WO2020147445A1

WO2020147445A1 - Rephotographed image recognition method and apparatus, computer device, and computer-readable storage medium

Info

Publication number: WO2020147445A1
Application number: PCT/CN2019/122696
Authority: WO
Inventors: 徐国诚
Original assignee: 深圳壹账通智能科技有限公司
Priority date: 2019-01-16
Filing date: 2019-12-03
Publication date: 2020-07-23
Also published as: CN109886275A

Abstract

A rephotographed image recognition method, comprising: acquiring an image to be recognized; carrying out a Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum feature map; extracting an LBP feature value from the Fourier spectrum feature map; according to the extracted LBP feature value, generating LBP histogram statistic data for representing an LBP feature value statistic probability; inputting the LBP histogram statistic data into a neural network model, and carrying out recognition processing on the LBP histogram statistic data by means of the neural network model to obtain a first rephotographing probability; and when the first rephotographing probability reaches a probability threshold value, determining that the image to be recognized is a rephotographed image.

Description

Recognition method, device, computer equipment and computer readable storage medium for remake image

Cross-reference of related applications

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on January 16, 2019. The application number is 2019100400045, and the application name is "Replicated Image Recognition Method, Apparatus, Computer Equipment, and Storage Medium". The entire content is by reference Incorporated in this application.

Technical field

This application relates to a method, device, computer equipment, and computer-readable storage medium for re-photographing image recognition.

Background technique

With the continuous development of computer technology and Internet technology, opening an account online or handling important business online may require users to shoot and upload corresponding images through mobile terminals or web cameras, such as the user’s ID image or the user’s real image .

However, the inventor realized that the uploaded image may not be a real image obtained by the user by shooting himself, or a credential image obtained by shooting a real ID, but a picture displayed on the display of a computer or mobile phone. As a result, there is a problem of forging user images. If these re-photographed images are not recognized, it will cause security problems for user information.

Summary of the invention

According to various embodiments disclosed in the present application, a method, apparatus, computer equipment, and computer-readable storage medium for recognizing a remake image are provided.

A method for recognizing a remake image, which is executed by a server, and the method includes:

Obtain the image to be recognized;

Perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

Extracting a local binary mode LBP feature value from the Fourier spectrum feature map;

Generate LBP histogram statistical data to represent the statistical probability of LBP feature values according to the extracted LBP feature values;

Input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability; and

When the first copying probability reaches the probability threshold, it is determined that the image to be recognized is a copying image.

A re-photographed image recognition device, the device includes:

Image acquisition module for acquiring the image to be recognized;

A transform module, which is used to perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

The feature value extraction module is configured to extract the feature value of the local binary mode LBP from the Fourier spectrum feature map;

A generating module for generating LBP histogram statistical data representing the statistical probability of the LBP feature value according to the extracted LBP feature value;

A processing module, configured to input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain the first remake probability; and

The remake image determination module is configured to determine that the to-be-identified image is a remake image when the first remake probability reaches a probability threshold.

A computer device includes a memory and a processor, the memory stores computer-readable instructions, and the processor implements the following steps when executing the computer-readable instructions:

Obtain the image to be recognized;

Inputting the LBP histogram statistical data into a neural network model, and performing identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability;

A non-volatile computer-readable storage medium having computer-readable instructions stored thereon, and when the computer-readable instructions are executed by a processor, the following steps are implemented:

Obtain the image to be recognized;

The details of one or more embodiments of the application are set forth in the drawings and description below. Other features and advantages of this application will become apparent from the description, drawings, and claims.

BRIEF DESCRIPTION

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings that need to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. A person of ordinary skill in the art can obtain other drawings based on these drawings without creative work.

Fig. 1 is an application scene diagram of the method for recognizing a re-photographed image according to one or more embodiments.

Fig. 2 is a schematic flowchart of a method for recognizing a re-photographed image according to one or more embodiments.

Fig. 3 is a schematic diagram of extracting LBP features according to one or more embodiments.

Fig. 4 is a schematic flowchart of the steps of training a neural network model according to another or more embodiments.

Fig. 5 is a structural block diagram of an image recognizing device according to one or more embodiments.

Fig. 6 is a structural block diagram of an image recognizing device according to another or more embodiments.

Fig. 7 is an internal structure diagram of a computer device according to one or more embodiments.

detailed description

In order to make the purpose, technical solutions and advantages of the present application more clear, the following describes the present application in further 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.

The image recognition method provided by this application can be applied in the application environment as shown in FIG. 1. Wherein, the terminal 102 communicates with the server 104 through the network through the network. The server 104 obtains the image to be recognized through the terminal 102, and performs Fourier transform on the obtained image to be recognized to obtain a Fourier spectrum feature map; the server 104 extracts LBP (Local Binary Pattern, local binary pattern) features from the Fourier spectrum feature map Calculate the statistical probability of the LBP feature value within the range of each feature value, and input the statistical probability into the neural network model. The neural network model can determine whether the image to be recognized is a remake image. The terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server 104 may be implemented by an independent server or a server cluster composed of multiple servers.

In one of the embodiments, as shown in FIG. 2, a method for recognizing a remake image is provided. Taking the method applied to the server in FIG. 1 as an example for description, the method includes the following steps:

S202: Acquire an image to be recognized.

The image to be recognized may be a 32-bit color image including RGB channels, in addition, it may also be a 64-bit or 128-bit color image including RGB channels.

In one of the embodiments, the server captures the image to be recognized through a camera installed on the terminal; or, the server obtains the image to be recognized from a local album stored in the terminal through the terminal.

In one of the embodiments, if the method for recognizing the re-photographed image is applied to a user registered account (such as a registered financial account), there may be two functional options in the operation page for displaying uploaded images on the terminal. One is to support real-time shooting by the user The function options of the image, and the other is the function option that supports the user to select the image from the album. If the user selects the function option of real-time image capture, the terminal calls the camera to capture the image of the user or user ID, and uses the captured image as the image to be recognized. If the user selects the function option of obtaining images from the album, the terminal obtains the image of the user or the user certificate from the local album, and uses the obtained image as the image to be recognized.

For example, when a user opens a credit card account or a securities account online, the account opening system will display an operation page for uploading the user's credential image and the user's real image. When uploading the real image of the user or the user ID photo, the camera of the terminal can be called for the user to take the image, or the corresponding image to be recognized can be selected from the album through the function options provided on the operation page, or the camera of the terminal can be called So that the user can take images.

S204: Perform Fourier transform on the acquired image to be identified to obtain a Fourier spectrum characteristic map.

The Fourier spectral characteristic map obtained after Fourier transform may be a single-channel 8-bit, 16-bit, or 32-bit grayscale image, and the spectral characteristic map reflects the aggregation of spectral energy points. The Fourier transform may be a discrete Fourier transform.

In one of the embodiments, S204 may specifically include: determining the size of the image to be recognized; when the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, perform the image recognition according to the preset standard size. Scaling; Decompose the scaled image to be recognized into three images of RGB channels; perform Fourier transform on the three images respectively, and synthesize the transformed image; convert the synthetic image into a single channel Fourier spectrum characteristic map.

For example, before performing Fourier transform on the acquired image, the terminal first determines the size of the image to be recognized, such as the image resolution or size. If the size of the image to be recognized is too large or too small, the image is enlarged or compressed to a preset standard size. The terminal decomposes the scaled image into three corresponding images according to the RGB three channels, performs Fourier transform on each image, and then synthesizes the changed images to obtain a combined feature map. The terminal performs gray scale conversion on this combined feature map, so that a single-channel Fourier spectrum feature map can be obtained. In the foregoing embodiment, compressing an image that is too large can reduce the amount of calculation in the recognition process.

S206: Extract the local binary mode LBP feature value from the Fourier spectrum feature map.

In one of the embodiments, the server evenly divides the Fourier spectrum feature map into a plurality of sub Fourier spectrum feature blocks, and extracts the LBP feature value at the pixel point of each sub Fourier spectrum feature block.

Specifically, the server divides the Fourier spectrum feature map into multiple sub-Fourier spectrum feature blocks, and divides each sub Fourier spectrum feature block into multiple pixel blocks. In each pixel block, the server judges whether the gray value of the non-central pixel is greater than the gray value of the central pixel; if it is, the gray value of the non-central pixel is set to the first value; if not, it is The gray value of the central pixel is set to the second value. The server performs a weighted summation on the gray values of the non-central pixel points in each pixel block after the gray value is set, and uses the result of the weighted summation as the LBP feature value of each pixel block.

For example, as shown in Figure 3, for a certain pixel in the sub-Fourier spectrum feature block, the server takes the pixel as the center pixel, and the gray value of the center pixel is compared with each field in the 3×3 window. The gray values of the pixels are compared. If the gray value of the field pixel is greater than or equal to the gray value of the center pixel, the gray value of the field pixel is set to 1; if the gray value of the field pixel is less than For the gray value of the central pixel, the gray value of the domain pixel is set to 0. Therefore, the gray value of the central pixel is calculated as 01111100 in binary and 124 in decimal. Through the above method, the LBP feature value of each sub-Fourier spectrum feature block can be obtained.

S208: Generate LBP histogram statistical data representing the statistical probability of the LBP feature value according to the extracted LBP feature value.

In one of the embodiments, the server may sort the LBP feature values in each sub-Fourier spectral feature block in order of size; sort the LBP feature values in each sub-Fourier spectral feature block after sorting according to a preset step size Divide evenly into multiple characteristic value ranges; calculate the statistical probability of LBP characteristic values belonging to each characteristic value range.

In one of the embodiments, in each sub Fourier spectrum feature block, an LBP histogram is established according to the statistical probability of the LBP feature value in each feature value range, and the probability corresponding to each LBP histogram is stored in a different sub-array. In A1, A2...An, the statistical probability of the LBP feature value of each sub-Fourier spectrum feature block within each preset feature value range is obtained, and the statistical probability may be LBP histogram statistical data. Before inputting the neural network model, the server saves the sub-arrays A1, A2...An in another array S.

S210: Input the statistical data of the LBP histogram into the neural network model, and perform identification processing on the statistical data of the LBP histogram through the neural network model to obtain the first remake probability.

In one of the embodiments, the server inputs the LBP histogram statistical data into the trained neural network model in the form of an array.

S212: When the first reproduction probability reaches the probability threshold, it is determined that the image to be recognized is a reproduction image.

For example, assuming that the probability threshold is 90%, when the first remake probability is greater than or equal to 90%, the image to be recognized is a remake image. When the first copy probability is less than 90%, the image to be recognized is not a copy image.

In one of the embodiments, in order to further ensure the accuracy of the recognition result, after S212, the method further includes: the server generates an image review request carrying the image to be recognized; and sends the image review request to the reviewer account, so that the reviewer can The image to be recognized is reviewed and the review result is fed back; the review result for the image review request that is fed back by the reviewer’s account is received; the review result carries the second probability that the image to be recognized is a remake image; obtains the first image corresponding to the machine recognition A weight and a second weight corresponding to the recheck identification; according to the first weight and the second weight, the first and second remake probability are weighted and summed to obtain the weighted sum of the remake probability; when the weighted sum is greater than or equal to When the weighted sum is preset, it is finally determined that the image to be recognized is a remake image.

The first weight and the second weight are two different values, and the sum of the first weight and the second weight is 1.

For example, in order to further ensure the accuracy of the recognition results, when the machine recognizes the image to be recognized as a remake, it will also send the image to be recognized to a professional, who will further review and recognize the image to be recognized, and then the feedback will contain the recognition result The result of the review, the recognition result may be the probability that the image to be recognized is a remake image. If the probability that the image to be recognized is a re-photographed image is high (for example, greater than 90%), it is finally determined that the image to be recognized is a re-photographed image. If the probability that the image to be recognized is a remake image is small (such as less than 90%), the two probabilities are weighted and summed to obtain the weighted sum of the remake probability, for example, the weighted sum is p=k ₁ ×p ₁ +k ₂ ×p ₂ (k ₁ and k ₂ are the first weight and the second weight, respectively, and p ₁ and p ₂ are the first and second remake probability, respectively), according to the weighted sum to determine whether the image to be recognized is a remake image. When the weighted sum p is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is a remake image. When the weighted sum p is less than the preset weighted sum, it is finally determined that the image to be recognized is not a remake image.

In one of the embodiments, the server may determine whether the image to be recognized is a remake image based on the result of the review and recognition. When the first reproduction probability reaches the probability threshold, the server may also not need to perform the process of rechecking and identifying, and directly determine that the image to be recognized is a reproduced image.

In one of the embodiments, this method can be applied to financial account opening. When it is finally determined that the image to be recognized is a rephotographed image, the server rejects the user’s account opening request; when it is determined that the image to be recognized is a non-replicated image, the server further reviews other materials submitted by the user, and if they meet the account opening requirements, the user’s request is allowed. Account opening request.

As an example, (1) first divide the Fourier spectral feature map into multiple sub-Fourier spectral feature blocks, and then divide the sub-Fourier spectral feature block into multiple pixel blocks; (2) in each pixel block, The gray value of the central pixel is compared with the gray value of the adjacent 8 pixels. If the gray value of any one of the 8 pixels is greater than the gray value of the central pixel, the corresponding pixel The gray value of the point is set to 1, otherwise it is set to 0. In this way, 8 pixels in the neighborhood of a 3×3 window can generate 8-bit binary numbers, that is, the LBP feature value of the center pixel of the window is obtained; (3) Then calculate the statistical histogram of each pixel block, the statistical histogram It is used to indicate the frequency of occurrence of the LBP feature value within each preset feature value range, that is, the statistical probability; wherein, the histogram can also be normalized. (4) Finally, according to the obtained statistical histogram of each pixel block, the LBP texture feature vector of the entire image to be recognized is obtained, and the LBP texture feature vector is input to the neural network model. Through the processing of the neural network, it can be obtained whether the image to be recognized is To remake an image.

In the foregoing embodiment, a Fourier spectrum feature map is generated according to the acquired image to be recognized, and the LBP feature value is extracted from the Fourier spectrum feature map, so as to obtain the local texture feature of the image to be recognized. Generate LBP histogram statistical data representing the statistical probability of LBP feature values, and input the LBP histogram statistical data into the neural network model. The neural network model can determine whether the image to be recognized is a remake image, thereby avoiding the image to be recognized as a remake computer Or the pictures displayed on the display screens of devices such as mobile phones, so as to ensure the authenticity of the images and avoid the hidden dangers of user information caused by third-party use of re-photographed images. In addition, through the above-mentioned embodiments, the accuracy of re-photographed image recognition can be effectively improved, and the accuracy is increased from the original 80% to 93%.

In one of the embodiments, as shown in FIG. 4, the method further includes:

S402: Obtain a sample image to be identified, label the sample image to be identified, and obtain a sample image to be identified including a label; the label is used to indicate whether the sample image to be identified is a remake image.

The sample image to be identified may be a 32-bit color image including RGB channels, in addition, it may also be a 64-bit or 128-bit color image including RGB channels. The size of the sample image to be identified is the preset standard size.

S404: Perform Fourier transform on the sample image to be identified containing the label to obtain a sample image of Fourier spectrum characteristics.

In one of the embodiments, before S404, the method may further include: the server randomly rotates the sample image to be identified containing the label from 0 to 360 degrees, and processes such as random zooming, adjusting brightness, chroma, and clarity, etc. Then, according to the processed sample image to be identified, the corresponding Fourier spectrum feature sample image is generated to increase the generalization ability of the neural network model, thereby improving the accuracy of remake recognition.

S406: Extract LBP feature sample values from the Fourier spectrum feature sample map.

In one of the embodiments, the Fourier spectrum feature sample map is evenly divided into a plurality of sub-Fourier spectrum feature sample blocks, and the LBP feature sample value is extracted from the pixels of each sub-Fourier spectrum feature sample block.

Specifically, the Fourier spectrum feature sample image is equally divided into a plurality of sub-Fourier spectrum feature sample blocks; each sub-Fourier spectrum feature sample block is equally divided into a plurality of sample pixel blocks; in each sample pixel Block, judge whether the gray value of the non-central pixel is greater than the gray value of the central pixel; if so, set the gray value of the non-central pixel to the first value; if not, set the gray value of the non-central pixel The gray value is set to the second value; the gray value of the non-central pixel in each sample pixel block after the gray value is set is weighted and summed; the result of the weighted sum is used as the LBP feature sample value of each sample pixel block .

For example, as shown in Figure 3, for a certain pixel in the sub-Fourier spectrum feature sample block, take the pixel as the center pixel, and the gray value of the center pixel is respectively compared with that in the 3×3 window. The gray values of pixels in each area are compared. If the gray value of the area pixel is greater than or equal to the gray value of the center pixel, the gray value of the area pixel is set to 1; if the gray value of the area pixel is If the value is less than the gray value of the central pixel, the gray value of the domain pixel is set to 0. Therefore, the gray value of the central pixel is calculated as 01111100 in binary and converted to decimal as 124, so each sub The LBP feature sample value of the Fourier spectrum feature sample block.

S408: Generate an LBP histogram statistical data sample for representing the statistical probability of the LBP feature value according to the extracted LBP feature value.

In one of the embodiments, S208 may specifically include: the server may sort the LBP feature sample values in each sub-Fourier spectrum feature sample block in order of size; and sort the sorted sub-Fourier spectrum feature sample graphs The LBP feature sample value in the block is evenly divided into multiple feature value ranges according to the preset step length; the statistical probability of the LBP feature sample value belonging to each feature value range is calculated.

In one of the embodiments, in each sub-Fourier spectrum feature sample block, the LBP histogram is established according to the statistical probability of the LBP feature value in each feature value range, and the probability corresponding to each LBP histogram is stored in a different In the sub-arrays A1, A2...An of, the statistical probability of the LBP feature sample value of each sub-Fourier spectrum feature sample block within each preset feature value range is obtained. Before inputting the neural network model, the server saves the sub-arrays A1, A2...An in a large array S.

S410: Input the LBP histogram statistical data sample into the neural network model, and perform identification processing on the LBP histogram statistical data sample through the neural network model to obtain the third remake probability.

In one of the embodiments, the server inputs the LBP histogram statistical data samples into the trained neural network model in the form of an array.

S412: Compare the difference between the third remake probability and the label, and adjust the parameters of the neural network model.

In one of the embodiments, S412 includes: determining the error between the third remake probability and the label; propagating the error back to the network layer of the neural network model to obtain the gradient of each network layer parameter; adjusting the neural network according to the obtained gradient The parameters of each network layer in the network model.

The server can determine the error between the third remake probability and the label through the loss function. The loss function can be any of the following: Mean Squared Error (Mean Squared Error), cross-entropy loss function, L2Loss function, and Focal Loss function.

It should be understood that, although the steps in the flowcharts of FIGS. 2 and 4 are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless clearly stated in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least part of the steps in Figures 2 and 4 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. These sub-steps or stages The execution order of is not necessarily performed sequentially, but may be performed alternately or alternately with at least a part of other steps or sub-steps or stages of other steps.

In one of the embodiments, as shown in FIG. 5, a remake image recognition device is provided, including: image acquisition module 502, transformation module 504, feature value extraction module 506, generation module 508, processing module 510, and remake image determination Module 512, of which:

The image acquisition module 502 is used to acquire the image to be recognized;

The transform module 504 is configured to perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

The feature value extraction module 506 is used to extract the feature value of the local binary mode LBP from the Fourier spectrum feature map;

The generating module 508 is configured to generate LBP histogram statistical data representing the statistical probability of the LBP feature value according to the extracted LBP feature value;

The processing module 510 is configured to input LBP histogram statistical data into a neural network model, and perform recognition processing on the LBP histogram statistical data through the neural network model to obtain the first remake probability;

The remake image determination module 512 is configured to determine that the image to be recognized is a remake image when the first remake probability reaches the probability threshold.

In one of the embodiments, the transformation module 504 is further configured to: determine the size of the image to be recognized; when the size of the image to be recognized is larger than the first preset size, or smaller than the second preset size, it is to be recognized according to the preset standard size The image is scaled; the scaled image to be recognized is decomposed into three images of RGB channel; the three images are respectively Fourier transformed, and the transformed image is synthesized; the synthesized image is converted into a single Fourier spectrum characteristic map of the channel.

In one of the embodiments, the feature value extraction module 506 is further configured to: divide the Fourier spectrum feature map into a plurality of sub-Fourier spectrum feature blocks; respectively divide each sub-Fourier spectrum feature block into a plurality of pixel blocks; In each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel; if so, set the gray value of the non-central pixel to the first value; if not, set the non-central pixel The gray value of the dot is set to the second value; the gray value of the non-central pixel point in each pixel block after the gray value is set is weighted and summed; the result of the weighted sum is used as the LBP feature value of each pixel block.

In one of the embodiments, as shown in FIG. 6, the device further includes: a request generating module 514, a sending module 516, a receiving module 518, a weight obtaining module 520, and a weighting module 522, where:

The request generation module 514 is configured to generate an image review request carrying the image to be recognized;

The sending module 516 is used to send an image review request;

The receiving module 518 is configured to receive a review result in response to an image review request; the review result carries a second remake probability that the image to be identified is a remake image;

The weight obtaining module 520 is configured to obtain the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition;

The weighting module 522 performs a weighted summation of the first remake probability and the second remake probability according to the first weight and the second weight respectively to obtain the weighted sum of the remake probability;

The re-photographed image determining module 512 is further configured to finally determine that the image to be recognized is a re-photographed image when the weighted sum is greater than or equal to the preset weighted sum.

In one of the embodiments, as shown in FIG. 6, the device further includes: an adjustment module 524, wherein:

The image acquisition module 502 is also used for the image acquisition module to acquire a sample image to be identified, annotate the sample image to be identified, and obtain a sample image to be identified that includes a label; the label is used to indicate whether the sample image to be identified is a remake image;

The transform module 504 is further configured to perform Fourier transform on the sample image to be identified containing the label to obtain a sample image of Fourier spectrum characteristics;

The feature value extraction module 504 block is also used to extract LBP feature sample values from the Fourier spectrum feature sample map;

The generating module 508 is further configured to generate an LBP histogram statistical data sample for representing the statistical probability of the LBP feature value according to the extracted LBP feature value;

The processing module 510 is further configured to input LBP histogram statistical data samples into the neural network model, and perform identification processing on the LBP histogram statistical data samples through the neural network model to obtain the third remake probability;

The adjustment module 524 is used to compare the difference between the third remake probability and the label, and adjust the parameters of the neural network model.

In one of the embodiments, the adjustment module 524 is further used to: determine the error between the third remake probability and the label; backpropagate the error to the network layer of the neural network model to obtain the gradient of each network layer parameter; The gradient of adjusts the parameters of each network layer in the neural network model.

For the specific limitation of the re-photographed image recognition device, please refer to the above-mentioned limitation on the re-photographed image recognition method, which will not be repeated here. Each module in the above-mentioned re-photographed image recognition device can be implemented in whole or in part by software, hardware, and a combination thereof. The above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

In one of the embodiments, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. 7. The computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor of the computer device is used to provide computing 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, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The database of the computer equipment is used to store data such as images to be identified and sample images to be identified. The network interface of the computer device is used to communicate with external terminals through a network connection. The computer-readable instruction is executed by the processor to realize a re-photographed image recognition method.

Those skilled in the art can understand that the structure shown in FIG. 7 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may Including more or less parts than shown in the figure, or combining some parts, or having a different part arrangement.

In one of the embodiments, a computer device is provided, including a memory and a processor, the memory stores computer-readable instructions, and the processor implements the following steps when executing the computer-readable instructions: acquiring an image to be recognized; Perform Fourier transform on the image to be recognized to obtain the Fourier spectrum feature map; extract the local binary mode LBP feature value from the Fourier spectrum feature map; generate the LBP histogram representing the statistical probability of the LBP feature value according to the extracted LBP feature value Figure statistical data; input the LBP histogram statistical data into the neural network model, and use the neural network model to identify the LBP histogram statistical data to obtain the first remake probability; when the first remake probability reaches the probability threshold, it is determined that the image to be recognized is Retake an image.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: determining the size of the image to be recognized; when the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, according to The image to be recognized is scaled by the preset standard size; the zoomed image to be recognized is decomposed into three images of RGB channels; the three images are respectively Fourier transformed, and the transformed images are synthesized; the synthesized processing The resulting image is converted into a single-channel Fourier spectrum characteristic map.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: divide the Fourier spectrum feature map into a plurality of sub Fourier spectrum feature blocks; each sub-Fourier spectrum feature block is equally divided into multiple Pixel block; in each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel; if yes, set the gray value of the non-central pixel to the first value; if not, then Set the gray value of the non-central pixel to the second value; perform a weighted summation on the gray value of the non-central pixel in each pixel block after the gray value is set; use the result of the weighted sum as the pixel block LBP characteristic value.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: generating and sending an image review request carrying the image to be identified; receiving a review result in response to the image review request; the review result carrying the image to be identified is a remake The second remake probability of the image; obtain the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition; perform a weighted summation of the first remake probability and the second remake probability according to the first weight and the second weight respectively , To obtain the weighted sum of the reproduction probability; when the weighted sum is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is the reproduced image.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: obtain the sample image to be identified, mark the sample image to be identified, and obtain the sample image to be identified containing the label; the label is used to indicate the sample image to be identified Whether it is a remake of the image; Fourier transform the sample image to be identified containing the label to obtain the Fourier spectrum feature sample image; extract the LBP feature sample value from the Fourier spectrum feature sample image; according to the extracted LBP feature value Generate LBP histogram statistical data samples that represent the statistical probability of LBP feature values; input LBP histogram statistical data samples into the neural network model, and use the neural network model to identify the LBP histogram statistical data samples to obtain the third remake probability; Compare the difference between the third remake probability and the label, and adjust the parameters of the neural network model.

In one of the embodiments, the processor further implements the following steps when executing the computer-readable instructions: determining the error between the third remake probability and the label; propagating the error back to the network layer of the neural network model to obtain the parameters of each network layer The gradient of; adjust the parameters of each network layer in the neural network model according to the obtained gradient.

In one of the embodiments, a computer-readable storage medium is provided, and computer-readable instructions are stored thereon. When the computer-readable instructions are executed by a processor, the following steps are implemented: acquiring an image to be recognized; The image undergoes Fourier transform to obtain the Fourier spectrum feature map; extract the local binary mode LBP feature value from the Fourier spectrum feature map; generate the LBP histogram statistical data representing the statistical probability of the LBP feature value according to the extracted LBP feature value ; Input the LBP histogram statistical data into the neural network model, and identify the LBP histogram statistical data through the neural network model to obtain the first remake probability; when the first remake probability reaches the probability threshold, the image to be identified is determined to be the remake image.

In one of the embodiments, when the computer-readable instructions are executed by the processor, the following steps are further implemented: determining the size of the image to be recognized; when the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, Scale the image to be recognized according to the preset standard size; decompose the scaled image to be recognized into three images with RGB channels; perform Fourier transform on the three images respectively, and synthesize the transformed images; The processed image is converted into a single-channel Fourier spectrum characteristic map.

In one of the embodiments, when the computer-readable instructions are executed by the processor, the following steps are further implemented: divide the Fourier spectrum feature map into multiple sub Fourier spectrum feature blocks; and divide each sub Fourier spectrum feature block equally into multiple blocks. Pixel blocks; in each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel; if yes, set the gray value of the non-central pixel to the first value; if not, Then the gray value of the non-central pixel is set to the second value; the gray value of the non-central pixel in each pixel block after the gray value is set is weighted and summed; the result of the weighted sum is used as each pixel block The characteristic value of LBP.

In one of the embodiments, when the computer-readable instructions are executed by the processor, the following steps are also implemented: generating and sending an image review request carrying the image to be identified; receiving a review result in response to the image review request; the review result carrying the image to be identified is The second remake probability of the remake image; obtain the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition; weight the first remake probability and the second remake probability according to the first weight and the second weight respectively Sum, to obtain the weighted sum of the probabilities of copying; when the weighted sum is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is the copying image.

In one of the embodiments, when the computer-readable instructions are executed by the processor, the following steps are also implemented: obtaining a sample image to be identified, marking the sample image to be identified, and obtaining a sample image to be identified containing a label; the label is used to indicate the sample to be identified Whether the image is a remake; the Fourier transform is performed on the sample image to be identified containing the label to obtain the Fourier spectrum feature sample image; the LBP feature sample value is extracted from the Fourier spectrum feature sample image; according to the extracted LBP feature Value generation is used to represent the LBP histogram statistical data sample of the statistical probability of the LBP feature value; the LBP histogram statistical data sample is input into the neural network model, and the LBP histogram statistical data sample is identified through the neural network model to obtain the third remake probability ; Compare the difference between the third remake probability and the label, and adjust the parameters of the neural network model.

In one of the embodiments, when the computer-readable instruction is executed by the processor, the following steps are also implemented: determine the error between the third remake probability and the label; back-propagate the error to the network layer of the neural network model to obtain each network layer The gradient of the parameter; adjust the parameters of each network layer in the neural network model according to the obtained gradient.

A person of ordinary skill in the art may understand that all or part of the process in the method of the foregoing embodiments may be completed by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions may be stored in a non-volatile computer In a readable storage medium, when the computer-readable instructions are executed, they may include the processes of the foregoing method embodiments. Wherein, any reference to the memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combinations of these technical features, they should be It is considered as the range described in this specification.

The above-mentioned embodiments only express several implementation manners of the present application, and their descriptions are more specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can also be made, which all fall within the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

Claims

A method for recognizing a remake image, which is executed by a server, and the method includes:

Obtain the image to be recognized;

Perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

Extracting a local binary mode LBP feature value from the Fourier spectrum feature map;

Generate LBP histogram statistical data to represent the statistical probability of LBP feature values according to the extracted LBP feature values;

Input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability; and

When the first copying probability reaches the probability threshold, it is determined that the image to be recognized is a copying image.
The method according to claim 1, wherein the performing Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map comprises:

Determining the size of the image to be recognized;

When the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, scaling the image to be recognized according to the preset standard size;

Decompose the zoomed image to be recognized into three images with RGB channels;

Perform Fourier transform on the three images respectively, and perform synthesis processing on the images obtained after the transformation; and

The image obtained by the synthesis process is converted into a single-channel Fourier spectrum characteristic map.
The method according to claim 1, wherein the extracting the local binary mode LBP feature value from the Fourier spectrum feature map comprises:

Divide the Fourier spectrum feature map into multiple sub Fourier spectrum feature blocks;

Dividing each of the sub-Fourier spectrum characteristic blocks into a plurality of pixel blocks;

In each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel;

If yes, set the gray value of the non-central pixel to the first value; if not, set the gray value of the non-central pixel to the second value;

Perform weighted summation on the gray values of the non-central pixels in each pixel block after the gray value is set; and

The result of the weighted summation is used as the LBP feature value of each pixel block.
The method according to any one of claims 1 to 3, wherein after the determining that the image to be recognized is a remake image, the method further comprises:

Generating and sending an image review request carrying the image to be recognized;

Receiving a review result in response to the image review request; the review result carries a second copy probability that the image to be identified is a copy image;

Acquiring the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition;

Performing a weighted summation of the first remake probability and the second remake probability according to the first weight and the second weight respectively to obtain the weighted sum of the remake probability; and

When the weighted sum is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is a remake image.
The method according to any one of claims 1 to 3, wherein the method further comprises:

Acquiring a sample image to be identified, marking the sample image to be identified, and obtaining a sample image to be identified including a label; the label is used to indicate whether the sample image to be identified is a remake image;

Performing Fourier transform on the sample image to be identified containing the label to obtain a sample image of Fourier spectrum characteristics;

Extracting LBP feature sample values from the Fourier spectrum feature sample map;

Generate LBP histogram statistical data samples used to represent the statistical probability of LBP feature values according to the extracted LBP feature values;

Input the LBP histogram statistical data sample into a neural network model, and perform identification processing on the LBP histogram statistical data sample through the neural network model to obtain a third remake probability; and

The difference between the third remake probability and the label is compared, and the parameters of the neural network model are adjusted.
The method according to claim 5, wherein the comparing the difference between the third remake probability and the label, and adjusting the parameters of the neural network model comprises:

Determine the error between the third remake probability and the tag;

Backpropagating the error to the network layer of the neural network model to obtain the gradient of each network layer parameter; and

The parameters of each network layer in the neural network model are adjusted according to the obtained gradient.
The method according to claim 1, wherein the extracting the local binary mode LBP feature value from the Fourier spectrum feature map comprises:

Evenly divide the Fourier spectrum feature map into multiple sub-Fourier spectrum feature blocks; and

Extract the local binary mode LBP feature value at each pixel of the sub-Fourier spectrum feature block.
The method according to claim 1, wherein the method further comprises:

Sort the LBP feature values in each of the sub-Fourier spectrum feature blocks in order of size;

Dividing the LBP feature values in each of the sub-Fourier spectrum feature blocks after sorting into a plurality of feature value ranges evenly according to a preset step; and

Calculate the statistical probability of the LBP feature value belonging to each of the feature value ranges.
A re-photographed image recognition device, characterized in that the device includes:

Image acquisition module for acquiring the image to be recognized;

A transform module, which is used to perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

The feature value extraction module is configured to extract the feature value of the local binary mode LBP from the Fourier spectrum feature map;

A generating module for generating LBP histogram statistical data representing the statistical probability of the LBP feature value according to the extracted LBP feature value;

A processing module, configured to input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability;

The remake image determination module is configured to determine that the to-be-identified image is a remake image when the first remake probability reaches a probability threshold.
The device according to claim 9, wherein the transformation module is further configured to:

Determining the size of the image to be recognized;

When the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, scaling the image to be recognized according to the preset standard size;

Decompose the zoomed image to be recognized into three images with RGB channels;

Perform Fourier transform on the three images respectively, and perform synthesis processing on the images obtained after the transformation;

The image obtained by the synthesis process is converted into a single-channel Fourier spectrum characteristic map.
A computer device including a memory and a processor, the memory storing computer readable instructions,

When the computer-readable instructions are executed by the processor, the one or more processors execute the following steps:

Obtain the image to be recognized;

Perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

Extracting a local binary mode LBP feature value from the Fourier spectrum feature map;

Generate LBP histogram statistical data to represent the statistical probability of LBP feature values according to the extracted LBP feature values;

Input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability; and

When the first copying probability reaches the probability threshold, it is determined that the image to be recognized is a copying image.
The computer device according to claim 11, wherein the processor further executes the following steps when executing the computer-readable instructions:

Determining the size of the image to be recognized;

When the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, scaling the image to be recognized according to the preset standard size;

Decompose the zoomed image to be recognized into three images with RGB channels;

Perform Fourier transform on the three images respectively, and perform synthesis processing on the images obtained after the transformation; and

The image obtained by the synthesis process is converted into a single-channel Fourier spectrum characteristic map.
The computer device according to claim 11, wherein the processor further executes the following steps when executing the computer-readable instructions:

Divide the Fourier spectrum feature map into multiple sub Fourier spectrum feature blocks;

Dividing each of the sub-Fourier spectrum characteristic blocks into a plurality of pixel blocks;

In each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel;

If yes, set the gray value of the non-central pixel to the first value; if not, set the gray value of the non-central pixel to the second value;

Perform weighted summation on the gray values of the non-central pixels in each pixel block after the gray value is set; and

The result of the weighted summation is used as the LBP feature value of each pixel block.
The computer device according to any one of claims 11 to 13, wherein the processor further executes the following steps when executing the computer-readable instruction:

Generating and sending an image review request carrying the image to be recognized;

Receiving a review result in response to the image review request; the review result carries a second copy probability that the image to be identified is a copy image;

Acquiring the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition;

Performing a weighted summation of the first remake probability and the second remake probability according to the first weight and the second weight respectively to obtain the weighted sum of the remake probability; and

When the weighted sum is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is a remake image.
The computer device according to claim 11, wherein the processor further executes the following steps when executing the computer-readable instructions:

Sort the LBP feature values in each of the sub-Fourier spectrum feature blocks in order of size;

Dividing the LBP feature values in each of the sub-Fourier spectrum feature blocks after sorting into a plurality of feature value ranges evenly according to a preset step; and

Calculate the statistical probability of the LBP feature value belonging to each of the feature value ranges.
A non-volatile computer-readable storage medium having computer-readable instructions stored thereon. When the computer-readable instructions are executed by a processor, the one or more processors execute the following steps:

Obtain the image to be recognized;

Perform Fourier transform on the acquired image to be recognized to obtain a Fourier spectrum characteristic map;

Extracting a local binary mode LBP feature value from the Fourier spectrum feature map;

Generate LBP histogram statistical data to represent the statistical probability of LBP feature values according to the extracted LBP feature values;

Input the LBP histogram statistical data into a neural network model, and perform identification processing on the LBP histogram statistical data through the neural network model to obtain a first remake probability; and

When the first copying probability reaches the probability threshold, it is determined that the image to be recognized is a copying image.
The storage medium according to claim 16, wherein the following steps are further executed when the computer-readable instructions are executed by the processor:

Determining the size of the image to be recognized;

When the size of the image to be recognized is larger than the first preset size or smaller than the second preset size, scaling the image to be recognized according to the preset standard size;

Decompose the zoomed image to be recognized into three images with RGB channels;

Perform Fourier transform on the three images respectively, and perform synthesis processing on the images obtained after the transformation; and

The image obtained by the synthesis process is converted into a single-channel Fourier spectrum characteristic map.
The storage medium according to claim 16, wherein the following steps are further executed when the computer-readable instructions are executed by the processor:

Divide the Fourier spectrum feature map into multiple sub Fourier spectrum feature blocks;

Dividing each of the sub-Fourier spectrum characteristic blocks into a plurality of pixel blocks;

In each pixel block, determine whether the gray value of the non-central pixel is greater than the gray value of the central pixel;

If yes, set the gray value of the non-central pixel to the first value; if not, set the gray value of the non-central pixel to the second value;

Perform weighted summation on the gray values of the non-central pixels in each pixel block after the gray value is set; and

The result of the weighted summation is used as the LBP feature value of each pixel block.
The storage medium according to any one of claims 16 to 18, wherein the following steps are further executed when the computer-readable instructions are executed by the processor:

Generating and sending an image review request carrying the image to be recognized;

Receiving a review result in response to the image review request; the review result carries a second copy probability that the image to be identified is a copy image;

Acquiring the first weight corresponding to the machine recognition and the second weight corresponding to the review recognition;

Performing a weighted summation of the first remake probability and the second remake probability according to the first weight and the second weight respectively to obtain the weighted sum of the remake probability; and

When the weighted sum is greater than or equal to the preset weighted sum, it is finally determined that the image to be recognized is a remake image.
The storage medium according to claim 16, wherein the following steps are further executed when the computer-readable instructions are executed by the processor:

Sort the LBP feature values in each of the sub-Fourier spectrum feature blocks in order of size;

Dividing the LBP feature values in each of the sub-Fourier spectrum feature blocks after sorting into a plurality of feature value ranges evenly according to a preset step; and

Calculate the statistical probability of the LBP feature value belonging to each of the feature value ranges.