A picture flipping identification method, system and device and a storage medium

Abstract

The application discloses a picture retouching recognition method, system and device and a storage medium. The method comprises the following steps: acquiring a first picture sample, determining three-dimensional posture information of the first picture sample, and labeling the first picture sample according to the three-dimensional posture information to obtain a training sample set; inputting the training sample set into a pre-constructed multi-branch convolutional neural network for training to obtain a trained picture retouching recognition model; acquiring a picture to be recognized, inputting the picture to be recognized into the picture retouching recognition model, and determining whether the picture to be recognized is a retouched picture according to a recognition result; wherein the multi-branch convolutional neural network comprises a pre-convolutional layer, a posture recognition branch network and a retouching recognition branch network. The application improves the accuracy of picture retouching recognition and can be widely applied to the technical field of computer vision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and in particular to a picture retake recognition method, system, device and storage medium. BACKGROUND

[0002] In the data verification of the retail field, in order to ensure the reliability of the data, the picture data obtained needs to be detected for retake to prevent related cheating behaviors. The existing retake detection technology generally detects the moire information of the retake picture to determine whether the picture is retaken, but the types of retake pictures are various, and when the resolution of the screen is high enough or the picture is printed on paper first and then retaken, the moire information of the retake picture is missing, which may cause the retake picture to be unable to be accurately recognized and detected, thereby affecting the authenticity of the data. SUMMARY

[0003] The present application aims to at least partly solve one of the problems in the prior art.

[0004] To this end, one object of the present application is to provide a picture retake recognition method which improves the accuracy of picture retake recognition.

[0005] Another object of the present application is to provide a picture retake recognition system.

[0006] In order to achieve the above technical objects, the technical solutions adopted by the embodiments of the present application include:

[0007] In one aspect, the present application provides a picture retake recognition method, comprising the following steps:

[0008] obtaining a first picture sample, determining the three-dimensional pose information of the first picture sample, and labeling the first picture sample according to the three-dimensional pose information to obtain a training sample set;

[0009] inputting the training sample set into a pre-constructed multi-branch convolutional neural network for training to obtain a trained picture retake recognition model;

[0010] obtaining a picture to be recognized, inputting the picture to be recognized into the picture retake recognition model, and determining whether the picture to be recognized is a retake picture according to the recognition result;

[0011] The multi-branch convolutional neural network includes a pre-convolutional layer, a pose recognition branch network and a retake recognition branch network, the pre-convolutional layer is used for feature extraction of the input picture, the pose recognition branch network is used for recognizing the three-dimensional pose of the input picture according to the extracted picture features, and the retake recognition branch network is used for recognizing whether the input picture is a retake picture according to the extracted picture features and the recognized three-dimensional pose.

[0012] Further, in an embodiment of the present application, the three-dimensional pose information is a three-dimensional pose angle of a foreground object in the first picture sample, and the step of determining the three-dimensional pose information of the first picture sample specifically comprises:

[0013] determining a foreground image of the first picture sample, the foreground image being an image region where the foreground object is located;

[0014] determining a plurality of key feature regions of the foreground image, and extracting feature pixel points in the key feature regions;

[0015] performing pose estimation according to a positional relationship of the feature pixel points to obtain the three-dimensional pose angle of the foreground object.

[0016] Further, in an embodiment of the present application, the step of labeling the first picture sample according to the three-dimensional pose information to obtain a training sample set specifically comprises:

[0017] determining a first label of the first picture sample according to the three-dimensional pose information;

[0018] determining a second label of the first picture sample according to whether the first picture sample is a retake;

[0019] constructing a training sample set according to the first picture sample, the first label and the second label.

[0020] Further, in an embodiment of the present application, the step of inputting the training sample set into a pre-constructed multi-branch convolutional neural network for training specifically comprises:

[0021] inputting the training sample set into the pre-convolutional layer to perform feature extraction on the first picture sample to obtain a first feature subgraph;

[0022] inputting the first feature subgraph into the pose recognition branch network to perform pose recognition, and determining a first loss value according to a pose recognition result and the first label, and further updating parameters of the pre-convolutional layer and the pose recognition branch network according to the first loss value through a back propagation algorithm;

[0023] inputting the first feature subgraph and the pose recognition result into the retake recognition branch network to perform retake recognition, and determining a second loss value according to a retake recognition result and the second label, and further updating parameters of the pre-convolutional layer and the retake recognition branch network according to the second loss value through the back propagation algorithm.

[0024] Further, in an embodiment of the present application, the training of the posture recognition branch network and the training of the retake recognition branch network are iteratively performed alternately, and the step of inputting the training sample set into the pre-constructed multi-branch convolutional neural network for training further comprises the following steps:

[0025] When the first loss value reaches a preset first threshold value and the second loss value reaches a preset second threshold value, the training is stopped.

[0026] Further, in an embodiment of the present application, the first loss value is calculated by the following formula:

[0027]

[0028] wherein, L1 represents the first loss value, N represents the number of first picture samples, euler i represents the first label of the i-th first picture sample, e i represents the posture recognition result of the i-th first picture sample.

[0029] Further, in an embodiment of the present application, the second loss value is calculated by the following formula:

[0030]

[0031] wherein, L2 represents the second loss value, N represents the number of first picture samples, y i represents the second label of the i-th first picture sample, when the i-th first picture sample is a retake, y i = 1, when the i-th first picture sample is not a retake, y i = 0, p i represents the probability that the i-th first picture sample is predicted to be a retake.

[0032] On the other hand, an embodiment of the present application provides a picture retake recognition system, comprising:

[0033] a training sample set determination module configured to obtain first picture samples, determine three-dimensional posture information of the first picture samples, and label the first picture samples according to the three-dimensional posture information to obtain a training sample set;

[0034] a multi-branch convolutional neural network training module configured to input the training sample set into a pre-constructed multi-branch convolutional neural network for training to obtain a trained picture retake recognition model;

[0035] a model recognition module configured to obtain a picture to be recognized, input the picture to be recognized into the picture retake recognition model, and determine whether the picture to be recognized is a retake picture according to a recognition result;

[0036] The multi-branch convolutional neural network comprises a pre-convolutional layer, a posture recognition branch network, and a copy recognition branch network, the pre-convolutional layer is configured to extract features of an input picture, the posture recognition branch network is configured to recognize a three-dimensional posture of the input picture according to the extracted picture features, and the copy recognition branch network is configured to recognize whether the input picture is a copy picture according to the extracted picture features and the recognized three-dimensional posture.

[0037] In another aspect, an embodiment of the present application provides a picture copy recognition device, comprising:

[0038] at least one processor;

[0039] at least one memory configured to store at least one program;

[0040] The at least one processor is configured to implement the picture copy recognition method when the at least one program is executed by the at least one processor.

[0041] In another aspect, an embodiment of the present application further provides a computer readable storage medium, wherein a processor executable program is stored, and the processor executable program is configured to execute the picture copy recognition method when executed by a processor.

[0042] The advantages and beneficial effects of the present application will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the present application:

[0043] The embodiment of the present application first acquires picture samples and labels a training sample set according to three-dimensional posture information, then inputs the training sample set into a pre-constructed multi-branch convolutional neural network, updates network parameters of the pre-convolutional layer and each branch network through joint training of the posture recognition branch network and the copy recognition branch network, and obtains a trained picture copy recognition model, so that the picture copy recognition model can be used for copy recognition of a picture to be recognized. The embodiment of the present application updates the network parameters of the pre-convolutional layer and each branch network through joint training of the posture recognition branch network and the copy recognition branch network, the training of the posture recognition branch network can optimize the pre-convolutional layer, so that the feature expression capability of the extracted picture features can be enhanced, and the accuracy of picture copy recognition is improved; the training of the copy recognition branch network considers the extracted picture features and the recognized three-dimensional posture, and the accuracy of picture copy recognition is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following introduces the drawings needed to be used in the embodiments of the present application. It should be understood that the drawings introduced in the following are only for facilitating the clear description of the technical solutions in the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without paying any creative labor.

[0045] Figure 1 A step flow chart of a picture copy recognition method provided by the embodiment of the present application;

[0046] Figure 2 A structural block diagram of a picture copy recognition system provided by the embodiment of the present application;

[0047] Figure 3 A structural block diagram of a picture copy recognition device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0048] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar notations represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application. For the step numbers in the following embodiments, they are only set for facilitating the description and explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0049] In the description of the present application, the meaning of multiple is two or more, and if the first and the second are described, it is only for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or the sequence of the indicated technical features. In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art.

[0050] With reference to Figure 1 The embodiment of the present application provides a picture copy recognition method, which specifically includes the following steps:

[0051] S101, acquiring a first picture sample, determining three-dimensional pose information of the first picture sample, and labeling the first picture sample according to the three-dimensional pose information to obtain a training sample set.

[0052] Specifically, in actual life scenarios, it can be found that compared with actually taken pictures, the pictures taken by copying are generally taken against a computer, mobile phone screen or printed paper, etc. The three-dimensional pose information (such as Euler angles of objects) of the copied pictures is abnormal compared with the real pictures, and the pictures also lack stereoscopic effect. Therefore, in the process of labeling the picture samples, the embodiment of the present application first determines the three-dimensional pose information of the picture samples, labels the picture samples according to the three-dimensional pose information, and simultaneously performs fitting of the three-dimensional pose information and identification of picture copying in the subsequent model training process.

[0053] Further, as an optional implementation, the three-dimensional pose information is a three-dimensional pose angle of the foreground object in the first picture sample, and the step of determining the three-dimensional pose information of the first picture sample specifically includes:

[0054] A1, determining a foreground image of the first picture sample, the foreground image being an image region where the foreground object is located;

[0055] A2, determining a plurality of key feature regions of the foreground image, and extracting feature pixel points in the key feature regions;

[0056] A3, performing pose estimation according to the positional relationship of the feature pixel points to obtain the three-dimensional pose angle of the foreground object.

[0057] Specifically, taking data verification in the retail and fast-moving consumer goods field as an example, the foreground image can be an image region where a shelf is located, and the key feature regions can be regions such as corners and layered regions of the shelf. The feature pixel points in these regions are extracted, and then pose estimation is performed according to the geometric structure relationship of the actual shelf and the positional relationship of the feature pixel points, so as to obtain the three-dimensional pose angle of the shelf, including the pitch angle, the yaw angle and the roll angle.

[0058] Further, as an optional implementation, the step of labeling the first picture sample according to the three-dimensional pose information to obtain a training sample set specifically includes:

[0059] B1, determining a first label of the first picture sample according to the three-dimensional pose information;

[0060] B2, determining a second label of the first picture sample according to whether the first picture sample is copied;

[0061] B3, constructing a training sample set according to the first picture sample, the first label and the second label.

[0062] Specifically, the first label can be denoted as (pitch, yaw, roll), wherein the pitch represents a pitch angle, the yaw represents a yaw angle, and the roll represents a roll angle; and the second label can be denoted as 1 or 0, wherein 1 represents that the corresponding picture sample is a flipped picture, and 0 represents that the corresponding picture sample is a real picture. Whether a picture is flipped can be determined according to a picture source, or is labeled by manual judgment. The training sample set is constructed by taking the first label and the second label as label data of the first picture sample, so as to form a training sample set.

[0063] In some optional embodiments, the first label and the second label can be combined into one multi-dimensional label, such as (1, pitch, yaw, roll).

[0064] S102, inputting the training sample set into a pre-constructed multi-branch convolutional neural network for training to obtain a trained picture flipping recognition model.

[0065] The multi-branch convolutional neural network comprises a pre-convolutional layer, a posture recognition branch network, and a flipping recognition branch network. The pre-convolutional layer is used for feature extraction of an input picture. The posture recognition branch network is used for recognizing a three-dimensional posture of the input picture according to the extracted picture features. The flipping recognition branch network is used for recognizing whether the input picture is a flipped picture according to the extracted picture features and the recognized three-dimensional posture.

[0066] Specifically, the embodiment of the present application proposes a multi-branch convolutional neural network comprising a pre-convolutional layer, a posture recognition branch network, and a flipping recognition branch network. The posture recognition branch network is used for fitting three-dimensional posture information of a picture sample. The flipping recognition branch network is used for distinguishing whether the picture sample is a flipped picture. The two branch networks share the pre-convolutional layer of the multi-branch convolutional neural network. The training of the posture recognition branch network can optimize the parameters of the pre-convolutional layer, thereby affecting the picture feature extraction of the pre-convolutional layer, enhancing the feature expression capability of the extracted picture features, and further strengthening the learning process of the flipping recognition branch network, and improving the accuracy of picture flipping recognition.

[0067] Further as an optional implementation, the step of inputting the training sample set into the pre-constructed multi-branch convolutional neural network for training specifically comprises:

[0068] C1, inputting the training sample set into the pre-convolutional layer to perform feature extraction on the first picture sample by the pre-convolutional layer to obtain a first feature subgraph;

[0069] C2, inputting the first feature subgraph into the posture recognition branch network for posture recognition, and determining a first loss value according to the posture recognition result and the first label, and further updating the parameters of the pre-convolutional layer and the posture recognition branch network by a back propagation algorithm according to the first loss value.

[0070] C3, inputting the first feature subgraph and the posture recognition result into the copy recognition branch network for copy recognition, and determining a second loss value according to the copy recognition result and the second label, and further updating the parameters of the pre-convolutional layer and the copy recognition branch network according to the second loss value.

[0071] Specifically, after inputting the picture sample in the training sample set into the initialized multi-branch convolutional neural network, the feature subgraph is extracted through the pre-convolutional layer, and then the feature subgraph is input into the posture recognition branch network to obtain the posture recognition result. The accuracy of the posture recognition branch network is evaluated according to the posture recognition result and the first label, so that the parameters of the pre-convolutional layer and the posture recognition branch network can be updated. The feature subgraph and the posture recognition result are input into the copy recognition branch network to obtain the copy recognition result. The accuracy of the copy recognition branch network is evaluated according to the copy recognition result and the second label, so that the parameters of the pre-convolutional layer and the copy recognition branch network can be updated.

[0072] For the posture recognition branch network and the copy recognition branch network, the accuracy of the recognition result can be measured by a loss function. The loss function is defined on a single training sample, and is used to measure the prediction error of a training sample. Specifically, the loss value of a training sample is determined by the label of the training sample and the prediction result of the model for the training sample. In actual training, there are many training samples in a training sample set. Therefore, a cost function is generally used to measure the overall error of the training sample set. The cost function is defined on the entire training sample set, and is used to calculate the average value of the prediction errors of all training samples, so as to better measure the prediction effect of the model. For a general machine learning model, based on the aforementioned cost function, a regular term that measures the complexity of the model can be added as the target function of training. Based on the target function, the loss value of the entire training sample set can be obtained. There are many commonly used loss functions, such as 0-1 loss function, square loss function, absolute loss function, logarithmic loss function, cross-entropy loss function, etc. All of them can be used as the loss function of the machine learning model, and will not be described one by one here. In the embodiment of the present application, any one of the loss functions can be selected to determine the loss value of training. Based on the loss value of training, the parameters of the model are updated by using the back propagation algorithm. When the loss value reaches the corresponding threshold value, the trained picture copy recognition model can be obtained.

[0073] Further as an optional implementation, the training of the posture recognition branch network and the training of the copy recognition branch network are alternately iterated. The step of inputting the training sample set into the pre-constructed multi-branch convolutional neural network for training further includes the following steps:

[0074] Training stops when the first loss value reaches a preset first threshold and the second loss value reaches a preset second threshold.

[0075] Specifically, the training of the pose recognition branch network and the training of the re-photographing recognition branch network are carried out alternately and iteratively. That is, the pose recognition result output by the first optimized pose recognition branch network is used as the input for the first training of the re-photographing recognition branch network. After the re-photographing recognition branch network updates the parameters of the previous convolutional layer, the pose recognition branch network is optimized for the second time. This process is repeated until both the first loss value and the second loss value reach the corresponding threshold, at which point training can be stopped.

[0076] As an optional implementation, the first loss value is calculated using the following formula:

[0077]

[0078] Where L1 represents the first loss value, N represents the number of the first image samples, and euler i Let e ​​represent the first label of the i-th first image sample. i This represents the pose recognition result of the i-th first image sample.

[0079] Specifically, euler i =(pitch) i yaw i , roll i ) represents the first label of the i-th first image sample, i.e., the pre-labeled pitch angle, yaw angle, and roll angle, e i =(a i b i c i ) represents the attitude recognition result of the i-th first image sample, namely the pitch angle, yaw angle and roll angle.

[0080] Understandably, the closer the identified pitch, yaw, and roll angles are to the labeled pitch, yaw, and roll angles, the smaller the value of the first loss value, indicating that the attitude recognition branch network has higher accuracy.

[0081] As an optional implementation, the second loss value is calculated using the following formula:

[0082]

[0083] Where L2 represents the second loss value, N represents the number of first image samples, and y i This represents the second label of the i-th first image sample. When the i-th first image sample is a reproduction, y i =1, when the i-th first image sample is not a reproduction.i = 0, p i Pi represents a probability that the i-th first picture sample is a flip.

[0084] S103, acquiring a picture to be identified, inputting the picture to be identified into the picture flip identification model, and determining whether the picture to be identified is a flip picture according to an identification result.

[0085] Specifically, after the picture to be identified is input into the picture flip identification model, the feature subgraph is extracted by the pre-convolutional layer (optimized after model training) first, and then the three-dimensional pose information is identified by the pose recognition branch network, and then the three-dimensional pose information and the feature subgraph are input into the flip identification branch network, so that whether the picture is a flip picture can be identified.

[0086] The method steps of the embodiments of the application are described above. It can be recognized that the network parameters of the pre-convolutional layer and each branch network are updated by the joint training of the pose recognition branch network and the flip identification branch network. Since the training of the pose recognition branch network can optimize the pre-convolutional layer, the feature expression capability of the extracted picture features can be enhanced, and thus the accuracy of picture flip identification is improved. Since the training of the flip identification branch network considers the extracted picture features and the identified three-dimensional pose at the same time, the accuracy of picture flip identification is further improved.

[0087] Referring to Figure 2 , the embodiments of the application provide a picture flip identification system, comprising:

[0088] A training sample set determination module is configured to acquire a first picture sample, determine three-dimensional pose information of the first picture sample, and label the first picture sample according to the three-dimensional pose information to obtain a training sample set.

[0089] A multi-branch convolutional neural network training module is configured to input the training sample set into a pre-constructed multi-branch convolutional neural network for training to obtain a trained picture flip identification model.

[0090] A model identification module is configured to acquire a picture to be identified, input the picture to be identified into the picture flip identification model, and determine whether the picture to be identified is a flip picture according to an identification result.

[0091] The multi-branch convolutional neural network comprises a pre-convolutional layer, a pose recognition branch network, and a flip identification branch network. The pre-convolutional layer is configured to extract features of an input picture. The pose recognition branch network is configured to identify a three-dimensional pose of the input picture according to the extracted picture features. The flip identification branch network is configured to identify whether the input picture is a flip picture according to the extracted picture features and the identified three-dimensional pose.

[0092] The contents in the method embodiments are applicable to the system embodiments, the system embodiments specifically implement the same functions as the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0093] Referring to Figure 3 The embodiment of the present application provides a picture flipping recognition device, which comprises:

[0094] at least one processor;

[0095] at least one memory for storing at least one program;

[0096] When the at least one program is executed by the at least one processor, the at least one processor implements the picture flipping recognition method.

[0097] The contents in the method embodiments are applicable to the device embodiments, the device embodiments specifically implement the same functions as the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0098] The embodiment of the present application further provides a computer readable storage medium, wherein a program executable by a processor is stored, and the program executable by the processor is used for executing the picture flipping recognition method when executed by the processor.

[0099] The computer readable storage medium of the embodiment of the present application can execute the picture flipping recognition method provided by the method embodiments of the present application, execute the steps of any combination of the method embodiments, and has the corresponding functions and beneficial effects of the method.

[0100] The embodiment of the present application further discloses a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of a computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method shown in the figure. Figure 1

[0101] ​In alternative embodiments, the functions / operations in the flow diagrams can occur in sequences other than those depicted. For example, two operations shown in succession can in fact be executed substantially concurrently or the operations can sometimes be executed in the reverse order depending upon the functionality / operations involved. Such variations are contemplated to be within the scope of the present application. Embodiments presented and described in the flow diagrams are examples only and are used to provide an enabling teaching for the present application. The processes disclosed are not limited to the order or specific blocks described. Alternative embodiments are contemplated, in which the order of the blocks is changed and where some blocks are performed in parallel rather than sequentially.

[0102] Moreover, while the present application has been described in the context of functional modules, it is to be understood that one or more of the functions and / or features described above can be integrated within a single physical device and / or software module, or one or more functions and / or features can be implemented in separate physical devices or software modules. It will also be appreciated that detailed discussion of the actual implementation of each module is not necessary for an enabling understanding of the application. Rather, the actual implementation is most readily derived from the description of the functionality of the various functional modules, in conjunction with the understanding of the properties, functions and interrelationships of the various functional modules presented in the context of the device disclosed herein. Therefore, the scope of the application is best understood from the appended claims, in conjunction with the full description and examples provided. It is to be understood that the specific concepts presented are merely illustrative of the application and are not intended to limit the scope of the application as defined by the claims. The scope of the application is defined by the claims and the full extent of equivalents to which such claims are entitled.

[0103] If the above functions are implemented in software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0104] The logic and / or steps represented in the flow diagrams or otherwise described herein, for example, can be embodied in non-transitory computer- readable media, executed by an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or in conjunction with which the instructions can be executed. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium.

[0105] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0106] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following technologies, known in the art, or combinations thereof, can be used: a discrete logic circuit having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), and / or the like.

[0107] In the above description of the present specification, reference has been made to descriptive terms such as "one embodiment / exemplification", "another embodiment / exemplification", or "some embodiments / exemplifications" etc. It is understood that such terms are not intended to mean that the described specific feature, structure, material or characteristic was included in at least one embodiment or exemplification. It is also understood that the specific features, structures, materials or characteristics described were not necessarily included in the same embodiment or exemplification. Furthermore, it is understood that the specific features, structures, materials or characteristics were described are not necessarily included in any one or more embodiments or exemplifications.

[0108] While the embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary and are not to be construed as limiting the scope of the application. The scope of the application is defined by the appended claims and their equivalents.

[0109] The above is the specific description of the preferred embodiment of the application, but the application is not limited to the above-mentioned embodiments, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the application, and these equivalent modifications or replacements are all included in the scope defined by the claims of the application.

Claims

1. A picture double exposure identification method, characterized in that, Includes the following steps: A first image sample is obtained, its three-dimensional pose information is determined, and the first image sample is labeled according to the three-dimensional pose information to obtain a training sample set. The training sample set is input into a pre-constructed multi-branch convolutional neural network for training to obtain a trained image reproduction recognition model. Obtain the image to be identified, input the image to be identified into the image reproduction recognition model, and determine whether the image to be identified is a reproduced image based on the recognition result; The multi-branch convolutional neural network includes a pre-convolutional layer, a pose recognition branch network, and a re-image recognition branch network. The pre-convolutional layer is used to extract features from the input image. The pose recognition branch network is used to identify the three-dimensional pose of the input image based on the extracted image features. The re-image recognition branch network is used to identify whether the input image is a re-image based on the extracted image features and the identified three-dimensional pose. The step of annotating the first image sample based on the three-dimensional pose information to obtain the training sample set specifically includes: The first label of the first image sample is determined based on the three-dimensional pose information; The second label of the first image sample is determined based on whether the first image sample is a re-photographed image; A training sample set is constructed based on the first image sample, the first label, and the second label; The step of inputting the training sample set into a pre-constructed multi-branch convolutional neural network for training specifically includes: The training sample set is input into the pre-convolutional layer, and the first feature sub-image is obtained by extracting features from the first image sample through the pre-convolutional layer. The first feature sub-image is input into the pose recognition branch network for pose recognition, and a first loss value is determined based on the pose recognition result and the first label. Then, the parameters of the pre-convolutional layer and the pose recognition branch network are updated based on the first loss value through the backpropagation algorithm. The first feature sub-image and pose recognition result are input into the re-photograph recognition branch network for re-photograph recognition. The second loss value is determined based on the re-photograph recognition result and the second label. Then, the parameters of the pre-convolutional layer and the re-photograph recognition branch network are updated based on the second loss value through the backpropagation algorithm. The three-dimensional pose information refers to the three-dimensional pose angle of the foreground object in the first image sample. The step of determining the three-dimensional pose information of the first image sample specifically includes: Determine the foreground image of the first image sample, wherein the foreground image is the image region where the foreground object is located; Identify multiple key feature regions of the foreground image and extract feature pixels from the key feature regions; The pose is estimated based on the positional relationship of the feature pixels to obtain the three-dimensional pose angle of the foreground object.

2. The image reproduction and recognition method according to claim 1, characterized in that, The training of the pose recognition branch network and the training of the re-image recognition branch network are performed alternately and iteratively. The step of inputting the training sample set into the pre-constructed multi-branch convolutional neural network for training also includes the following steps: Training stops when the first loss value reaches a preset first threshold and the second loss value reaches a preset second threshold.

3. The image reproduction and recognition method according to claim 1, characterized in that, The first loss value is calculated using the following formula: wherein L1 represents a first loss value, N represents a number of the first picture samples, euler i represents a first label of the i-th first picture sample, e i represents a pose recognition result of the i-th first picture sample.

4. The image reproduction and recognition method according to claim 1, characterized in that, The second loss value is calculated using the following formula: wherein L2 represents a second loss value, N represents a number of first picture samples, y i represents a second label of the i-th first picture sample, y i = 1 when the i-th first picture sample is a flip, and y i = 0 when the i-th first picture sample is not a flip, p i represents a probability that the i-th first picture sample is predicted to be a flip.

5. An image reproduction and recognition system, characterized in that, A method for implementing an image reproduction and recognition method as described in any one of claims 1 to 4 includes: The training sample set determination module is used to acquire a first image sample, determine the three-dimensional pose information of the first image sample, and annotate the first image sample according to the three-dimensional pose information to obtain a training sample set. A multi-branch convolutional neural network training module is used to input the training sample set into a pre-constructed multi-branch convolutional neural network for training, so as to obtain a trained image reproduction recognition model. The model recognition module is used to acquire the image to be recognized, input the image to be recognized into the image reproduction recognition model, and determine whether the image to be recognized is a reproduced image based on the recognition result. The multi-branch convolutional neural network includes a pre-convolutional layer, a pose recognition branch network, and a re-image recognition branch network. The pre-convolutional layer is used to extract features from the input image. The pose recognition branch network is used to identify the 3D pose of the input image based on the extracted image features. The re-image recognition branch network is used to identify whether the input image is a re-image based on the extracted image features and the identified 3D pose.

6. A photo reproduction and recognition device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements an image reproduction and recognition method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a processor-executable program, characterized in that, The program executable by the processor is used, when executed by the processor, to perform an image reproduction and recognition method as described in any one of claims 1 to 4.

Images