阅读说明：本技术 基于自动编码器的击键动力学身份认证与识别方法及系统 (Keystroke dynamics identity authentication and identification method and system based on automatic encoder ) 是由 章行健 薛质 施勇 于 2021-07-30 设计创作，主要内容包括：本发明涉及一种基于自动编码器的击键动力学身份认证与识别方法及系统,所述方法包括以下步骤：1)获取对象的键入的按键键值和击键动力学数据,校验所述按键键值是否与预设键值相同,若是,则执行步骤2),若否,则产生识别不通过的提示信息；2)根据预设的采样精度对所述击键动力学数据进行处理,生成击键模式灰度图；3)以所述击键模式灰度图作为预先训练好的自动编码器模型的输入,实现身份认证与识别。与现有技术相比,本发明具有对数据集要求低、识别准确度高等优点。(The invention relates to a keystroke dynamics identity authentication and identification method and a system based on an automatic encoder, wherein the method comprises the following steps: 1) acquiring key values and keystroke dynamics data input by an object, checking whether the key values are the same as preset key values, if so, executing the step 2), and if not, generating prompt information for failing identification; 2) processing the keystroke dynamics data according to preset sampling precision to generate a keystroke mode gray graph; 3) and the keystroke mode gray-scale image is used as the input of a pre-trained automatic encoder model to realize identity authentication and identification. Compared with the prior art, the method has the advantages of low requirement on the data set, high identification accuracy and the like.)

1. A keystroke dynamics identity authentication and identification method based on an automatic encoder is characterized by comprising the following steps:

1) acquiring key values and keystroke dynamics data input by an object, checking whether the key values are the same as preset key values, if so, executing the step 2), and if not, generating prompt information for failing identification;

2) processing the keystroke dynamics data according to preset sampling precision to generate a keystroke mode gray graph;

3) and the keystroke mode gray-scale image is used as the input of a pre-trained automatic encoder model to realize identity authentication and identification.

2. The automated encoder-based keystroke dynamics identity authentication and identification method of claim 1, wherein the keystroke dynamics data comprises a time of depression and a time of release of each key.

3. The autoencoder-based keystroke dynamics identity authentication and recognition method of claim 1, wherein the sampling precision is on the order of milliseconds.

4. The automated encoder-based keystroke dynamics authentication and identification method of claim 1, wherein the keystroke pattern gray scale map represents the length of time each key is depressed at an equivalent sampling density in white, the typing time difference between the key and the next key at an equivalent sampling density in gray, and the black represents zero-valued fill at a set image width.

5. The autoencoder-based keystroke dynamics identity authentication and recognition method of claim 1, wherein the autoencoder model comprises a network of autoencoders for extracting high-dimensional features and a classification layer that implements classification based on the high-dimensional features.

6. The autoencoder-based keystroke dynamics identity authentication and recognition method of claim 1, wherein after each recognition is completed, data that is classified as entered by the user is added to the training database and the autoencoder model is updated.

7. The automated encoder-based keystroke dynamics identity authentication and recognition method of claim 6, wherein the total amount of data in the training database is constant.

8. A keystroke dynamics identity authentication and identification system based on an autoencoder, comprising:

the keystroke dynamics data acquisition module is used for acquiring a keystroke key value and keystroke dynamics data which are input by an object, verifying whether the keystroke key value is the same as a preset key value, if so, saving the keystroke key value, and if not, abandoning the keystroke key value;

the gray level image generation module is used for processing the keystroke dynamics data stored in the keystroke dynamics data acquisition module according to the preset sampling precision to generate a keystroke mode gray level image;

the automatic encoder model training module is used for maintaining a training database, storing a keystroke pattern gray-scale image in the training database and training on the basis of the training database to obtain an automatic encoder model;

and the identity authentication and identification module classifies the keystroke mode gray level images to be identified, which are acquired in real time, by utilizing the trained automatic encoder model, so as to realize identity authentication and identification.

9. The automated encoder-based keystroke dynamics identity authentication and recognition system of claim 8, further comprising:

and the model updating module is used for adding the data which is classified and recognized as the data input by the user into the training database after the identification is finished every time, and sending a model updating instruction to the automatic encoder model training module.

10. A computer-readable storage medium comprising one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing the autoencoder-based keystroke dynamics identity authentication and recognition method of any of claims 1-7.

Technical Field

The invention relates to the technical field of identity verification, in particular to a keystroke dynamics identity authentication and identification method and system based on an automatic encoder.

Background

The identity authentication and identification technology protects information from illegal leakage and tampering and protects system functions from abuse by ensuring that the physical identity of an operator corresponds to the digital identity, and is an important technology in the field of information security.

The identity authentication and identification technology can be divided into single-factor authentication and double-factor authentication according to authentication conditions; software authentication and hardware authentication can be classified according to whether hardware is used; the authentication information can be classified into static authentication and dynamic authentication. The most common and basic method of identity authentication at present is static password authentication. In static password authentication, a user finishes an authentication process by inputting a preset correct password, the method is easy to guess and leak, and the password needs to be transmitted in a computer memory and a network in the authentication process, so that the security of the password is difficult to guarantee. Thus, a two-factor authentication method based on static password authentication has been developed to facilitate various other authentication means.

In recent years, biometric identification technology has become a research focus as a means of identity identification, because it is not easy to be forgotten and cracked like a password, and is not easy to be stolen and transferred like a held object. In particular, the biometric features used for identification need to meet basic requirements of universality, uniqueness, stability, collectability and the like, and the identification system established by the biometric features needs to be balanced in robustness, acceptability and deceptibility, so that the identification based on the biometric features can be realized. The current mainstream biometric identification technology includes face identification, fingerprint identification, iris identification, hand shape identification, palm print identification, voice identification and the like. Early in the eighties of the last century, studies by the U.S. natural foundation and the U.S. national standards institute have shown that keyboard typing patterns contain unique features that can be used for identification. Based on this study, the application of keystroke dynamics in identity authentication techniques was established.

Keystroke dynamics authentication studies can be divided into two major directions, fixed text studies and free text studies. The former analyzes fixed data typed by a user, and the latter analyzes arbitrary text typed by the user during the operation of the system. Compared with biological characteristics such as fingerprints and irises, the method has the advantages that: keystroke behavior data can be obtained without additional equipment; data acquisition is transparent to the user; the data can be continuously obtained. However, limited by the small size and instability of keystroke dynamics data, especially the lack of sample size and coverage and interference of factors in the subject with factors outside the environment, the accuracy of the current keystroke dynamics model still cannot be used alone as a means of identity authentication and identification like a human face. In the existing identity authentication method based on keystroke dynamics, keystroke timing characteristics, namely the characteristics of the keystroke duration, the keystroke pressing interval, the keystroke releasing interval and the like of a testee and statistical characteristics thereof are generally analyzed, and the keystroke modes of the testee are derived and distinguished by using the principles of statistical theory, machine learning, deep learning and the like, so that the identity authentication and identification functions are realized.

The statistical theory is a method adopted by early research of keystroke behavior characteristic analysis, and represents the identity authentication and identification method based on the probability distribution of time required by continuous input, which is proposed by Gaines et al in 1980. The method usually calculates the mean, variance and correlation coefficient of the sample feature vector, and uses hypothesis testing and distance measurement to obtain the classification result. Although some documents claim to achieve better classification effect, the sample size used by the documents is small, and the conclusion confidence is not high.

Statistical methods can only obtain rough decision information, while machine learning can obtain richer information from the sample vector space. Machine learning is a common method for keystroke behavior feature analysis, including: the identity Authentication and identification method implemented by Ulinskasas et al Using k-adjacent space classifier [ Ulinskas M, Wniak M, Damaegius R.analysis of Keystone Dynamics for Fatitue Recognition [ C ]// International Conference on Computational Science & Its applications, 2018 Ho et al, the identity Authentication and identification method implemented by a class of Bayesian classifier [ Ho, Jianke, Kang, et al, one-class environmental benefits with duration Authentication for access User Authentication, 2018 fuzzy logic [ J ]. Applied to the Authentication of the identity Authentication and identification method implemented by the aid User based on a class of Bayesian classifier [ Wood ] J.application of forest management of scientific theory, 2018, Kresting et al, the random identification method implemented by Ulinscription International Journal of identity Authentication and identification of forest management Ru, Kresting, 2018, and the random identification method implemented by Keystone of forest management, Kresting, and class of forest management, and identification method implemented by Krestine, No. 8, Krestine, No. 4, No. 7, an Eloff J.enhanced paper Authentication through Fuzzy Logic [ J ]. IEEE Expert: Intelligent Systems and Thr Applications,1997 ], 2015-year method of identification and recognition by Ceker et al based on Gaussian mixture model [ Ceker H, Updhyaya S.enhanced registration of key dynamics using Gaussian mixture models [ C ]// Military Communications conference. IEEE,2015 ] and the like.

In the deep learning field, the continuous identity authentication of free text of keystroke dynamics [ reed effect peak, Zhang Sheng, Yisheng Wei ] is realized by adopting a Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) method in 2018 by Luxian peak and the like, the continuous identity authentication of free text based on the keystroke modes of the CNN and the RNN [ J ]. Qinghua university bulletin (Nature science edition), 2018,58(12): 1072-.

The prior keystroke dynamics identity authentication and identification technology has two defects, one is that a supervised classification method is mostly adopted, prior knowledge and manual labels are needed, and the method is not suitable for practical application scenes with a lack of sample quantity; secondly, most of the methods adopt key stroke dynamics time sequence characteristics, have limitations on characteristic collection and scale, and need to develop a new characteristic extraction method.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provide the keystroke dynamics identity authentication and identification method and system based on the automatic encoder, which have low requirement on a data set and high identification accuracy.

The purpose of the invention can be realized by the following technical scheme:

in a first aspect, the invention provides a keystroke dynamics identity authentication and identification method based on an automatic encoder, which comprises the following steps:

1) acquiring key values and keystroke dynamics data input by an object, checking whether the key values are the same as preset key values, if so, executing the step 2), and if not, generating prompt information for failing identification;

2) processing the keystroke dynamics data according to preset sampling precision to generate a keystroke mode gray graph;

3) and the keystroke mode gray-scale image is used as the input of a pre-trained automatic encoder model to realize identity authentication and identification.

Further, the keystroke dynamics data includes a time of depression and a time of release of each key.

Further, the sampling precision is in milliseconds.

Further, in the keystroke pattern grayscale map, the time length of each key pressed at the same sampling density is represented by a white portion, the typing time difference between the key and the next key at the same sampling density is represented by a gray portion, and a black portion represents zero-valued fill at the set image width.

Further, the auto-encoder model includes an auto-encoder network for extracting high-dimensional features and a classification layer that implements classification based on the high-dimensional features.

Further, after each recognition, data entered by the user himself/herself as a result of the classification recognition is added to the training database, and the automatic encoder model is updated.

Further, the total amount of data in the training database is constant.

In a second aspect, the present invention provides a keystroke dynamics identity authentication and identification system based on an automatic encoder, comprising:

the keystroke dynamics data acquisition module is used for acquiring a keystroke key value and keystroke dynamics data which are input by an object, verifying whether the keystroke key value is the same as a preset key value, if so, saving the keystroke key value, and if not, abandoning the keystroke key value;

the gray level image generation module is used for processing the keystroke dynamics data stored in the keystroke dynamics data acquisition module according to the preset sampling precision to generate a keystroke mode gray level image;

the automatic encoder model training module is used for maintaining a training database, storing a keystroke pattern gray-scale image in the training database and training on the basis of the training database to obtain an automatic encoder model;

and the identity authentication and identification module classifies the keystroke mode gray level images to be identified, which are acquired in real time, by utilizing the trained automatic encoder model, so as to realize identity authentication and identification.

Further, the system further comprises:

and the model updating module is used for adding the data which is classified and recognized as the data input by the user into the training database after the identification is finished every time, and sending a model updating instruction to the automatic encoder model training module.

In a third aspect, the invention provides a computer-readable storage medium, comprising one or more programs for execution by one or more processors of an electronic device, the one or more programs including instructions for performing an auto-encoder based keystroke dynamics identity authentication and recognition method as described above.

Compared with the prior art, the invention has the following beneficial effects:

1. the invention adopts the principle of an Automatic Encoder (AE) of self-supervision on the classification method, does not depend on prior knowledge and labels, reduces the requirements of the classifier on the data set due to the self characteristics of the automatic encoder, can still well extract the characteristic information of a normal sample on the data set of a small number of abnormal samples, and is more suitable for practical application scenes.

2. In the feature extraction method, the invention introduces image information and replaces the traditional key stroke dynamics time sequence features with gray level images, thereby realizing a feature extraction means which is more efficient and more beneficial to classification.

3. After each recognition, the data input by the user is added into the training data, and the automatic encoder model is updated, so that the model is fit with the behavior habit of the user, stable self-adaptive adjustment aiming at the change of user proficiency and the like is realized, and the classification precision and the user experience are improved.

4. According to the invention, through the key stroke dynamics data gray level imaging and the introduction of the automatic encoder as a classifier, the calculation cost is greatly saved, the recognition accuracy is improved, the practical application scene is better fitted, and a new way of key stroke dynamics identity authentication and recognition is expanded.

Drawings

FIG. 1 is a schematic flow diagram of the present invention;

FIG. 2 is a gray scale diagram of the keystroke pattern generated by the present invention.

Detailed Description

The invention is described in detail below with reference to the figures and specific embodiments. The present embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and a specific operation process are given, but the scope of the present invention is not limited to the following embodiments.

Example 1

As shown in fig. 1, the present embodiment provides a keystroke dynamics identity authentication and identification method based on an automatic encoder, which includes the following steps:

1) acquiring key values and keystroke dynamics data of key-in of an object, checking whether the key values are the same as preset key values, if so, executing the step 2), and if not, generating prompt information for failing identification. The keystroke dynamics data includes the time of depression and the time of release of each key.

Specifically, the first key pressing is taken as the starting time, the corresponding key-stroke behavior timestamps are respectively recorded, the accuracy is as millisecond, whether the content of the key values is correct or not is verified after the input is finished, and if the content of the key values is correct, the next step is carried out.

2) And processing the keystroke dynamics data according to the preset sampling precision to generate a keystroke mode gray-scale map.

As shown in fig. 2, in the keystroke pattern gray-scale graph obtained by the method, the white part represents the time length of each key value pressed under the same sampling density; the grey part represents the keying time difference between the key value and the next key value under the same sampling density; the black part is filled with zero value under the set image width for ensuring the image rendering effect. For example, if the key value a is pressed for 0.05 seconds and the difference between the key value a and the next key value N is 0.03, the white portion corresponding to the key value a is 5 pixels, the gray portion corresponding to the key value a is 3 pixels, and the black portion corresponding to the key value a is 2 pixels at a sampling density of 0.01 and an image width of 10.

3) And the keystroke mode gray-scale image is used as the input of a pre-trained automatic encoder model to realize identity authentication and identification.

The automatic encoder model employed by the present invention includes an automatic encoder network for extracting high-dimensional features and a classification layer that implements classification based on the high-dimensional features. The automatic encoder network is one kind of neural network, and its basic idea is to directly use one or more layers of neural networks to map the input data to obtain the output vector as the feature extracted from the input data, and obtain the best feature of the reconstructed data, i.e. the implicit feature of the data, by two operations of encoding and decoding. The classification layer is added to the autoencoder network, and finally, the autoencoder can be used as a classifier.

The automatic encoder model is obtained by optimization based on a pre-stored database for training, in this embodiment, 100 pieces of data are stored in the database for training. The user needs to enter training data in advance in the initial stage, and high-dimensional features are extracted through an automatic encoder network to obtain a reliable classification model.

After the trained automatic encoder model is obtained, the current typing characteristics of the user can be classified, the data identified as the user is passed, and the data identified as the non-user is rejected, so that the identity authentication and identification function based on keystroke dynamics is finally realized.

In a preferred embodiment, after the automatic encoder model is established, the data input by the user himself/herself classified is added to the training data according to the classification result of the authentication and identification module for each classification, and the oldest data item is deleted while the total amount of the training data is kept constant. The method is beneficial to enabling the classification model to fit the behavior habit of the user, realizing stable self-adaptive adjustment aiming at the change of user proficiency and the like, and improving the classification precision and the user experience.

The above functions, if implemented in the form of software functional units and sold or used as a separate product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

Example 2

The embodiment provides a keystroke dynamics identity authentication and identification system based on an automatic encoder, which comprises a keystroke dynamics data acquisition module, a gray level image generation module, an automatic encoder model training module and an identity authentication and identification module, wherein the keystroke dynamics data acquisition module is used for acquiring a keystroke key value and keystroke dynamics data input by an object, verifying whether the keystroke key value is the same as a preset key value, if so, saving, and if not, discarding; the gray level image generation module is used for processing the keystroke dynamics data stored in the keystroke dynamics data acquisition module according to the preset sampling precision to generate a keystroke mode gray level image; the automatic encoder model training module maintains a training database, the training database stores a keystroke pattern gray-scale image, and an automatic encoder model is obtained based on the training database; and the identity authentication and identification module classifies the keystroke mode gray-scale image to be identified, which is acquired in real time, by using the trained automatic encoder model, so as to realize identity authentication and identification.

Referring to fig. 1, the specific process of performing identity authentication and identification by using the system includes:

step one, a user enters data (usually a password);

secondly, recording input of a user and keystroke dynamics data corresponding to the input by the system;

converting the recorded keystroke dynamics data into a gray image by the system through a gray image generation module;

if the data is training data, storing the data in the local, and generating a classification model for identity authentication and identification through an automatic encoder model training module;

and step five, the identity authentication and identification module classifies the data by utilizing the classification model generated by the automatic encoder model training module, so that the identity authentication and identification function is realized.

In a preferred embodiment, the system further comprises: and the model updating module is used for adding the data which is classified and recognized as the data input by the user into the training database after the identification is finished every time, and sending a model updating instruction to the automatic encoder model training module.

The foregoing detailed description of the preferred embodiments of the invention has been presented. It should be understood that numerous modifications and variations could be devised by those skilled in the art in light of the present teachings without departing from the inventive concepts. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.