WO2021128682A1

WO2021128682A1 - Machine learning-based web page rendering method and apparatus, and computer device

Info

Publication number: WO2021128682A1
Application number: PCT/CN2020/088015
Authority: WO
Inventors: 温桂龙
Original assignee: 深圳壹账通智能科技有限公司
Priority date: 2019-12-23
Filing date: 2020-04-30
Publication date: 2021-07-01
Also published as: CN111177622A

Abstract

Disclosed by the present application are a machine learning-based web page rendering method and apparatus, a computer device and a storage medium. The method comprises: collecting a first face image of a user by using a preset camera; calculating an image similarity value between the first face image and a specified face image; if the similarity value is not greater than a preset image similarity threshold, then inputting the first face image into a linear classifier in a preset predictive model for calculation to obtain feature information outputted by the linear classifier; inputting the feature information into a non-linear classifier for calculation to obtain web page rendering information composed of multiple pieces of sub-information; correspondingly sending the multiple pieces of sub-information to a plurality of servers; and receiving a plurality of sub-data packets correspondingly returned by the plurality of servers and combining same into a rendering data packet, and using the rendering data packet to render a web page. Therefore, web page rendering is adaptive to special groups of people (such as elderly people), and the efficiency of web page usage is increased.

Description

Web page rendering method, device and computer equipment based on machine learning

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on December 23, 2019, the application number is 201911342654.1, and the application title is "Machine Learning-based Web Page Rendering Method, Apparatus, and Computer Equipment", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the computer field, and in particular to a web page rendering method, device, computer equipment and storage medium based on machine learning.

Background technique

With the rapid development of the Internet industry, people are now accustomed to browsing web pages on the Internet using computer equipment. The inventor realized that the traditional web page (or web page rendering method) only has a few fixed style web page settings, such as day mode web page, night mode web page, theme mode web page, etc., but the number of these web page settings is limited, and these Web page settings need to be manually set by the user, and cannot be automatically adjusted according to the user's characteristics. In addition, for the elderly, due to physiological factors such as reduced eyesight, they are more suitable for web pages with simple layout and larger font size. However, the web pages of traditional technologies cannot be automatically adjusted according to the characteristics of the user population, resulting in low efficiency for users to use the web pages.

technical problem

The main purpose of this application is to provide a web page rendering method, device, computer equipment, and storage medium based on machine learning, aiming to make web page rendering adaptive to special groups of people (such as elderly people) and improve web page usage efficiency.

Technical solutions

In order to achieve the above-mentioned purpose of the invention, this application proposes a web page rendering method based on machine learning, which includes the following steps:

Use the preset camera to collect the user's first face image;

According to a preset image similarity calculation method, calculate an image similarity value between the first face image and a prestored designated face image, and determine whether the similarity value is greater than a preset image similarity threshold;

If the similarity value is not greater than the preset image similarity threshold, the first face image is input into the linear classifier in the preset prediction model for calculation, so as to obtain the features output by the linear classifier Information, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset non-linear classifier; the linear classifier is pre-processed by including facial images and Trained on the sample data of the feature information corresponding to the face image;

The feature information is input into the non-linear classifier for calculation, so as to obtain the web page rendering information composed of multiple sub-information output by the non-linear classifier, wherein the multiple sub-information includes at least the color sub-information of the web page, Icon style sub-information and layout style sub-information; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

Sending the multiple sub-information and returning the request information of the sub-data package used for web page rendering to multiple corresponding servers;

Receiving the multiple sub-data packets correspondingly returned by the multiple servers, combining the multiple sub-data packets into a rendering data packet, and rendering the webpage by using the rendering data packet.

This application provides a web page rendering device based on machine learning, including:

The first face image acquisition unit is configured to use a preset camera to collect the user's first face image;

The image similarity value judgment unit is used to calculate the image similarity value between the first face image and a pre-stored designated face image according to a preset image similarity calculation method, and determine whether the similarity value is Greater than the preset image similarity threshold;

The feature information acquiring unit is configured to, if the similarity value is not greater than the preset image similarity threshold, input the first face image into the linear classifier in the preset prediction model for calculation, so as to obtain the The feature information output by the linear classifier, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset nonlinear classifier; the linear classifier It is pre-trained through sample data including facial images and feature information corresponding to facial images;

The web page rendering information acquisition unit is configured to input the feature information into the nonlinear classifier for calculation, so as to obtain web page rendering information composed of multiple sub-information output by the nonlinear classifier, wherein the multiple sub-information Including at least color sub-information, icon style sub-information, and layout style sub-information of the webpage; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

A sub-information sending unit, configured to send the multiple sub-information and return request information of the sub-data package used for web page rendering to multiple corresponding servers;

The web page rendering unit is configured to receive the multiple sub-data packets correspondingly returned by the multiple servers, combine the multiple sub-data packets into a rendering data packet, and render the web page by using the rendering data packet.

The present application provides a computer device including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above-mentioned machine learning-based webpage rendering method when the computer program is executed.

The present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method for rendering a webpage based on machine learning are realized.

Beneficial effect

This application realizes the adaptation of web page rendering to special people (such as the elderly) and improves the efficiency of web page use.

Description of the drawings

FIG. 1 is a schematic flowchart of a webpage rendering method based on machine learning according to an embodiment of the application;

2 is a schematic block diagram of the structure of a web page rendering apparatus based on machine learning according to an embodiment of the application;

FIG. 3 is a schematic block diagram of the structure of a computer device according to an embodiment of the application.

The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

The best mode of the present invention

In order to make the purpose, technical solutions, and advantages of this application clearer and clearer, the following further describes the application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

1, an embodiment of the present application provides a method for rendering a webpage based on machine learning, including the following steps:

S1. Use a preset camera to collect the user's first face image;

S2, according to the preset image similarity calculation method, calculate the image similarity value between the first face image and a pre-stored designated face image, and determine whether the similarity value is greater than the preset image similarity Threshold

S3. If the similarity value is not greater than the preset image similarity threshold, input the first face image into the linear classifier in the preset prediction model for calculation, thereby obtaining the linear classifier output The feature information includes at least the predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and the preset nonlinear classifier; the linear classifier is pre-processed by including facial The image and the sample data of the feature information corresponding to the face image are trained;

S4. Input the feature information into the non-linear classifier for calculation, so as to obtain the web page rendering information composed of multiple sub-information output by the non-linear classifier, wherein the multiple sub-information at least includes the color sub-information of the web page. Information, icon style sub-information, and layout style sub-information; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

S5. Send the multiple sub-information and return the request information of the sub-data package used for web page rendering to multiple corresponding servers;

S6. Receive the multiple sub-data packets correspondingly returned by the multiple servers, combine the multiple sub-data packets into a rendering data package, and render the webpage by using the rendering data package.

As described in step S1 above, a preset camera is used to collect the user's first face image. This application can provide a suitable webpage according to the actual situation of the user, so as to improve the efficiency of the user's use of the meeting. Accordingly, the preset camera is used to collect the user's first face image. The first face image will be used as a basis for obtaining user information and be accurately analyzed.

As described in step S2 above, according to the preset image similarity calculation method, calculate the image similarity value between the first face image and a pre-stored designated face image, and determine whether the similarity value is greater than the preset image similarity value. Set the image similarity threshold. The method for acquiring the designated face image can be any feasible method, for example: acquiring the user's login information (such as user account, user ID number, etc.), according to the preset login information-the correspondence of the face image Relationship, retrieve the designated face image corresponding to the login information from the preset face image database. In addition, one of the purposes of judging whether the similarity value is greater than a preset image similarity threshold includes at least screening out inaccurate users (for example, natural person A uses the account of user B to log in) to improve The accuracy of the rendering of the web pages of this application. Wherein, the number of the designated face image is one. The image similarity value is used to measure whether the first face image is similar to the pre-stored specified face image. If the similarity value is high (that is, the similarity value is greater than the preset image similarity threshold), it indicates the specified face The image is the user's face image, that is, the user's data is obtained in advance, so there is no need to further analyze the first face image, and directly according to the preset correspondence relationship between the face image and the data packet, obtain the specified face A designated data package corresponding to the image, wherein the designated data package is used to render a web page; the designated data package is used to render the web page to provide a suitable web page. Among them, this application pre-collects the face images of multiple users and the customary web page rendering data (ie data packets) of multiple users, so as to form the correspondence between the face images and the data packets, so that when knowing the user’s After the face image, the corresponding data packet can be obtained according to the corresponding relationship between the face image and the data packet. If the similarity value is low, the first face image needs to be further analyzed. Wherein, the image similarity calculation method is, for example, performing scaling processing on the first face image according to a preset scaling method to obtain a scaled picture, wherein the scaled picture is the same size as the specified face image Same (the same size means that the length and width of the picture are the same, thereby facilitating the comparison of pictures); acquiring multiple first feature lengths in the zoomed picture, and acquiring multiple second features in the specified face image Length, wherein each of the plurality of first characteristic lengths and the plurality of second characteristic lengths includes at least a face length and a face width, and the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence; Generate the first matrix [U1,U2,...,Un] and the second matrix [P1,P2,...,Pn], where U1,U2,...,Un are the multiple first feature lengths , P1, P2,..., Pn are the multiple second characteristic lengths, there are n first characteristic lengths and n second characteristic lengths in total; according to the formula: M=1/|||[U1,U2, ...,Un] [P1,P2,...,Pn] ^T ||-n|, calculate the image similarity value M between the first face image and the specified face image.

As described in step S3 above, if the similarity value is not greater than the preset image similarity threshold, the first face image is input into the linear classifier in the preset prediction model for calculation, so as to obtain the The feature information output by the linear classifier, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset nonlinear classifier. This application adopts a segmented prediction model, that is, a prediction model formed by sequentially connecting the linear classifier and the preset nonlinear classifier to process the first face image. Since the segmented setting is adopted, it is beneficial to quickly analyze the cause of the error when the prediction model makes an error. The linear classifier can be any feasible model, for example, a support vector machine (linear kernel), a Bayesian classifier, a single-layer perceptron, and so on. The non-linear classifier can be any feasible model, such as a neural network-based classifier, decision tree, etc.

As described in step S4 above, the feature information is input into the nonlinear classifier for calculation, so as to obtain the web page rendering information composed of a plurality of sub-information output by the nonlinear classifier, wherein the plurality of sub-information is at least Including webpage color sub-information, icon style sub-information and layout style sub-information. Wherein, the non-linear classifier may adopt a supervised learning method and be trained using pre-collected sample data, the sample data including feature information and multiple sub-information corresponding to the feature information. The non-linear classifier is used to predict appropriate web page rendering information for the user according to the feature information. In addition, this application adopts a method of rendering information on a webpage composed of multiple sub-information to realize the diversification of webpage settings.

As described in step S5 above, sending the multiple sub-information and returning the request information of the sub-data packet used for web page rendering to multiple corresponding servers. This application uses multiple servers to provide sub-information separately, so that the data required for web page rendering is scattered and distributed in different servers. Since each server only needs to provide sub-data packets, the computational pressure is relieved and more types can be provided. The number of sub-data packages for the web pages will be increased by an order of magnitude (because each server can focus on a certain sub-data package, so the number of sub-data packages increases, and the types of web pages that can be rendered after the combination Also greatly increased). Wherein, the sub-data package contains data required to render a webpage, such as webpage parameters, pictures of icons, and so on. Further, the webpage is generated by staged rendering, such as the rendering of the sidebar of the webpage, the rendering of the top menu bar, and the rendering of the icons of the entire webpage. The staged rendering only means that the rendering is carried out by project, but does not mean The rendering sequence, for example, the rendering of the sidebar of the webpage and the rendering of the top menu bar can be performed at the same time or at different times. Correspondingly, each of the sub-data packets is only used for one microstructure (item) of the rendered webpage, where the microstructure refers to the smallest unit that can be rendered separately when the webpage is rendered, such as the sidebar The function realization of the designated button, the icon rendering of the whole webpage, and so on.

As described in the above step S6, the multiple sub-data packets correspondingly returned by the multiple servers are received, the multiple sub-data packets are combined into a rendering data packet, and the webpage is rendered using the rendering data packet. Rendering a web page requires a rendering packet. By receiving the multiple sub-data packets correspondingly returned by the multiple servers and combining them into a rendering data packet, the computer can extract data from the rendering data packet when rendering the webpage, and then render the webpage.

In one embodiment, the step S2 of calculating the image similarity value between the first face image and a pre-stored designated face image according to a preset image similarity calculation method includes:

S201: Perform scaling processing on the first face image according to a preset scaling method to obtain a scaled picture, wherein the eyebrow and eye spacing of the scaled picture is equal to the eyebrow and eye spacing of the designated face image;

S202. Acquire multiple first feature lengths in the zoomed picture, and acquire multiple second feature lengths in the designated face image, wherein the multiple first feature lengths and the multiple second features The lengths include at least face length and face width, and the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence;

S203. Generate a first matrix [U1, U2,..., Un] and a second matrix [P1, P2,..., Pn], where U1, U2,..., Un are the first Characteristic lengths, P1, P2,..., Pn are the plurality of second characteristic lengths, there are n first characteristic lengths and n second characteristic lengths in total;

S204. According to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] ^T ||-n|, calculate the first face image The image similarity value M with the specified face image.

As described above, the calculation of the image similarity value between the first face image and a pre-stored designated face image is realized. Eyebrow spacing refers to the distance between a person's eyebrows and eyes. Different people may have different eyebrow spacing. According to the preset scaling method, this application performs scaling processing on the first face image to obtain a scaled picture, wherein the eyebrow and eye spacing of the scaled picture is equal to the eyebrow and eye spacing of the designated face image, so that the first The face image can be compared with the pre-stored specified face image in feature length. Then acquire multiple first feature lengths in the zoomed picture, and acquire multiple second feature lengths in the specified face image; generate a first matrix [U1, U2,..., Un] and a second matrix [U1, U2,..., Un] Matrix [P1,P2,...,Pn], according to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] ^T ||-n |, calculate the image similarity value M between the first face image and the specified face image. If the first face image is similar to the pre-stored designated face image, then the value of ||[U1,U2,...,Un] [P1,P2,...,Pn] ^T || will be It is close to n, so the value of M will be close to infinity, so the image similarity value M can be used as a standard to measure the image similarity value. In addition, due to the structural characteristics of the human body, if a certain feature length of face A is greater than face B, in general, all feature lengths of face A are greater than face B, so this application adopts the formula: M=1 /|||[U1,U2,...,Un] [P1,P2,...,Pn] ^T ||-n|, can amplify the difference in feature length, thereby improving the calculation of image similarity accuracy.

In one embodiment, the step S201 of performing zooming processing on the first face image according to a preset zooming method to obtain a zoomed picture includes:

S2011. Use a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and use the facial feature point detection model to perform detection on the specified face image Perform detection to obtain multiple second facial feature points;

S2012. Generate a first minimum circumscribed rectangle of the plurality of first facial feature points, and generate a second minimum circumscribed rectangle of the plurality of second facial feature points;

S2013: Perform scaling processing on the first face image by using an equal-scale scaling method, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, thereby obtaining a scaled picture.

As described above, it is realized that the first face image is zoomed according to the preset zooming method, so as to obtain a zoomed picture. Accurate scaling is a prerequisite for image similarity calculation and is extremely important. This application uses a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and uses the facial feature point detection model to detect the specified face image Perform detection to obtain a plurality of second facial feature points; generate a first minimum circumscribed rectangle of the plurality of first facial feature points, and generate a second minimum circumscribed rectangle of the plurality of second facial feature points; adopt The method of proportional scaling performs scaling processing on the first face image so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, which improves the accuracy of scaling. Wherein, the facial feature point detection model may be any model, for example, a model based on neural network model training. The facial feature point detection model is obtained, for example, in the following manner: calling sample data, and dividing the sample data into training data and test data; wherein, the sample data is only composed of face images and face images placed in a standard posture. The composition of the labeled feature points; use the training data and use the stochastic gradient descent method to train the preset neural network model to obtain the facial feature point intermediate detection model; use the test data to verify the facial feature point intermediate detection model; if the verification passes, Then, the intermediate detection model for facial feature points is recorded as the facial feature point detection model. Among them, the facial feature point is, for example, the midpoint of the pupil. Since the structure of each face is similar, they all have the same feature points. Of course, the location of the feature points varies from person to person. The present application adopts the method of generating the first minimum circumscribed rectangle of the plurality of first facial feature points and the method of generating the second minimum circumscribed rectangle of the plurality of second facial feature points, using all facial feature points. Then, the first face image is scaled by a method of equal scaling, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, so as to obtain a scaled image to achieve accurate Zoom.

In one embodiment, the facial feature point detection model is only trained by a face image placed in a preset standard posture, the designated face image is placed in a standard posture, and the preset facial feature points are used. The detection model detects the first face image to obtain a plurality of first facial feature points; and uses the facial feature point detection model to detect the specified face image, thereby obtaining a plurality of second facial features. The step S2011 of facial feature points includes:

S20111. Use a preset facial feature point detection model to detect the first face image to obtain a first number, and use the facial feature point detection model to detect the second face image to obtain A second number, wherein the first number refers to the number of facial feature points in the first face image, and the second number refers to the number of facial feature points in the designated face image;

S20112. Determine whether the absolute value of the difference between the first quantity and the second quantity is less than a preset difference threshold;

S20113. If the absolute value of the difference between the first number and the second number is not less than a preset difference threshold, perform one or more rotation processing on the first face image to obtain a picture Or multiple rotated images;

S20114. Use the facial feature point detection model to detect the rotated image in sequence until the absolute value of the difference between the number of detected facial feature points minus the second number is less than a preset difference threshold, and The facial feature point output by the facial feature point detection model for the last time is recorded as the first facial feature point;

S20115. Obtain the first facial feature point, and obtain the second facial feature point generated by the facial feature point detection model to detect the second face image.

As described above, it is achieved that the first face image is detected by using a preset facial feature point detection model, so as to obtain a plurality of first facial feature points; and, the facial feature point detection model is used to detect the first face image. Specify the face image for detection, thereby obtaining multiple second facial feature points. The facial feature point detection model used in this application is only trained from the face image placed in a preset standard posture, so that the facial feature point detection model can be trained in a short time (because only the face image placed in the standard posture , Does not consider other poses, thus effectively reducing the number of training samples), but correspondingly, the facial feature point detection model can only detect face images placed in standard poses. Therefore, the first face image needs to be rotated to be placed in a standard posture. Wherein, the facial feature point detection model may be any model, for example, a model based on neural network model training. The facial feature point detection model is obtained, for example, in the following manner: calling sample data, and dividing the sample data into training data and test data; wherein, the sample data is only composed of face images and face images placed in a standard posture. The composition of the labeled feature points; use the training data and use the stochastic gradient descent method to train the preset neural network model to obtain the facial feature point intermediate detection model; use the test data to verify the facial feature point intermediate detection model; if the verification passes, Then, the intermediate detection model for facial feature points is recorded as the facial feature point detection model. Wherein, the rotation processing may be any feasible rotation processing, for example, taking the center of the face image as the rotation center, rotating clockwise or counterclockwise by a preset angle. The preset angle is, for example, 1-180 degrees, preferably 90 degrees. Among them, it is precisely because the facial feature point detection model is only trained by the face image placed in the preset standard posture, the first face image placed in the non-standard posture will not be able to detect the facial feature points (or Only a small number of facial feature points can be detected). Therefore, the first face image is obtained by detecting the first face image by judging the preset facial feature point detection model, and the placement posture of the first face image can be obtained. If the absolute value of the difference between the first quantity minus the second quantity is not less than the preset difference threshold, it indicates that the placement posture of the first face image is not a standard posture, and therefore the first face image is not in a standard posture. The image is rotated one or more times; the rotated image is detected in turn using the facial feature point detection model until the absolute value of the difference between the number of detected facial feature points minus the second number is less than a preset Difference threshold, and record the last facial feature point output by the facial feature point detection model as the first facial feature point; obtain the first facial feature point, and obtain the facial feature point detection model for all The second facial image is detected to generate second facial feature points. Therefore, on the basis of ensuring the realization of facial feature point extraction, the training time of the facial feature point detection model is reduced.

In one embodiment, the image similarity value between the first face image and a pre-stored designated face image is calculated according to a preset image similarity calculation method, and it is determined whether the similarity value is greater than After step S2 of the preset image similarity threshold, the method includes:

S21. If the similarity value is greater than a preset image similarity threshold, obtain a specified data packet corresponding to the specified face image according to the preset correspondence between the face image and the data packet, wherein the specified face image Data packets are used to render web pages;

S22. Render a web page using the specified data package.

As described above, the use of the specified data package to render the web page is realized. If the similarity value is greater than the preset image similarity threshold, it indicates that the first face image is similar to the specified face image, that is, the specified face image can represent the user to which the first face image belongs Therefore, it is sufficient to directly use the specified data packet associated with the specified face image. This application pre-saves the user’s information by pre-storing the designated face image, so that the user only needs to use the prediction model to process the first face image when rendering a web page for the first time. The pre-stored specified data package can be directly called when rendering a web page once, thereby reducing unnecessary computing consumption and improving the rendering speed of the web page at the same time.

In one embodiment, if the similarity value is not greater than a preset image similarity threshold, the first face image is input into a linear classifier in a preset prediction model for calculation, thereby obtaining The feature information output by the linear classifier, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and the preset nonlinear classifier before step S3, include:

S211. Obtain pre-collected training data and test data, where both the training data and the test data include a facial image and feature information corresponding to the facial image;

S212. Adopt a stochastic gradient descent method and use the training data to train a preset initial linear classifier, so as to obtain an intermediate linear classifier;

S213. Use the test data to verify the intermediate linear classifier, and determine whether the intermediate linear classifier passes the verification;

S214: If the verification of the intermediate linear classifier passes, the intermediate linear classifier is used as the final linear classifier.

As mentioned above, the final linear classifier is obtained. The linear classifier can be any feasible model, for example, a support vector machine (linear kernel), a Bayesian classifier, a single-layer perceptron, and so on. This application uses a supervised learning method to train the linear classifier, and uses training data and test data that both include a facial image and feature information corresponding to the facial image, which are used for training and verification, respectively. The stochastic gradient descent method is used to train a preset initial linear classifier using the training data. The stochastic gradient descent method refers to randomly selecting multiple training data to replace all the training data for training, so as to improve the training speed. The test data is then used to verify the intermediate linear classifier. If the intermediate linear classifier passes the verification, it indicates that the intermediate linear classifier is available. Therefore, the intermediate linear classifier is used as the final linear classifier.

In one embodiment, the web page rendering method is applied to a rendering terminal, the rendering terminal and the multiple servers are connected to each other through a gateway, and the step S5 of correspondingly sending the multiple sub-information to the multiple servers, include:

S501: Encrypt the multiple sub-information separately by using a first encryption method previously agreed with multiple servers, so as to obtain multiple first sub-ciphertexts;

S502. Combine the multiple first sub-ciphertexts into an intermediate ciphertext.

S503: Encrypt the intermediate ciphertext by adopting a second encryption method agreed with the gateway in advance, so as to obtain a final ciphertext;

S504. Send the final ciphertext to the gateway, and request the gateway to decrypt and split the final ciphertext to obtain multiple first sub-ciphertexts, and combine the multiple first sub-ciphertexts Correspondingly sent to multiple servers.

As described above, the multiple sub-information is correspondingly sent to multiple servers. This application improves information security by means of secondary encryption. Both the first encryption method and the second encryption method may be any feasible encryption methods, including but not limited to symmetric encryption (for example, hash encryption) and asymmetric encryption. The rendering terminal and the multiple servers respectively pre-appoint the first encryption method, so the multiple servers can decrypt the first sub-ciphertext; the rendering terminal and the gateway pre-appoint the second encryption method, so the gateway can decrypt the final ciphertext. However, the gateway cannot know the specific information of the sub-ciphertext, and the server can only decrypt the corresponding first sub-ciphertext. Therefore, whether it is a third party who intends to steal information (need to intercept the final ciphertext, and also need to obtain the first encryption method and the second encryption method, so the final ciphertext is more difficult to be cracked), or gateways and multiple servers, it is difficult Obtain complete information, thereby improving information security.

2, an embodiment of the present application provides a web page rendering device based on machine learning, including:

The first facial image acquisition unit 10 is configured to use a preset camera to acquire a user's first facial image;

The image similarity value judgment unit 20 is configured to calculate the image similarity value between the first face image and a pre-stored designated face image according to a preset image similarity calculation method, and determine the similarity value Whether it is greater than the preset image similarity threshold;

The feature information acquiring unit 30 is configured to, if the similarity value is not greater than a preset image similarity threshold, input the first face image into a linear classifier in a preset prediction model for calculation, thereby obtaining The feature information output by the linear classifier, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset nonlinear classifier; the linear classifier The device is pre-trained through sample data including facial images and feature information corresponding to the facial images;

The web page rendering information acquisition unit 40 is configured to input the feature information into the nonlinear classifier for calculation, thereby obtaining web page rendering information composed of multiple sub-information output by the nonlinear classifier, wherein the multiple sub-information The information includes at least color sub-information, icon style sub-information, and layout style sub-information of the webpage; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

The sub-information sending unit 50 is configured to send the multiple sub-information and return the request information of the sub-data package used for web page rendering to multiple corresponding servers;

The web page rendering unit 60 is configured to receive the multiple sub-data packets correspondingly returned by the multiple servers, combine the multiple sub-data packets into a rendering data packet, and render the web page using the rendering data packet.

The operations performed by the above-mentioned units respectively correspond to the steps of the machine learning-based web page rendering method of the aforementioned embodiment, and will not be repeated here.

In an embodiment, the image similarity value judgment unit 20 includes:

A zooming subunit, configured to perform zooming processing on the first face image according to a preset zooming method to obtain a zoomed picture, wherein the eyebrow and eye spacing of the zoomed picture is equal to the eyebrow and eye spacing of the designated face image;

The feature length obtaining subunit is used to obtain multiple first feature lengths in the zoomed picture, and multiple second feature lengths in the designated face image, wherein the multiple first feature lengths and all Each of the plurality of second characteristic lengths includes at least a face length and a face width, and the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence;

The matrix generation subunit is used to generate the first matrix [U1,U2,...,Un] and the second matrix [P1,P2,...,Pn], where U1,U2,...,Un are all The plurality of first characteristic lengths, P1, P2,..., Pn are the plurality of second characteristic lengths, and there are n first characteristic lengths and n second characteristic lengths in total;

The image similarity value M is used to calculate the subunit according to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] ^T ||-n|, Calculate the image similarity value M between the first face image and the specified face image.

The operations performed by the aforementioned sub-units respectively correspond to the steps of the machine learning-based web page rendering method of the foregoing embodiment, and will not be repeated here.

In one embodiment, the zoom subunit includes:

The facial feature point detection module is configured to use a preset facial feature point detection model to detect the first face image, so as to obtain a plurality of first facial feature points; and, use the facial feature point detection model to Performing detection on the designated face image to obtain a plurality of second facial feature points;

A circumscribed rectangle generating module, configured to generate a first minimum circumscribed rectangle of the plurality of first facial feature points, and generate a second minimum circumscribed rectangle of the plurality of second facial feature points;

The scaling module is configured to perform scaling processing on the first face image by adopting an equal scaling method, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, thereby obtaining a scaled picture.

The operations performed by the above-mentioned modules respectively correspond to the steps of the machine learning-based web page rendering method of the foregoing embodiment, and will not be repeated here.

In one embodiment, the facial feature point detection model is only trained by a face image placed in a preset standard posture, the designated face image is placed in a standard posture, and the facial feature point detection module includes :

The quantity detection sub-module is configured to detect the first face image using a preset facial feature point detection model to obtain the first quantity, and use the facial feature point detection model to perform detection on the second face image Performing detection to obtain a second number, where the first number refers to the number of facial feature points in the first face image, and the second number refers to the number of facial feature points in the designated face image;

The difference threshold judging sub-module is used to judge whether the absolute value of the difference of the first quantity minus the second quantity is less than a preset difference threshold;

The rotation sub-module is configured to perform one or more rotation processing on the first face image if the absolute value of the difference between the first amount and the second amount is not less than a preset difference threshold, To get one or more rotated images;

The first facial feature point marking sub-module is used to sequentially use the facial feature point detection model to detect the rotated image until the absolute value of the difference between the number of detected facial feature points minus the second number is less than A preset difference threshold, and record the facial feature point output last time by the facial feature point detection model as the first facial feature point;

The feature point acquisition sub-module is configured to acquire the first facial feature point, and acquire the facial feature point detection model to detect the second face image, thereby generating the second facial feature point.

The operations performed by the above-mentioned sub-modules respectively correspond to the steps of the machine learning-based webpage rendering method of the foregoing embodiment, and will not be repeated here.

In one embodiment, the device includes:

A designated data packet acquisition unit, configured to, if the similarity value is greater than a preset image similarity threshold, acquire designated data corresponding to the designated face image according to the preset correspondence relationship between the face image and the data packet Package, wherein the specified data package is used to render a web page;

The designated data packet rendering unit is used to render the web page using the designated data packet.

In one embodiment, the device includes:

A data acquisition unit for acquiring pre-collected training data and test data, where the training data and test data both include facial images and feature information corresponding to the facial images;

An intermediate linear classifier acquiring unit, configured to adopt a stochastic gradient descent method and use the training data to train a preset initial linear classifier to obtain an intermediate linear classifier;

A verification unit, configured to verify the intermediate linear classifier using the test data, and determine whether the intermediate linear classifier passes the verification;

The linear classifier marking unit is configured to use the intermediate linear classifier as the final linear classifier if the verification of the intermediate linear classifier passes.

In one embodiment, the web page rendering method is applied to a rendering terminal, the rendering terminal and the multiple servers are connected to each other through a gateway, and the sub-information sending unit 50 includes:

The first encryption subunit is configured to separately encrypt the plurality of sub-information by adopting the first encryption method respectively agreed with the multiple servers to obtain multiple first sub-ciphertexts;

A ciphertext combination subunit for combining the plurality of first sub-ciphertexts into an intermediate ciphertext;

The second encryption subunit is used to encrypt the intermediate ciphertext by adopting the second encryption method agreed with the gateway in advance, so as to obtain the final ciphertext;

The final ciphertext sending subunit is used to send the final ciphertext to the gateway, and request the gateway to decrypt and split the final ciphertext to obtain a plurality of first sub-ciphertexts, and send the The multiple first sub-ciphertexts are correspondingly sent to multiple servers.

The webpage rendering device based on machine learning of the present application realizes webpage rendering adaptive to special people (for example, elderly people), and improves webpage usage efficiency.

3, an embodiment of the present application also provides a computer device. The computer device may be a server, and its internal structure may be as shown in the figure. The computer equipment includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor designed by the computer is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data used in the web page rendering method based on machine learning. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to realize a web page rendering method based on machine learning. The above-mentioned processor executes the above-mentioned machine learning-based webpage rendering method, wherein the steps included in the method respectively correspond to the steps of executing the machine learning-based webpage rendering method of the aforementioned embodiment, and will not be repeated here.

An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for rendering a webpage based on machine learning is implemented, wherein the steps included in the method are the same as those in the previous implementation. The steps of the web page rendering method based on machine learning correspond one to one, so I won’t repeat them here. The computer-readable storage medium may be non-volatile or volatile.

A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments and methods can be implemented by a computer program instructing relevant hardware. When the computer program is executed, it may include the processes as in the above-mentioned embodiments of the methods. . It should be noted that in this article, the terms "include", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements, It also includes other elements not explicitly listed, or elements inherent to the process, device, article, or method. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of this application, or directly or indirectly used in other related technical fields, are similarly included in the scope of patent protection of this application.

Claims

A web page rendering method based on machine learning, including:

Use the preset camera to collect the user's first face image;

According to a preset image similarity calculation method, calculate an image similarity value between the first face image and a prestored designated face image, and determine whether the similarity value is greater than a preset image similarity threshold;

If the similarity value is not greater than the preset image similarity threshold, the first face image is input into the linear classifier in the preset prediction model for calculation, so as to obtain the features output by the linear classifier Information, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset non-linear classifier; the linear classifier is pre-processed by including facial images and Trained on the sample data of the feature information corresponding to the face image;

The feature information is input into the non-linear classifier for calculation, so as to obtain the web page rendering information composed of multiple sub-information output by the non-linear classifier, wherein the multiple sub-information includes at least the color sub-information of the web page, Icon style sub-information and layout style sub-information; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

Sending the multiple sub-information and returning the request information of the sub-data package used for web page rendering to multiple corresponding servers;

Receiving the multiple sub-data packets correspondingly returned by the multiple servers, combining the multiple sub-data packets into a rendering data packet, and rendering the webpage by using the rendering data packet.
The method for rendering a webpage based on machine learning according to claim 1, wherein the image similarity between the first face image and a pre-stored specified face image is calculated according to a preset image similarity calculation method Value steps include:

Performing a scaling process on the first face image according to a preset scaling method to obtain a scaled picture, wherein the eyebrow and eye spacing of the scaled picture is equal to the eyebrow and eye spacing of the designated face image;

Acquire multiple first feature lengths in the zoomed picture, and acquire multiple second feature lengths in the specified face image, wherein the multiple first feature lengths and the multiple second feature lengths are both Comprising at least a face length and a face width, and the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence;

Generate the first matrix [U1,U2,...,Un] and the second matrix [P1,P2,...,Pn], where U1,U2,...,Un are the multiple first feature lengths , P1, P2,..., Pn are the plurality of second characteristic lengths, and there are n first characteristic lengths and n second characteristic lengths in total;

According to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] T ||-n|, calculate the first face image and the specified The image similarity value M of the face image.
The method for rendering a webpage based on machine learning according to claim 2, wherein the step of performing zooming processing on the first face image according to a preset zooming method to obtain a zoomed picture comprises:

Use a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and use the facial feature point detection model to detect the specified face image , So as to obtain multiple second facial feature points;

Generating a first minimum circumscribed rectangle of the plurality of first facial feature points, and generating a second minimum circumscribed rectangle of the plurality of second facial feature points;

The scaling process is performed on the first face image by an equal scaling method, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, thereby obtaining a scaled picture.
The method for rendering a web page based on machine learning according to claim 3, wherein the facial feature point detection model is only trained by a face image placed in a preset standard posture, and the designated face image is in a standard posture. Placing, said using a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and using the facial feature point detection model to detect the designated person The step of detecting the face image to obtain multiple second facial feature points includes:

Use a preset facial feature point detection model to detect the first face image to obtain a first number, and use the facial feature point detection model to detect the second face image to obtain a second face image. Number, wherein the first number refers to the number of facial feature points in the first face image, and the second number refers to the number of facial feature points in the designated face image;

Judging whether the absolute value of the difference between the first quantity and the second quantity is less than a preset difference threshold;

If the absolute value of the difference between the first quantity minus the second quantity is not less than the preset difference threshold, the first face image is rotated one or more times to obtain one or more Rotated images;

Use the facial feature point detection model to detect the rotated image in turn, until the absolute value of the difference between the number of detected facial feature points minus the second number is less than a preset difference threshold, and the face The last facial feature point output by the feature point detection model is recorded as the first facial feature point;

Acquire the first facial feature point, and obtain the second facial feature point generated by the facial feature point detection model to detect the second face image.
The method for rendering a webpage based on machine learning according to claim 1, wherein the image similarity between the first face image and a pre-stored specified face image is calculated according to a preset image similarity calculation method After the step of determining whether the similarity value is greater than the preset image similarity threshold, the method includes:

If the similarity value is greater than the preset image similarity threshold, the specified data packet corresponding to the specified face image is acquired according to the preset correspondence between the face image and the data packet, wherein the specified data packet Used to render web pages;

Render the web page using the specified data package.
The method for rendering a web page based on machine learning according to claim 1, wherein if the similarity value is not greater than a preset image similarity threshold, the first face image is input into a preset prediction model The calculation is performed in the linear classifier in the linear classifier to obtain the feature information output by the linear classifier. The feature information includes at least the predicted age range; wherein the prediction model is composed of the linear classifier and a preset nonlinear Before the steps of sequential connection of classifiers, include:

Acquiring pre-collected training data and test data, where the training data and test data both include a facial image and feature information corresponding to the facial image;

Adopting a stochastic gradient descent method and using the training data to train a preset initial linear classifier to obtain an intermediate linear classifier;

Use the test data to verify the intermediate linear classifier, and determine whether the intermediate linear classifier passes the verification;

If the verification of the intermediate linear classifier is passed, the intermediate linear classifier is used as the final linear classifier.
The method for rendering a webpage based on machine learning according to claim 1, wherein the method for rendering the webpage is applied to a rendering terminal, the rendering terminal and the multiple servers are connected to each other through a gateway, and the multiple sub The corresponding steps of sending information to multiple servers include:

Encrypting the multiple sub-information separately using a first encryption method previously agreed with multiple servers to obtain multiple first sub-ciphertexts;

Combining the plurality of first sub-ciphertexts into an intermediate ciphertext;

Encrypt the intermediate ciphertext by adopting the second encryption method agreed with the gateway in advance, so as to obtain the final ciphertext;

Send the final ciphertext to the gateway, and request the gateway to decrypt and split the final ciphertext to obtain a plurality of first sub-ciphertexts, and send the plurality of first sub-ciphertexts correspondingly To multiple servers.
A web page rendering device based on machine learning includes:

The first face image acquisition unit is configured to use a preset camera to collect the user's first face image;

The image similarity value judgment unit is used to calculate the image similarity value between the first face image and a pre-stored designated face image according to a preset image similarity calculation method, and determine whether the similarity value is Greater than the preset image similarity threshold;

The feature information acquiring unit is configured to, if the similarity value is not greater than the preset image similarity threshold, input the first face image into the linear classifier in the preset prediction model for calculation, so as to obtain the The feature information output by the linear classifier, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset nonlinear classifier; the linear classifier It is pre-trained through sample data including facial images and feature information corresponding to facial images;

The web page rendering information acquisition unit is configured to input the feature information into the nonlinear classifier for calculation, so as to obtain web page rendering information composed of multiple sub-information output by the nonlinear classifier, wherein the multiple sub-information Including at least color sub-information, icon style sub-information, and layout style sub-information of the webpage; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

A sub-information sending unit, configured to send the multiple sub-information and return request information of the sub-data package used for web page rendering to multiple corresponding servers;

The web page rendering unit is configured to receive the multiple sub-data packets correspondingly returned by the multiple servers, combine the multiple sub-data packets into a rendering data packet, and render the web page by using the rendering data packet.
A computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when the processor executes the computer program:

Use the preset camera to collect the user's first face image;

According to a preset image similarity calculation method, calculate an image similarity value between the first face image and a prestored designated face image, and determine whether the similarity value is greater than a preset image similarity threshold;

If the similarity value is not greater than the preset image similarity threshold, the first face image is input into the linear classifier in the preset prediction model for calculation, so as to obtain the features output by the linear classifier Information, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset non-linear classifier; the linear classifier is pre-processed by including facial images and Trained on the sample data of the feature information corresponding to the face image;

The feature information is input into the non-linear classifier for calculation, so as to obtain the web page rendering information composed of multiple sub-information output by the non-linear classifier, wherein the multiple sub-information includes at least the color sub-information of the web page, Icon style sub-information and layout style sub-information; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

Sending the multiple sub-information and returning the request information of the sub-data package used for web page rendering to multiple corresponding servers;

Receiving the multiple sub-data packets correspondingly returned by the multiple servers, combining the multiple sub-data packets into a rendering data packet, and rendering the webpage by using the rendering data packet.
The computer device according to claim 9, wherein the calculating the image similarity value between the first face image and a pre-stored specified face image according to a preset image similarity calculation method comprises:

Performing a scaling process on the first face image according to a preset scaling method to obtain a scaled picture, wherein the eyebrow and eye spacing of the scaled picture is equal to the eyebrow and eye spacing of the designated face image;

Acquire multiple first feature lengths in the zoomed picture, and acquire multiple second feature lengths in the specified face image, wherein the multiple first feature lengths and the multiple second feature lengths are both Including at least face length and face width, the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence;

Generate the first matrix [U1,U2,...,Un] and the second matrix [P1,P2,...,Pn], where U1,U2,...,Un are the multiple first feature lengths , P1, P2,..., Pn are the plurality of second characteristic lengths, and there are n first characteristic lengths and n second characteristic lengths in total;

According to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] T ||-n|, calculate the first face image and the specified The image similarity value M of the face image.
10. The computer device according to claim 10, wherein the zooming process is performed on the first face image according to a preset zooming method to obtain a zoomed picture, comprising:

Use a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and use the facial feature point detection model to detect the specified face image , So as to obtain multiple second facial feature points;

Generating a first minimum circumscribed rectangle of the plurality of first facial feature points, and generating a second minimum circumscribed rectangle of the plurality of second facial feature points;

The scaling process is performed on the first face image by an equal scaling method, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, thereby obtaining a scaled picture.
The computer device according to claim 11, wherein the facial feature point detection model is only trained by a face image placed in a preset standard posture, the designated face image is placed in a standard posture, and the use The preset facial feature point detection model detects the first face image to obtain a plurality of first facial feature points; and uses the facial feature point detection model to detect the specified face image, Thereby, multiple second facial feature points are obtained, including:

Use a preset facial feature point detection model to detect the first face image to obtain a first number, and use the facial feature point detection model to detect the second face image to obtain a second face image. Number, wherein the first number refers to the number of facial feature points in the first face image, and the second number refers to the number of facial feature points in the designated face image;

Judging whether the absolute value of the difference between the first quantity and the second quantity is less than a preset difference threshold;

If the absolute value of the difference between the first quantity minus the second quantity is not less than the preset difference threshold, the first face image is rotated one or more times to obtain one or more Rotated images;

Use the facial feature point detection model to detect the rotated image in turn, until the absolute value of the difference between the number of detected facial feature points minus the second number is less than a preset difference threshold, and the face The last facial feature point output by the feature point detection model is recorded as the first facial feature point;

Acquire the first facial feature point, and obtain the second facial feature point generated by the facial feature point detection model to detect the second face image.
8. The computer device according to claim 9, wherein the image similarity value between the first face image and a pre-stored designated face image is calculated according to a preset image similarity calculation method, and determined After describing whether the similarity value is greater than the preset image similarity threshold, it includes:

If the similarity value is greater than the preset image similarity threshold, the specified data packet corresponding to the specified face image is acquired according to the preset correspondence between the face image and the data packet, wherein the specified data packet Used to render web pages;

Render the web page using the specified data package.
9. The computer device according to claim 9, wherein if the similarity value is not greater than a preset image similarity threshold, the first face image is input to a linear classifier in a preset prediction model In order to obtain the feature information output by the linear classifier, the feature information includes at least the predicted age range; wherein the prediction model is sequentially connected by the linear classifier and the preset nonlinear classifier. Before completion, including:

Acquiring pre-collected training data and test data, where the training data and test data both include a facial image and feature information corresponding to the facial image;

Adopting a stochastic gradient descent method and using the training data to train a preset initial linear classifier to obtain an intermediate linear classifier;

Use the test data to verify the intermediate linear classifier, and determine whether the intermediate linear classifier passes the verification;

If the verification of the intermediate linear classifier is passed, the intermediate linear classifier is used as the final linear classifier.
A computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

Use the preset camera to collect the user's first face image;

According to a preset image similarity calculation method, calculate an image similarity value between the first face image and a prestored designated face image, and determine whether the similarity value is greater than a preset image similarity threshold;

If the similarity value is not greater than the preset image similarity threshold, the first face image is input into the linear classifier in the preset prediction model for calculation, so as to obtain the features output by the linear classifier Information, the feature information includes at least a predicted age range; wherein the prediction model is formed by sequentially connecting the linear classifier and a preset non-linear classifier; the linear classifier is pre-processed by including facial images and Trained on the sample data of the feature information corresponding to the face image;

The feature information is input into the non-linear classifier for calculation, so as to obtain the web page rendering information composed of multiple sub-information output by the non-linear classifier, wherein the multiple sub-information includes at least the color sub-information of the web page, Icon style sub-information and layout style sub-information; the non-linear classifier is pre-trained through sample data including feature information and multiple sub-information corresponding to the feature information;

Sending the multiple sub-information and returning the request information of the sub-data package used for web page rendering to multiple corresponding servers;

Receiving the multiple sub-data packets correspondingly returned by the multiple servers, combining the multiple sub-data packets into a rendering data packet, and rendering the webpage by using the rendering data packet.
15. The computer-readable storage medium according to claim 15, wherein said calculating the image similarity value between the first face image and a pre-stored specified face image according to a preset image similarity calculation method, include:

Performing a scaling process on the first face image according to a preset scaling method to obtain a scaled picture, wherein the eyebrow and eye spacing of the scaled picture is equal to the eyebrow and eye spacing of the designated face image;

Acquire multiple first feature lengths in the zoomed picture, and acquire multiple second feature lengths in the specified face image, wherein the multiple first feature lengths and the multiple second feature lengths are both Including at least face length and face width, the plurality of first characteristic lengths correspond to the plurality of second characteristic lengths in a one-to-one correspondence;

Generate the first matrix [U1,U2,...,Un] and the second matrix [P1,P2,...,Pn], where U1,U2,...,Un are the multiple first feature lengths , P1, P2,..., Pn are the plurality of second characteristic lengths, and there are n first characteristic lengths and n second characteristic lengths in total;

According to the formula: M=1/|||[U1,U2,...,Un] [P1,P2,...,Pn] T ||-n|, calculate the first face image and the specified The image similarity value M of the face image.
16. The computer-readable storage medium according to claim 16, wherein the zooming process is performed on the first face image according to a preset zooming method to obtain a zoomed picture, comprising:

Use a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and use the facial feature point detection model to detect the specified face image , So as to obtain multiple second facial feature points;

Generating a first minimum circumscribed rectangle of the plurality of first facial feature points, and generating a second minimum circumscribed rectangle of the plurality of second facial feature points;

The scaling process is performed on the first face image by an equal scaling method, so that the area of the first minimum circumscribed rectangle is equal to the area of the second minimum circumscribed rectangle, thereby obtaining a scaled picture.
18. The computer-readable storage medium of claim 17, wherein the facial feature point detection model is only trained by a face image placed in a preset standard posture, and the designated face image is placed in a standard posture, Said using a preset facial feature point detection model to detect the first face image to obtain a plurality of first facial feature points; and using the facial feature point detection model to detect the specified face image Perform detection to obtain multiple second facial feature points, including:

Use a preset facial feature point detection model to detect the first face image to obtain a first number, and use the facial feature point detection model to detect the second face image to obtain a second face image. Number, wherein the first number refers to the number of facial feature points in the first face image, and the second number refers to the number of facial feature points in the designated face image;

Judging whether the absolute value of the difference between the first quantity and the second quantity is less than a preset difference threshold;

If the absolute value of the difference between the first quantity minus the second quantity is not less than the preset difference threshold, the first face image is rotated one or more times to obtain one or more Rotated images;

Use the facial feature point detection model to detect the rotated image in turn, until the absolute value of the difference between the number of detected facial feature points minus the second number is less than a preset difference threshold, and the face The last facial feature point output by the feature point detection model is recorded as the first facial feature point;

Acquire the first facial feature point, and obtain the second facial feature point generated by the facial feature point detection model to detect the second face image.
15. The computer-readable storage medium according to claim 15, wherein said calculating the image similarity value between the first face image and a pre-stored specified face image according to a preset image similarity calculation method, And after judging whether the similarity value is greater than the preset image similarity threshold, the method includes:

If the similarity value is greater than the preset image similarity threshold, the specified data packet corresponding to the specified face image is acquired according to the preset correspondence between the face image and the data packet, wherein the specified data packet Used to render web pages;

Render the web page using the specified data package.
The computer-readable storage medium according to claim 15, wherein if the similarity value is not greater than a preset image similarity threshold, the first face image is input into a preset prediction model The calculation is performed in the linear classifier to obtain the feature information output by the linear classifier. The feature information includes at least a predicted age range; wherein the prediction model is composed of the linear classifier and a preset nonlinear Before the sequence is connected, it includes:

Acquiring pre-collected training data and test data, where the training data and test data both include a facial image and feature information corresponding to the facial image;

Adopting a stochastic gradient descent method and using the training data to train a preset initial linear classifier to obtain an intermediate linear classifier;

Use the test data to verify the intermediate linear classifier, and determine whether the intermediate linear classifier passes the verification;

If the verification of the intermediate linear classifier is passed, the intermediate linear classifier is used as the final linear classifier.