TW202014917A - Authentication method and electronic device using the same - Google Patents

Abstract

An authentication method is provided. The authentication method includes: sending a motion prompt message; activating an image capturing element of an electronic device for acquiring a first video stream; calculating first optical flow information according to two successive images within the first video stream; inputting the first optical flow information into a neural network model in order to obtain an authentication result, where the neural network model includes a motion tag corresponding to the motion prompt message, at least one second optical flow information corresponding to the motion tag and identification registration information corresponding to each second optical flow information. In addition, an electronic device using the authentication method is also provided.

Description

Identity verification method and electronic device using the method

The invention relates to an identity verification method and an electronic device using the method.

In the past, identity verification on electronic devices used to input personal passwords through input devices such as keyboards, touch screens, or mice. The process was quite cumbersome and often resulted in user inconveniences due to forgotten passwords.

In recent years, many biometric-based identity verification methods have been gradually developed, including fingerprint recognition, voiceprint recognition, iris recognition, and even facial recognition, each with its own advantages and disadvantages. For example, facial recognition is to extract facial features from static photos for comparison, but the accuracy of the current facial recognition algorithm needs to be improved, and it is not suitable for users of all ethnic groups (For example, the accuracy rate will vary depending on the user's skin color).

Therefore, how to provide a fast and accurate identity verification method is the goal of those skilled in the art.

In view of this, the present invention provides an identity verification method and an electronic device using the identity verification method, which is suitable for users of various ethnic groups and is fast and accurate.

The identity verification method of the embodiment of the present invention is applicable to electronic devices, and includes the following steps: sending out an action prompt message; activating the image capturing element of the electronic device to obtain the first video stream; according to the two consecutive video stream Image, calculate and obtain the first optical flow information; input the first optical flow information to the neural network-like model to obtain the identity verification result, wherein the neural network-like model includes an action label corresponding to the action prompt message, and at least one first corresponding to the action label Two optical flow information, and identity registration information corresponding to each second optical flow information.

In an embodiment of the present invention, the above-mentioned identity verification method further includes the following steps before the step of issuing the action prompt message: collecting at least one second video stream of at least one user according to different multiple action tags; Two consecutive images in a second video stream, calculating and obtaining at least one second optical flow information of at least one second video stream; and using at least one second optical flow information to train a neural network model.

In an embodiment of the invention, the above action prompt message corresponds to one of the action tags.

In an embodiment of the invention, the step of training the neural network model using at least one second optical flow information further includes the following steps: defining identity registration information to the corresponding second optical flow information.

In an embodiment of the present invention, the aforementioned processor triggers an execution event of the electronic device according to the identity verification result.

The electronic device of the embodiment of the present invention includes a prompt component, an image capturing component and a processor. The prompting component is used for sending out motion prompting messages. After the action prompt message is sent, the image capturing element is activated to obtain the first video stream. The processor signal is connected to the prompt component and the image capturing component, and the processor is used to run a neural network model, wherein the neural network model includes an action label corresponding to the action prompt message and at least one second optical flow information corresponding to the action label , And the identity registration information corresponding to each second optical flow information, the processor is used to perform the following steps: calculate and obtain the first optical flow information according to the two consecutive images in the first video stream; input the first optical flow information to the nerve-like Network model to obtain identity verification results.

Based on the above, the identity verification method and the electronic device using the method proposed in the embodiments of the present invention utilize the characteristics that different individuals' actions corresponding to the same action command will be different, by capturing the video string of the user's action Flow, calculate the optical flow information of two adjacent images in the video stream, and use this optical flow information to input a trained neural network-like model as the basis for judgment during identity verification. According to this, it can be applied to various appearance groups and has a high verification speed.

In order to make the above-mentioned features and advantages of the present invention more obvious and understandable, the embodiments are specifically described below in conjunction with the accompanying drawings for detailed description as follows.

The actions of different individuals corresponding to the same action instructions (for example, nodding, shaking or turning head) will also be different. If the action is taken, these differences will be reflected in the optical flow corresponding to the video stream (optical flow) ) In the news. Therefore, the identity verification method of the embodiment of the present invention uses the optical flow information corresponding to the video stream as the basis for identity verification, and then uses the trained classifier through machine learning to accurately obtain the results of the identity verification.

FIG. 1 is a schematic diagram of an electronic device according to an embodiment of the invention.

1, the electronic device 100 includes a processor 110, an image capturing element 120, a prompt element 130, and a storage element 140, wherein the image capturing element 120, the prompt element 130, and the storage element 140 are all connected to the processor 110, and The processor 110 is used to run a type of neural network model 150. In some embodiments, the electronic device 100 is, for example, a personal computer, a server, a notebook computer, a tablet computer, a smartphone, etc., and the invention is not limited thereto.

The processor 110 is used to cooperate with other elements in the electronic device 100 to complete the identity verification method of the embodiment of the present invention. The processor 110 may be, for example, a dual-core, quad-core, or eight-core central processing unit (CPU), system-on-chip (SOC), application processor (application processor), or media Processor (media processor), microprocessor (microprocessor), digital signal processor (digital signal processor) programmable controller, application specific integrated circuits (application specific integrated circuits, ASIC), programmable logic device (programmable logic device, PLD) or other similar devices or a combination of these devices, the present invention does not limit the type of processor used in implementation. In some embodiments, the processor 110 is responsible for the overall operation of the electronic device 100, for example.

The image capturing element 120 is used to shoot video streams. The image capture device 120 may be, for example, built-in or external to the electronic device 100, and is equipped with a charge coupled device (Charge Coupled Device, CCD), a complementary metal-oxide semiconductor (CMOS) device, or other types of The imaging lens of the photosensitive element, but the present invention is not limited to this. In some embodiments, the image capturing element 110 is, for example, a camera embedded above the screen of the electronic device 100.

The prompting component 130 is used for sending out action prompting messages to prompt the user to perform corresponding actions. The prompting element 130 may be, for example, a display or a speaker, etc., and sends motion prompting messages with dynamic images, static images, text, or sound, but the present invention is not limited thereto.

The storage element 140 is used to record data required by the identity verification method, such as user identity registration data, weight values or parameters of the classifier, and the like. The storage element 140 can be, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory (Flash memory), Hard disks or other similar devices or combinations of these devices, but the invention is not limited thereto.

Since the electronic device 100 can be used to execute the identity verification method according to the embodiment of the present invention, the following will describe the identity verification method in detail with various components of the electronic device 100. However, it is worth mentioning that the present invention does not limit the device to which the identity verification method is applicable.

For the sake of clarity, the identity verification method is divided into two parts for description in the following, which are the training of the neural network-like model 150 and the actual identity verification steps. In some embodiments, the training of the neural network-like model 150 is, for example, used to adjust various parameters and weight values used by the processor 110 when running the neural network-like model 150, and the actual identity verification step is, for example, The trained neural network-like model 150 is used for identity verification.

FIG. 2 shows a flowchart of training a neural network model according to an embodiment of the invention.

Referring to FIG. 2, in step S210, the processor 110 collects at least one second video stream of at least one user according to different multiple action tags. In some embodiments, the action label includes, for example, nodding, shaking, turning, or drawing a circle on the face.

In some embodiments, step S210 is, for example, a user registration step. The processor 110 may prompt the user to make one of the action labels (such as nodding, shaking his head, turning his head, or drawing a circle on his face, etc.) through the prompt element 130, Then, the second video stream is photographed through the image capturing element 120 to collect the second video stream corresponding to the action performed by the user according to the above action prompt message. In some embodiments, the user's second video stream may also be pre-stored in the storage element 140 or in a cloud hard drive (not shown), and the processor 110 will step from the storage element 140 or cloud in step S210. The hard disk is used to obtain the second video stream of the user, and the invention is not limited thereto.

For example, the four users "A", "B", "C", and "D" are registered, and the processor 110 collects the four uses of "A", "B", "C", and "D" The second video stream corresponds to different action tags. Taking the user "A" as an example, the processor 110 may prompt the user "A" to perform a nod action several times with the action label as a nod, and correspondingly obtain the second video stream of the user "A" nod; The label for shaking his head prompts the user "A" to shake his head several times, and correspondingly obtains the second video stream of the user "A" shaking his head, and so on. In this way, the processor 110 respectively obtains multiple video streams of "A", "B", "C", and "D" nods, multiple video streams of shaking heads, and so on.

In some embodiments, the processor 110 stores, for example, all the collected second video streams in the storage element 140, and organizes the second video streams with folder names. For example, when the user "A" is registered, the processor 110 may, for example, record all the second video streams of the user "A" whose action is nodded in the "nod" subfolder in the "A" folder (For example, the data path is A -> nod); record all the second video streams of the registered user "A" action label as shaking his head in the "shaking head" subfolder of the "A" folder (for example, data The path is A -> shaking his head); when the user "B" is registered, the processor 110 can, for example, record all the second video streams of the registered user "B" action label as nod in the "B" folder "Nod" subfolder (for example, the data path is B -> nod), and so on.

In step S220, the processor 110 calculates and obtains the second optical flow information corresponding to each second video stream collected by the action tag according to the two consecutive images in each second video stream. In detail, for each second video stream it collects, the processor 110 first samples (for example, but not limited to, 30 images per second) to obtain multiple images in the second video stream. Subsequently, the processor 110 calculates the displacement vectors of the optical flow points of the two consecutive images in the images as the second optical flow information of the corresponding action label.

For example, when the processor 110 samples the second video stream to obtain 10 images, the processor 110 calculates multiple displacement vectors for multiple optical flow points between the second image and the first image. Multiple displacement vectors of multiple optical flow points between the three images and the second image, and so on, the processor 110 will obtain multiple displacement vectors of multiple optical flow points between the two pairs of images, and The multiple displacement vectors of the multiple optical flow points between the nine two pairs of images can be used as the second optical flow information of the second video stream.

Regarding the definition method of optical flow point and the calculation method of optical flow information, those with ordinary knowledge in the field should have sufficient ability to implement it according to relevant literature, so I will not repeat them here.

For example, after collecting the second video streams of the four users "A", "B", "C", and "D", the processor 110 calculates all the second video streams separately in the above manner, for example The second optical flow information, and each second optical flow information will correspond to one of the action labels of one of the users in "A", "B", "C", and "D".

In step S230, the processor 110 uses the obtained second optical flow information to train the neural network-like model 150. In order to train the neural network-like model 150, in step S231, the processor 110 first defines an identity registration information to the corresponding second optical flow information, and then uses each second optical flow information and its identity registration in step S233 Information to train a neural network-like model 150 (eg, optimize multiple weight values or parameters in the classifier). It must be noted that those with ordinary knowledge in the field can obtain and implement specific methods of training the neural network model 150 based on their knowledge of machine learning, so the relevant details will not be repeated here.

In some embodiments, the identity registration information assigned to each piece of second optical flow information includes, for example, the personal identity information of the user to which it belongs, such as name and the like. For example, the identity registration information corresponding to all the second optical flow information that the user "A" nods includes, for example, the name of "A", and the identity registration information corresponding to all the second optical flow information that the user "B" shakes his head includes, for example, The name of "B", and so on.

In some embodiments, the trained neural network-like model 150 is, for example, a long-short-term memory (LSTM) time recurrent neural network (RNN), which includes memory capability Neurons. According to this, the time-recurrent neural network of trained long- and short-term memory can provide more accurate classification of optical flow information. It must be noted that the invention does not limit the type of neural network model 150 here.

The neural network model 150 obtained after completing the above step S230 is equivalent to the weight value or parameter of the optimized neural network model 150, and the processor 110 will optimize these neural network models The weight value or parameter of the road model 150 is recorded in the storage element 140.

FIG. 3 illustrates a flowchart of identity verification according to an embodiment of the invention.

Referring to FIG. 3, in step S310, the prompting element 130 issues an action prompt message to prompt the user to perform an action, and the action prompt message corresponds to one of the action labels. After the action prompt message is sent, the image capturing element 120 is activated. In step S320, the image capturing element 120 shoots the user performing the action corresponding to the action prompt message to obtain a first video stream, and The first video stream is sent to the processor 110.

In some embodiments, the neural network-like model 150 uses multiple motion labels, such as nodding, shaking, turning, or drawing a circle on the face, for example. Therefore, the processor 110 will send an action prompt message through the prompt element 130 when verifying the identity to prompt the user to perform one of the action labels. After the processor 110 sends out an action prompt message, for example, the image capturing element 120 may be used to obtain the first video stream within a certain period of time (for example, the default is 10 seconds, but not limited to this) after the action prompt is issued. flow. For example, the image capturing element 120 can continue to shoot when the electronic device 100 is awakened or a specific application is opened, and the processor 110 obtains a video stream at a specific time from the time point when the action prompt message is issued. For subsequent verification.

In step S330, the processor 110 calculates and obtains the first optical flow information according to two consecutive images in the first video stream. In this step, the detailed method of calculating the first optical flow information based on the adjacent two consecutive images in the first video stream is similar to the calculation of the second optical flow information of one of the second video streams in the foregoing step S220 Way, so I won’t go into details here.

In step S340, the processor 110 runs the neural network-like model 150 according to the calculated first optical flow information to output the identity verification result. In some embodiments, the trained neural network-like model 150 can be used to classify the input first optical flow information to one of the personal identity information, or to other user information, to other User information means that this first optical flow information cannot be classified to any registered user. For example, the identity registration information used when training the neural network model 150 includes "A", "B", "C", and "D", so the neural network model 150 can calculate the input after training The first optical flow information belongs to the probability of each user of "A", "B", "C" and "D". If the input optical flow information belongs to the user "A" with the highest probability and is higher than the preset threshold, the neural network model 150, for example, will classify the input first optical flow information to the user "A" ; If the first optical flow information entered belongs to the user "B" with the highest probability and is higher than the preset threshold, the neural network model 150, for example, will classify the input optical flow information to the user "B" If the entered first optical flow information belongs to the user "C" with the highest probability and is higher than the preset threshold, the neural network model 150, for example, will classify the input first optical flow information to use "C"; if the entered first optical flow information belongs to the user "D" with the highest probability and is higher than the preset threshold, the neural network model 150 will, for example, classify the input first optical flow information Class to user "D"; if the probability that the first optical flow information entered belongs to "A", "B", "C" and "D" is not higher than the preset threshold, then the neural network model 150, for example, classifies the input first optical flow information as "other".

In step S350, the processor 110 triggers an execution event of the electronic device 100 according to the identity verification result output by the neural network-like model.

In some embodiments, the identity verification result shows, for example, whether the user passed the verification. For example, when the identity verification result classifies the first optical flow information as one of the identity registration information (for example, the name of one of the users such as "A", "B", "C" or "D"), or When the optical flow information is classified as "registered", the processor 110 outputs the identity verification result of the user passing the verification. On the other hand, when the identity verification result classifies the first optical flow information as having no registered information (for example, "other"), or classifies the first optical flow information as "unregistered", the processor 110 will Output the user's identity verification result that failed verification. Accordingly, the processor 110 may determine whether to trigger an execution event of the electronic device 100 according to the identity verification result, for example, the execution event is, for example, to allow a user to log in to the electronic device 100 or execute a specific application program.

In summary, the identity verification method and the electronic device using the method provided in the embodiments of the present invention take advantage of the characteristics that the actions performed by different individuals corresponding to the same action instruction will be different. Video streaming, calculate the optical flow information of video streaming, and use this optical flow information as the basis for identity verification. According to this, it can be applied to various appearance groups and has a high verification speed. In addition, in the embodiment of the present invention, an artificial neural network model for deep learning is further used, which greatly improves the accuracy when verifying the identity using optical flow information.

Although the present invention has been disclosed as above with examples, it is not intended to limit the present invention. Any person with ordinary knowledge in the technical field can make some changes and modifications without departing from the spirit and scope of the present invention. The scope of protection of the present invention shall be subject to the scope defined in the appended patent application.

100: Electronic device 110: Processor 120: Image capture component 130: Prompt component 140: Storage component 150: Neural network model S210, S220, S230, S231, S233: Neural network model trained in identity verification method Steps S310, S320, S330, S340, S350: steps of identity verification in the identity verification method

FIG. 1 is a schematic diagram of an electronic device according to an embodiment of the invention. FIG. 2 shows a flowchart of training a neural network model according to an embodiment of the invention. FIG. 3 illustrates a flowchart of identity verification according to an embodiment of the invention.

S310, S320, S330, S340, S350: steps of identity verification in the identity verification method

Claims (10)

An identity verification method is suitable for an electronic device. The identity verification method includes: sending out an action prompt message; activating an image capturing element of the electronic device to obtain a first video stream; according to the first video stream S2 continuous images, calculate and obtain a first optical flow information; and input the first optical flow information to a type of neural network model to obtain an identity verification result, where the type of neural network model includes one of the prompt messages corresponding to the action The action tag, at least one second optical flow information corresponding to the action tag, and identity registration information corresponding to each of the at least one second optical flow information. According to the identity verification method described in item 1 of the patent application scope, before the step of issuing the action prompt message, the method further includes: collecting at least one second video stream of at least one user according to different multiple action tags; For each of the two consecutive images in the at least one second video stream, calculating and obtaining the at least one second optical flow information of the at least one second video stream; and using the at least one second optical flow information to train the type of nerve Network model. The identity verification method as described in item 2 of the patent application scope, wherein the action prompt message corresponds to one of the action tags. The identity verification method as described in item 2 of the patent application scope, wherein the step of training the neural network model using the at least one second optical flow information includes: defining an identity registration information to the corresponding second optical flow information. The identity verification method as described in item 1 of the patent application scope, wherein the processor triggers an execution event of the electronic device according to the identity verification result. An electronic device includes: a prompt component for sending out a motion prompt message; an image capturing component, after the motion prompt message is sent, the image capturing component is activated to obtain a first video stream; And a processor, the signal is connected to the prompt component and the image capturing component, and the processor is used to run a type of neural network model, wherein the type of neural network model includes an action label corresponding to the action prompt message, corresponding to At least one second optical flow information of the motion tag and an identity registration information corresponding to each of the at least one second optical flow information, the processor is used to perform the following steps: Based on two consecutive images in the first video stream, the calculation is obtained A first optical flow information; and input the first optical flow information to the neural network model to obtain an identity verification result. The electronic device as described in item 6 of the patent application scope, wherein before the prompting element sends out the action prompting message, the processor is further used to perform the following steps: collect at least one of at least one user according to different multiple action tags Second video stream; calculating the at least one second optical flow information of the at least one second video stream according to two consecutive images in each of the at least one second video stream; and using the at least one second Optical flow information trains this type of neural network model. The electronic device as described in item 7 of the patent application scope, wherein the action prompt message sent by the prompt element corresponds to one of the action tags. The electronic device as described in item 7 of the patent application scope, wherein when using the at least one second optical flow information to train the neural network model, the processor is further used to perform the following steps: define an identity registration information to the corresponding The second optical flow information. The electronic device as described in item 6 of the patent application scope, wherein the processor triggers an execution event of the electronic device according to the identity verification result.

Images