network to obtain a first fingerprint feature corresponding to the first fingerprint image and a second fingerprint feature corresponding to the second fingerprint image, the first fingerprint feature and the second fingerprint feature having equal dimensionality;
performing, by the fingerprint recognition device, dimensionality reduction on the first fingerprint feature and the second fingerprint feature to respectively obtain a third fingerprint feature and a fourth fingerprint feature, the third fingerprint feature and the fourth fingerprint feature having equal dimensionality that is smaller than the dimensionality of the first fingerprint feature and the second fingerprint feature;
calculating, by the fingerprint recognition device, a cosine distance between the third fingerprint feature and the fourth fingerprint feature;
comparing, by the fingerprint recognition device, the calculated cosine distance with a preset threshold;
when the cosine distance is greater than the preset threshold, determining, by the fingerprint recognition device, that the first fingerprint image and the second fingerprint image belong to the same fingerprint; and
when the cosine distance is less than or equal to the preset threshold, determining, by the fingerprint recognition device, that the first fingerprint image and the second fingerprint image belong to different fingerprints.

2. The method of claim 1 , wherein the auto encode decode network comprises at least one encoding layer, and the method further comprises:
training the auto encode decode network by:
training an encoding feature parameter for each encoding layer in the at least one encoding layer with an unlabeled fingerprint sample to obtain an encoding feature representation parameter corresponding to each encoding layer;
performing a data reconstruction on the encoding feature representation parameter corresponding to each encoding layer to obtain fingerprint reconstruction data of the unlabeled fingerprint sample using a decoding layer corresponding to each respective encoding layer;
determining a reconstruction error between the fingerprint reconstruction data and the unlabeled fingerprint sample; and
adjusting the encoding feature representation parameter corresponding to each encoding layer according to the reconstruction error; and
stopping training of the auto encode decode network when the reconstruction error reaches a minimum value for each encoding layer, to obtain a trained auto encode decode network.

3. The method of claim 2 , wherein a last encoding layer of the trained auto encode decode network is connected with a classifier, and the method further comprises:
inputting a labeled fingerprint sample into the trained auto encode decode network to obtain a first output result;
inputting the first output result to the classifier and training the classifier with the labeled fingerprint sample; and
stopping training the classifier when a reconstruction error between an output result of the classifier and the labeled fingerprint sample reaches a minimum value.

4. The method of claim 2 , wherein a last encoding layer of the trained auto encode decode network is connected with a classifier, and the method further comprises:
inputting a labeled fingerprint sample to the trained auto encode decode network to obtain a second output result;
inputting the second output result to the classifier, training the classifier with the labeled fingerprint sample and fine-tuning the encoding feature representation parameter corresponding to the encoding layer of the trained auto encode decode network; and
stopping training of the classifier and the fine-tuning of the encoding feature representation parameter corresponding to the encoding layer when a reconstruction error between an output result of the classifier and the labeled fingerprint sample reaches a minimum value.

5. The method of claim 2 , further comprising:
extracting an encoding feature representation parameter having a first setting dimensionality of the unlabeled fingerprint sample with the trained auto encode decode network; and
performing linear discriminant analysis LDA training on the encoding feature representation parameter having the first setting dimensionality to obtain a projection matrix having a second setting dimensionality of the LDA.

6. A fingerprint recognition apparatus, comprising:
a processor; and
a non-transitory computer-readable medium configured to store instructions executable by the processor, wherein, the processor, when executing the instructions, is configured to cause the fingerprint recognition apparatus to:
retrieve a first fingerprint image acquired by a fingerprint sensor and a second fingerprint image stored in a database;
perform feature extraction on a first fingerprint image and the second fingerprint image with an auto encode decode network to obtain a first fingerprint feature corresponding to the first fingerprint image and a second fingerprint feature corresponding to the second fingerprint image, the first fingerprint feature and the second fingerprint feature having equal dimensionality;
perform dimensionality reduction on the first fingerprint feature and the second fingerprint feature to respectively obtain a third fingerprint feature and a fourth fingerprint feature, the third fingerprint feature and the fourth fingerprint feature having equal dimensionality that is smaller than the dimensionality of the first fingerprint feature and the second fingerprint feature;
calculate a cosine distance between the third fingerprint feature and the fourth fingerprint feature;
compare the calculated cosine distance with a preset threshold;
when the cosine distance is greater than the preset threshold, determine that the first fingerprint image and the second fingerprint image belong to the same fingerprint; and
when the cosine distance is less than or equal to the preset threshold, determine that the first fingerprint image and the second fingerprint image belong to different fingerprints.

7. The apparatus of claim 6 , wherein the auto encode decode network comprises at least one encoding layer, and the processor when executing the instructions is further configured to:
train the auto encode decode network, and the processor is further configured to:
train an encoding feature parameter for each encoding layer in the at least one encoding layer with an unlabeled fingerprint sample to obtain an encoding feature representation parameter corresponding to each encoding layer;
perform a data reconstruction on the encoding feature representation parameter corresponding to each encoding layer to obtain fingerprint reconstruction data of the unlabeled fingerprint sample through a decoding layer corresponding to each respective encoding layer;
determine a reconstruction error between the fingerprint reconstruction data and the unlabeled fingerprint sample; and
adjust the encoding feature representation parameter corresponding to the encoding layer according to the reconstruction error; and
stop training the auto encode decode network, when the reconstruction error reaches a minimum value, to obtain a trained auto encode decode network.

8. The apparatus of claim 7 , wherein a last encoding layer of the trained auto encode decode network i

Assignees

• Xiaomi Inc

Inventors

• Zhang Tao
• Wang Pingze
• Zhang Shengkai

Classifications

• G06V30/19173

  Classification techniques · CPC title

• G06V40/1347 Primary

  Preprocessing; Feature extraction · CPC title

• G06V40/1365 Primary

  Matching; Classification · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06F18/24137

  Distances to cluster centroïds · CPC title