TWI831059B - Fingerprint identification method, fingerprint identification device and information processing device - Google Patents

Abstract

The present invention mainly discloses a fingerprint identification method, which includes the following steps: performing an image matching operation on a fingerprint image and a reference fingerprint image; when the fingerprint image and the reference fingerprint image match each other, using a fingerprint The image classification model performs an image classification operation on the fingerprint image; and when the image classification operation confirms that the fingerprint image is a real fingerprint image, it is determined that the fingerprint recognition is successful. Simply put, in addition to the traditional fingerprint recognition process, the fingerprint recognition method of the present invention further performs a real (living body) fingerprint recognition process, thereby realizing fingerprint recognition on the basis of living body recognition, effectively improving the security of fingerprint recognition. .

Description

Fingerprint identification method, fingerprint identification device and information processing device

The present invention relates to the technical field of fingerprint identification, and in particular, to a highly secure fingerprint identification method.

Biometric identification technology collects the inherent physiological characteristics of the human body as the basis for individual biological identification, such as: iris characteristics, facial characteristics, voiceprint characteristics, and fingerprint characteristics. Among various biometric identification technologies, fingerprint identification technology is the most mature and is widely used in various mobile devices in the form of a fingerprint identification circuit. When performing fingerprint recognition, the fingerprint recognition circuit completes fingerprint recognition and user identity recognition by matching features of a fingerprint image collected in real time with a pre-recorded fingerprint template.

Practical experience points out that in the process of applying fingerprint recognition technology to achieve user identity recognition, a third party can successfully deceive the fingerprint recognition circuit by holding a copy of the user's fingerprint, thereby identifying the user through fingerprint recognition. Currently, there are many known methods for obtaining or manufacturing fingerprint replicas. The first method is to obtain the fingerprint of the target finger by pressing it against a clay called Plastiline. The second method is to press the target's finger against a fingerprint reader used at airports, banks and border crossings, and then use a color printer to print out the fingerprint image. The third method is to leave a fingerprint of the target finger on a transparent object (such as a glass plate or glass), and then take a photo of the fingerprint left on the transparent object to obtain a fingerprint image of the target finger.

Furthermore, with the advancement of 3D printing technology, 3D printers can be used to create fake fingers with fingerprints, thereby using the fake fingers to successfully deceive the fingerprint recognition circuit and identify the user's identity through fingerprint recognition. It must be noted that with the advancement of financial technology, people can complete various financial-related behaviors without going to the counter, such as: online payment, online transfer, ATM withdrawal, online account opening, online stock investment, using the Internet Banks manage personal financial accounts, etc. Among them, the vast majority of financial technologies combine fingerprint, facial or iris recognition. In other words, in order to prevent a third party from being able to log into the user's financial account with a copy of the user's fingerprint to conduct illegal online financial activities, it is necessary to improve the shortcomings of the conventional fingerprint recognition circuit so that it can pass the liveness detection Or identification to detect non-living fingerprint copies, so that non-living fingerprint copies cannot be used for fingerprint recognition and user identity identification.

It can be seen from the above description that a highly secure fingerprint identification method is urgently needed in this field.

The main purpose of the present invention is to provide a fingerprint identification method, which is applied in a fingerprint identification device, so that the fingerprint identification device sequentially performs a fingerprint identification process on a fingerprint image and a living (real) fingerprint identification process. This effectively improves the security of fingerprint recognition.

To achieve the above object, the present invention proposes an embodiment of the fingerprint identification method, which includes:

Perform an image matching operation on a fingerprint image and a reference fingerprint image;

When it is determined that the fingerprint image and the reference fingerprint image match each other, performing an image classification operation on the fingerprint image using a fingerprint image classification model; and

When the image classification operation confirms that the fingerprint image is a real fingerprint image, it is determined that the fingerprint identification is successful.

In one embodiment, the fingerprint image classification model includes:

A feature extraction unit used to extract a plurality of sub-image features from the fingerprint image, thereby determining an image feature representing the fingerprint image based on the plurality of sub-image features; and

A classifier is used to perform a classification operation on the image feature to classify the image feature into the real fingerprint image or a fake fingerprint image.

In one embodiment, the sub-image features include features selected from the group consisting of ridge features, valley features, gray level co-occurrence matrix features, direction field features, global histogram features, and local histogram features. at least one of them.

In one embodiment, the fingerprint image classification model is generated using the following steps:

Input a plurality of sample images into a machine learning model to obtain a predicted category information for each sample image; wherein the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image, and Each of the sample images carries a category label;

Adjust the model parameters of the machine learning model according to the difference between the predicted category information and the category label of the sample image, thereby obtaining an adjusted machine learning model;

Use the adjusted machine learning model to classify multiple test images to obtain a classification accuracy;

Repeat all the aforementioned steps if the classification accuracy does not reach an accuracy threshold; and

When the classification accuracy reaches the accuracy threshold, the adjusted machine learning model is used as the fingerprint image classification model.

Furthermore, the present invention also provides a fingerprint identification device, which includes:

a sensor array; and

A fingerprint identification circuit coupled to the sensor array is used to drive the sensor array to perform an image acquisition operation on a finger to obtain a fingerprint image, and includes:

An identification unit used to perform an image matching operation on the fingerprint image and a reference fingerprint image;

A classification unit used to perform an image classification operation on the fingerprint image using a fingerprint image classification model; and

A determination unit used to determine whether the fingerprint image is a real fingerprint image according to the result of the image classification operation, and then determine that the fingerprint recognition is successful.

In one embodiment, the fingerprint image classification model includes:

A feature extraction unit used to extract a plurality of sub-image features from the fingerprint image, thereby determining an image feature representing the fingerprint image based on the plurality of sub-image features; and

A classifier is used to perform a classification operation on the image feature to classify the image feature into the real fingerprint image or a fake fingerprint image.

In one embodiment, the sub-image features include features selected from the group consisting of ridge features, valley features, gray level co-occurrence matrix features, direction field features, global histogram features, and local histogram features. at least one of them.

In one embodiment, the fingerprint image classification model is generated using the following steps:

Input a plurality of sample images into a machine learning model to obtain a predicted category information for each sample image; wherein the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image, and Each of the sample images carries a category label;

Adjust the model parameters of the machine learning model according to the difference between the predicted category information and the category label of the sample image, thereby obtaining an adjusted machine learning model;

Use the adjusted machine learning model to classify multiple test images to obtain a classification accuracy;

Repeat all the aforementioned steps if the classification accuracy does not reach an accuracy threshold; and

When the classification accuracy reaches the accuracy threshold, the adjusted machine learning model is used as the fingerprint image classification model.

Furthermore, the present invention also provides an information processing device, which is characterized by having the fingerprint identification device of the present invention as described above. In a feasible embodiment, the information processing device is selected from a tablet computer, a notebook computer, an all-in-one computer, a smart phone, a smart watch, an electronic financial transaction device, a clock card device, an access control device, and an electronic door. An electronic device in a group of locks.

In order to enable the review committee to further understand the structure, characteristics, purpose, and advantages of the present invention, drawings and detailed descriptions of preferred embodiments are attached below.

Figure 1 shows a block diagram of a fingerprint identification device according to the present invention. As shown in FIG. 1 , the fingerprint identification device 1 includes a sensor array 11 and a fingerprint identification circuit 12 coupled to the sensor array 11 , and the fingerprint identification circuit 12 has an identification unit 121 and a classification unit 122 and a determining unit 123. According to the design of the present invention, after driving the sensor array 11 to perform an image acquisition operation on a finger to obtain a fingerprint image, the fingerprint identification circuit 12 uses the identification unit 121 to compare the fingerprint image with a reference fingerprint. An image matching operation is performed on the image to confirm whether the fingerprint image matches the reference fingerprint image. Moreover, when it is determined that the fingerprint image and the reference fingerprint image match each other, the classification unit 122 uses a fingerprint image classification model to perform an image classification operation on the fingerprint image to confirm the fingerprint image. Be a real fingerprint image or a fake fingerprint image. Further, when it is confirmed that the fingerprint image is a real fingerprint image, the determining unit 123 determines that the fingerprint recognition is successful.

Figure 2 shows a flow chart of a fingerprint identification method of the present invention. As shown in Figure 2, the fingerprint identification method of the present invention first performs step S1: performing an image matching operation on a fingerprint image and a reference fingerprint image. Next, the method flow executes step S2: when it is determined that the fingerprint image and the reference fingerprint image match each other, use a fingerprint image classification model to perform an image classification operation on the fingerprint image. Finally, the method flow executes step S3: when the image classification operation confirms that the fingerprint image is a real fingerprint image, it is determined that the fingerprint identification is successful.

Figure 3 is a block diagram of the classification unit shown in Figure 1. It should be understood that in order to enable the fingerprint identification device 1 as shown in FIG. 1 to sequentially perform a fingerprint identification process on a fingerprint image and a living (real) fingerprint identification process, the present invention specifically makes the fingerprint identification circuit 12 have A recognition unit 121, a classification unit 122 and a determination unit 123, and the classification unit 122 includes a fingerprint image classification model 1221. In an embodiment, the fingerprint image classification model includes: a feature extraction unit and a classifier.

FIG. 4 is a flow chart of the image classification operation performed by the fingerprint image classification model shown in FIG. 3 . As shown in Figure 4, when performing the image classification operation, first, the feature extraction unit of the fingerprint image classification model 1221 extracts a plurality of sub-image features from the fingerprint image, thereby based on the plurality of sub-images The feature is determined to represent an image feature of the fingerprint image (ie, step S21 is performed). In a feasible embodiment, the sub-image features may include one or a combination of one or more of ridge features, valley features, gray level co-occurrence matrix features, direction field features, global histogram features, and local histogram features. Then, the classifier performs a classification operation on the image feature to classify the image feature into the real fingerprint image or a fake fingerprint image (ie, step S22 is performed).

The present invention does not limit the algorithm of the classifier. In a feasible embodiment, the algorithm used to construct the classifier may be a Support Vector Machine (SVM), softmax, Convolutional Neural Networks (CNN), or Bayesian classification. French Naive Bayes Classification).

In addition, common algorithms that can be used to build the classifier also include: fully convolutional neural network (FCN), region-based convolutional neural network (Region-based convolutional neural network, R-CNN), Fast region-based convolutional neural network (fast R-CNN), or Faster region-based convolutional neural network (Faster region-based convolutional neural network) , faster R-CNN), region-based convolutional network algorithm using mask (Mask R-CNN), real-time object detection algorithm YOLOv1, object detection algorithm YOLOv2, object detection algorithm YOLOv3, real-time object detection framework algorithm (Single-shot multiBox detector, SSD) and many other types, the present invention is not particularly limited.

It should be understood that the fingerprint image classification model uses a plurality of sample images (training samples) to perform classification training on a machine learning model and uses a plurality of test images (test samples) to classify the trained machine learning model. Completed after testing. Figure 5 is a flow chart for generating the fingerprint image classification model shown in Figure 3. As shown in Figure 5, the fingerprint image classification model 1221 is generated using the following steps:

Step S1a: Input a plurality of sample images into a machine learning model to obtain a predicted category information for each sample image; wherein the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image. image, and each sample image carries a category label;

Step S2a: Adjust the model parameters of the machine learning model according to the difference between the predicted category information and the category label of the sample image, thereby obtaining an adjusted machine learning model;

Step S3a: Use the adjusted machine learning model to classify a plurality of test images to obtain a classification accuracy;

Step S4a: Repeat all the aforementioned steps if the classification accuracy does not reach an accuracy threshold; and

Step S5a: When the classification accuracy reaches the accuracy threshold, use the adjusted machine learning model as the fingerprint image classification model.

Electronic engineers who are familiar with the design and production of image recognition software should know that the accuracy threshold is an empirical value, such as 93%. In addition, it should be supplemented that the plurality of sample images are usually organized into a training sample set, and include the plurality of sample images including a plurality of forged fingerprint images and the real fingerprint image, each of the samples Images all carry a category label. On the other hand, the plurality of test images are organized into a test sample set. Moreover, in step S2a, a host computer used to perform fingerprint image classification training on the machine learning model evaluates the difference between the predicted category information and the category label based on the loss function, thereby adjusting the result according to the difference. Describe the model parameters of the machine learning model. Finally, after multiple iterative trainings (ie, steps S1a to S3a are repeated multiple times), the machine learning model evaluated by the test set is used as the fingerprint image classification model 1221 included in the classification unit 122 .

In this way, the above has completely and clearly explained a fingerprint identification method of the present invention; and from the above, it can be seen that the present invention has the following advantages:

(1) The present invention discloses a fingerprint identification method, which is applied in a fingerprint identification device so that the fingerprint identification device sequentially performs a fingerprint identification process on a fingerprint image and a living (real) fingerprint identification process, thereby Effectively improve the security of fingerprint recognition.

(2) Moreover, the present invention also proposes a fingerprint identification device, which includes a sensor array and a fingerprint identification circuit; it is characterized in that the fingerprint identification circuit has an identification unit, a classification unit and a determination unit, so as to be able to A fingerprint identification method of the present invention is performed as described above.

(3) Further, the present invention also proposes an information processing device, which has the fingerprint identification device of the present invention as described above, and the information processing device is selected from the group consisting of tablet computers, notebook computers, all-in-one computers, and smart phones. It is an electronic device among the group consisting of mobile phones, smart watches, financial transaction electronic devices, clock-in devices, access control devices, and electronic door locks.

It must be emphasized that the foregoing disclosed in this case are preferred embodiments. Any partial changes or modifications derived from the technical ideas of this case and easily inferred by those familiar with the art do not deviate from the patent of this case. category of rights.

To sum up, regardless of the purpose, means and effects of this case, it shows that it is completely different from the conventional technology, and that the invention is practical first, and indeed meets the patent requirements for inventions. I sincerely ask the review committee to take a clear look and grant the patent as soon as possible for your benefit. Society is a prayer for the Supreme Being.

1:Fingerprint recognition device 11: Sensor array 12:Fingerprint recognition circuit 121:Identification unit 122: Taxon 1221:Fingerprint image classification model 123: Determine unit S1: Perform an image matching operation on a fingerprint image and a reference fingerprint image S2: When it is determined that the fingerprint image and the reference fingerprint image match each other, use a fingerprint image classification model to perform an image classification operation on the fingerprint image. S3: When the image classification operation confirms that the fingerprint image is a real fingerprint image, determine that the fingerprint identification is successful. S21: The feature extraction unit extracts a plurality of sub-image features from the fingerprint image, thereby determining an image feature representing the fingerprint image based on the plurality of sub-image features. S22: The classifier performs a classification operation on the image feature to classify the image feature into the real fingerprint image or a fake fingerprint image. S1a: Input a plurality of sample images into a machine learning model to obtain a predicted category information for each sample image; wherein the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image , and each sample image carries a category label S2a: Adjust the model parameters of the machine learning model according to the difference between the predicted category information and the category label of the sample image, thereby obtaining an adjusted machine learning model S3a: Use the adjusted machine learning model to classify multiple test images and obtain a classification accuracy rate S4a: Repeat all the above steps if the classification accuracy does not reach an accuracy threshold. S5a: When the classification accuracy reaches the accuracy threshold, use the adjusted machine learning model as the fingerprint image classification model

Figure 1 is a block diagram of a fingerprint identification device according to the present invention; Figure 2 is a flow chart of a fingerprint identification method according to the present invention; Figure 3 is a block diagram of a taxon shown in Figure 1; Figure 4 is a flow chart of an image classification operation performed by a fingerprint image classification model shown in Figure 3; and Figure 5 is a flow chart for generating the fingerprint image classification model shown in Figure 3.

S1: Perform an image matching operation on a fingerprint image and a reference fingerprint image

S2: When it is determined that the fingerprint image and the reference fingerprint image match each other, use a fingerprint image classification model to perform an image classification operation on the fingerprint image.

S3: When the image classification operation confirms that the fingerprint image is a real fingerprint image, determine that the fingerprint identification is successful.

Claims (8)

A fingerprint identification method is applied to a fingerprint identification device including a sensor array and a fingerprint identification circuit, and is executed by the fingerprint identification circuit; the fingerprint identification method includes: comparing a fingerprint image with a Perform an image matching operation with reference to the fingerprint image; when it is determined that the fingerprint image and the reference fingerprint image match each other, use a fingerprint image classification model to perform an image classification operation on the fingerprint image, wherein , the fingerprint image classification model includes a feature extraction unit and a classifier, and the image classification operation includes the following steps: enabling the feature extraction unit to extract a plurality of sub-image features from the fingerprint image, thereby according to the complex number The sub-image feature determines an image feature representing the fingerprint image; and the classifier is enabled to perform a classification operation on the image feature to classify the image feature as a real fingerprint image or a fake fingerprint image. image; and in the case where the image classification operation confirms that the fingerprint image is a real fingerprint image, it is determined that the fingerprint identification is successful. The fingerprint identification method according to claim 1, wherein the sub-image features include ridge features, valley features, gray level co-occurrence matrix features, direction field features, global histogram features, and local histograms. At least one of the groups composed of graph features. The fingerprint identification method according to claim 1, wherein the fingerprint image classification model is generated by using the following steps: inputting a plurality of sample images into a machine learning model to obtain a predicted category information for each of the sample images. ; Wherein, the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image, and each of the sample images carries a category label; according to the predicted category information of the sample image and Based on the difference between the category labels, adjust the model parameters of the machine learning model to obtain an adjusted machine learning model; use the adjusted machine learning model to classify a plurality of test images to obtain a classification accuracy; When the classification accuracy does not reach an accuracy threshold, all the aforementioned steps are repeatedly performed; and when the classification accuracy reaches the accuracy threshold, the adjusted machine learning model is used as the Fingerprint image classification model. A fingerprint identification device, which includes: a sensor array; and a fingerprint identification circuit coupled to the sensor array to drive the sensor array to perform an image acquisition operation on a finger to obtain a fingerprint image Image; It is characterized in that the fingerprint identification circuit further performs a fingerprint identification method on the fingerprint image, and the fingerprint identification method includes the following steps: performing an image matching operation on the fingerprint image and a reference fingerprint image; When it is determined that the fingerprint image and the reference fingerprint image match each other, an image classification operation is performed on the fingerprint image using a fingerprint image classification model, wherein the fingerprint image classification model includes a feature extraction unit and a classifier, and the image classification operation includes the following steps: enabling the feature extraction unit to extract a plurality of sub-image features from the fingerprint image, thereby determining a feature representing the fingerprint image based on the plurality of sub-image features. an image feature; and enabling the classifier to perform a classification operation on the image feature to classify the image feature as a real fingerprint image or a fake fingerprint image; and confirming in the image classification operation If the fingerprint image is a real fingerprint image, it is determined that the fingerprint identification is successful. The fingerprint identification device according to claim 4, wherein the sub-image features include ridge features, valley features, gray level co-occurrence matrix features, direction field features, global histogram features, and local histograms. At least one of the groups composed of graph features. The fingerprint identification device according to claim 4, wherein the fingerprint image classification model is generated by using the following steps: inputting a plurality of sample images into a machine learning model to obtain a predicted category information for each of the sample images. ; Wherein, the plurality of sample images include a plurality of forged fingerprint images and the real fingerprint image, and each of the sample images carries a category label; According to the difference between the predicted category information and the category label of the sample image, the model parameters of the machine learning model are adjusted, thereby obtaining an adjusted machine learning model; using the adjusted machine learning model to analyze a plurality of Classify the test image to obtain a classification accuracy; repeat all the aforementioned steps when the classification accuracy does not reach an accuracy threshold; and when the classification accuracy reaches the accuracy threshold , using this adjusted machine learning model as the fingerprint image classification model. An information processing device, characterized by having a fingerprint identification device as described in any one of claims 4 to 6. The information processing device as described in claim 7, wherein the information processing device is selected from the group consisting of tablet computers, notebook computers, all-in-one computers, smart phones, smart watches, financial transaction electronic devices, clock-in devices, and access control devices. An electronic device in the group consisting of a device and an electronic door lock.

Images