CN112258481A - Portal photo reproduction detection method - Google Patents
Portal photo reproduction detection method Download PDFInfo
- Publication number
- CN112258481A CN112258481A CN202011145416.4A CN202011145416A CN112258481A CN 112258481 A CN112258481 A CN 112258481A CN 202011145416 A CN202011145416 A CN 202011145416A CN 112258481 A CN112258481 A CN 112258481A
- Authority
- CN
- China
- Prior art keywords
- photo
- reproduction
- detecting
- copying
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 34
- 238000000034 method Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000002372 labelling Methods 0.000 claims abstract description 21
- 238000012360 testing method Methods 0.000 claims abstract description 15
- 238000003062 neural network model Methods 0.000 claims abstract description 11
- 238000012795 verification Methods 0.000 claims abstract description 10
- 238000001914 filtration Methods 0.000 claims abstract description 6
- 238000010606 normalization Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 7
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000012797 qualification Methods 0.000 claims description 6
- 238000012550 audit Methods 0.000 claims description 5
- 238000013434 data augmentation Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 2
- 230000002779 inactivation Effects 0.000 claims description 2
- 238000007689 inspection Methods 0.000 claims description 2
- 238000010200 validation analysis Methods 0.000 claims description 2
- 230000006870 function Effects 0.000 description 11
- 238000012552 review Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000004880 explosion Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000008033 biological extinction Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 210000004205 output neuron Anatomy 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The application provides a method for detecting the reproduction of a door picture, which comprises the following steps: carrying out characteristic marking on the obtained photo with the copying characteristic, wherein the characteristic marking at least comprises one of fuzzy characteristic marking, moire phenomenon characteristic marking, black edge characteristic marking, watermark characteristic marking and mobile phone frame characteristic standard; dividing the photos subjected to feature labeling into a training set, a test set and a verification set; carrying out graying treatment; training the training set through the constructed neural network model, testing through the testing set, and verifying through the verification set until the model is converged to obtain a door photograph copying detector; outputting the reproduction probability of the photo to be detected through a door face photograph reproduction detector; determining a detection threshold value according to the source of the photo to be detected; and detecting and filtering the copied photos according to the copying probability and the detection threshold value. The copying detector is constructed according to the characteristics of the copied picture, and the detection precision of the copied picture is improved.
Description
Technical Field
The application belongs to the technical field of data processing, and particularly relates to a door photo reproduction detection method.
Background
The fresh electronic commerce industry requires that a shop face photo of a shop where a merchant is located is uploaded as a qualification certificate when the merchant is registered, and meanwhile, a salesperson of an enterprise needs to upload a shop face photo of a visited merchant as a visiting basis when the salesperson visits the merchant on line, and subsequently, auditors need to perform online audit. Usually, an enterprise requires that only a mobile phone original camera is called when a photo is uploaded, and a shooting place is required to be consistent with uploaded store position information so as to avoid some common cheating means, but at present, many merchants and sales staff cheat by uploading and copying the photo. The pixels of the existing mobile phone cameras are higher and higher, the similarity between the copied photos and the real photos is higher and higher, and the auditors are easy to mislead, so that direct economic loss is brought to enterprises. Therefore, it is an urgent need of the fresh electronic business enterprise to identify whether the uploaded door face photos belong to the reproduction.
The existing copying detection mode is manual review and review by a review model constructed through machine learning, the pure manual review has the advantages of high detection precision and low speed, the average review speed is 1.3 minutes per piece, and the machine learning is adopted for review, so that the detection speed is increased, and the detection precision is reduced.
Disclosure of Invention
In order to solve at least one of the above technical problems, the present application provides a door photograph copying detection method based on a convolutional neural network, where the convolutional neural network is one of the most widely applied algorithms in the field of image processing at present, and the convolutional neural network extracts features of an image in a convolutional manner, performs feature dimension reduction in a pooling manner, and finally inputs the extracted features into a full connection layer for classification. The mode of judging whether the photo belongs to the reproduction or not through the convolutional neural network has high accuracy and high speed.
The application of the method for detecting the copying of the door picture comprises the following steps: carrying out feature labeling on the obtained photo with the copying characteristic, wherein the feature labeling at least comprises at least one of fuzzy feature labeling, moire phenomenon feature labeling, black edge feature labeling, watermark feature labeling and mobile phone frame feature standards; dividing the photos subjected to feature labeling into a training set, a test set and a verification set; carrying out graying processing on the picture; training the training set through a built neural network model, testing through the test set, and verifying through the verification set until the model is converged to obtain a door face photograph copying detector; outputting the reproduction probability of the photo to be detected through the door face photograph reproduction detector; determining a detection threshold value according to the source of the photo to be detected; and detecting and filtering the copied photos according to the copying probability and the detection threshold value.
Preferably, according to 8: 1: a scale of 1 divides the photographs into a training set, a test set, and a validation set.
Preferably, the picture is grayed according to a conversion formula G ═ pR + qG + wB, where G represents a single channel of a grayscale map, RGB represents three channels of a color map, respectively, p is set to 0.340 to 0.344, q is set to 0.505 to 0.509, w is set to 0.149 to 0.153, and p + q + w is 1.
Preferably, the method further includes, after the graying the picture: and carrying out data augmentation on the photo by changing the brightness of the photo.
Preferably, the changing of the brightness of the photograph includes: increasing the picture brightness to 1.5 times the original brightness and/or decreasing the picture brightness to 0.7 times the original brightness.
Preferably, the method further includes, after the graying the picture: all photographs were normalized for size.
Preferably, the constructed neural network model includes: the system comprises a first convolution layer, two basic block modules and two fully-connected layers, wherein each basic block module is formed by 4 channels in parallel, the first channel comprises 3 convolution layers of 3 x 3, the second channel comprises 2 convolution layers of 3 x 3, the third channel comprises a convolution layer of 3 x 3, the fourth channel is an identity module, and a pooling layer is added behind the first convolution layer and the two basic block modules.
Preferably, in the constructed neural network model, a dropout layer for neuron inactivation is connected behind the first full-connection layer, and a relu function is adopted as an activation function in the neural network model.
Preferably, a batch normalization layer is provided before the activation function.
Preferably, the determining the detection threshold according to the source of the photo to be detected comprises: and if the source of the photo to be detected is the store qualification inspection business, setting the detection threshold value to be 50%. And if the source of the photo to be detected is a sale visit audit business, setting the detection threshold value to be 90%.
The method and the device have the advantages that the copying characteristics are found through summarization, data marking is carried out according to the copying characteristics, a data set special for copying detection is obtained, data are preprocessed in a special data augmentation mode of image graying and changing of the brightness degree of the image, and the detection precision can be effectively improved. On the other hand, different copying threshold values are defined according to different application scenes, and the application is more flexible and efficient.
Drawings
FIG. 1 is a flow chart of a preferred embodiment of a method for detecting a portrait impression of a door of the present application.
Fig. 2 is a diagram of a convolutional neural network model structure according to the embodiment shown in fig. 1 of the present application.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present application clearer, the technical solutions in the embodiments of the present application will be described in more detail below with reference to the accompanying drawings in the embodiments of the present application. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all embodiments of the present application. The embodiments described below with reference to the drawings are exemplary and intended to be used for explaining the present application, and should not be construed as limiting the present application. All other embodiments obtained by a person of ordinary skill in the art without any inventive work based on the embodiments in the present application are within the scope of protection of the present application. Embodiments of the present application will be described in detail below with reference to the drawings.
The application provides a door face photo reproduction detection method which is mainly applied to two aspects:
1. and (3) auditing the face of the enterprise platform: the system is used for filtering the uploaded copied cheated photos, and entering the subsequent auditing step when the photos identified as normal photos are filtered;
2. sale anti-cheating: and the system is used for detecting the pictures uploaded during the sales visit and delivering the results to the wind control department.
Aiming at the existing problems, the invention provides a method for detecting the reproduction of a door photograph, which can quickly and accurately detect whether the door photograph is a reproduced photograph or not so as to stop cheating and avoid economic loss of enterprises.
The technical scheme adopted by the invention for solving the technical problem is as follows: a method for detecting a portrait reproduction, as shown in fig. 1, includes the following steps:
s1, acquiring a door picture, summarizing the unique characteristics of the copied picture, marking data according to the characteristics, and constructing a data set special for copying detection;
step S2, performing special preprocessing aiming at the characteristics of the copied photo;
s3, constructing a special neural network model aiming at the characteristics of the photos to be copied, training the model by using the constructed data set, and obtaining a face photo copying detector after the model is converged;
and step S4, detecting the door face photograph according to the detector acquired in the step S3, outputting a threshold value, and judging the photograph to be recognized as a copied photograph when the threshold value of the photograph to be recognized is greater than or equal to a set probability threshold value.
In step S1, since deep learning model training requires a large number of labeled training samples, it is particularly important to acquire a door picture to be detected. The method comprises the steps of collecting photos of a shop front photo uploaded by a sales visit client and a qualification audit shop front photo uploaded by the client, analyzing the photos, summarizing unique characteristics of the copied photos, wherein the characteristics comprise that the copied photos are more fuzzy, the copied photos have a special Moire phenomenon, the peripheries of the copied photos have an obvious black edge phenomenon, part of the copied photos have a possibility of a non-camera watermark phenomenon, and part of the copied photos contain a mobile phone frame, and manually labeling the collected photos according to the characteristics to obtain a special data set for copying detection.
In some alternative embodiments, the labeled data is expressed as 8: 1: the ratio of 1 is split into a training set, a test set and a verification set. Training set data samples for model fitting; the verification set is a sample set reserved in the model training process and can be used for adjusting the hyper-parameters of the model and carrying out preliminary evaluation on the capability of the model, and in the neural network, the verification data set is used for searching the optimal network depth, or determining the stopping point of a back propagation algorithm or selecting the number of hidden layer neurons in the neural network; the test set is used to evaluate the generalization ability of the model final model. But not as a basis for algorithm-related selection of parameters, selection features, and the like.
In step S2, since the feature differences between the real photo and the copied photo are not correlated with the color information, the color information may cause more noise. Therefore, the original RGB three-channel photograph needs to be converted into a single-channel gray image, and the conversion formula is: g & ltpR & gt + qG & ltwB & gt (G & ltg & gt + qG & ltwB & gt represents a single channel of a gray scale image, and RGB & ltRGB & gt respectively represents three channels of a color image), wherein p is set to be 0.340-0.344, q is set to be 0.505-0.509, w is set to be 0.149-0.153, and p + q & ltw & gt is 1, and p is set to be 0.342, q is set to be 0.507, and w is set to be 0.151 in order to detect a shot picture.
Data augmentation often brings about improvement of model accuracy, but the data augmentation mode also needs attention. The conventional method of randomly cutting photos and adding noise is not suitable because it tends to cause loss of features in the application scene. And because some photos in the real photos are shot in the strong light environment or the night environment, the proportion of the photos is low, but the influence on the classification result is large, so that the precision can be effectively improved by changing the brightness of the photos to simulate the strong light environment and the night environment, the effect is best when the brightness of the photos is increased to 1.5 times or the brightness of the photos is reduced to 0.7 time in the test process, or the two aspects are expanded simultaneously, and the data set can be expanded to 3 times of the original data set.
And (3) carrying out data normalization processing, uniformly normalizing all the photos to 227 × 227, wherein the too large photos can cause difficulty in model training, and the too small photos can cause too much loss of reproduction features, and then inputting the photos into a subsequent CNN network.
In step S3, the present application constructs a dedicated network model for the features unique to the copied photo, and the network uses a plurality of different convolution kernels to process in parallel to extract more features.
As shown in fig. 2, the network includes 1 convolutional layer, two basic blocks and two fully connected layers. Basicblock consists of 4 channels in parallel, the first channel containing 3 × 3 convolutional layers instead of one 7 × 7 convolutional layer, the second channel containing 2 3 × 3 convolutional layers instead of one 5 × 5 convolutional layer, the third channel containing one 3 × 3 convolutional layer, and the fourth channel being an identity module. The pooling layer is added after both the first layer convolution and the two basicblocks. After the use is carried out the experiment and is verified repeatedly from the reproduction photo data set of establishing, the network depth design of this application is at 9 layers at last, and the degree of depth and the width of network increase again behind 9 layers are not promoting the recognition effect of model, can increase unnecessary calculated amount simultaneously.
The number of convolution kernels in each layer is respectively 64, 128, 256 and 256; each pooling layer adopts a maximum pooling mode, the pooling size is 2 x 2, and the step length is 2; the number of output neurons of the full connection layer is 128 and 2 respectively, wherein 2 is the number of categories for outputting softmax, namely, the reproduction and the normal category.
In some optional embodiments, a dropout layer is further connected to the first fully-connected layer to improve the generalization of the model and avoid overfitting, the dropout layer randomly inactivates part of neurons of the network during training, and corresponding weights do not participate in forward and backward propagation in a single training process, so that the neural network can learn general commonality and avoid overfitting. The rate parameter of dropout is set to 0.5, i.e., the probability of each neuron being inactive is 0.5.
In some optional embodiments, the activation functions in the network are all relu functions, and the relu activation functions have better performance than traditional sigmoid and tanh activation functions, and can effectively avoid gradient extinction and gradient explosion, and the calculation formula is as follows:
f(x)＝max(0，x)。
in some optional embodiments, a batch normalization layer is added behind each convolution layer and the first full-connection layer to accelerate model training and improve precision, wherein batch normalization is to enable an input value of nonlinear transformation to fall into an input sensitive region through a certain normalization means, so that gradient explosion is avoided and the convergence speed of the model is accelerated; the two main parameters momentum and epsilon of the batch normalization layer during training are set to 0.99 and 0.001, respectively. Usually, the batch normalization layer is placed behind relu when the network is designed, but the effect of placing the batch normalization layer in front of relu is better when training and recognition of the copying detection are carried out through a large amount of tests and analysis and summary of experimental data.
It can be understood that the Batch Normalization layer Batch Normalization (BN) is used to solve the problem of gradient disappearance/explosion, the problem that the distribution of each layer of input changes during training, and more generally, the problem that the parameters of the current layer change, and the distribution of each layer of input changes during training (internal covariate shift), so the Batch Normalization layer is a skilled method for training neural network, can solve some problems during training, and makes the distribution of data consistent, so that the model is easy and stable during training.
During model training, a cross-validation mode is used to select the final hyper-parameters, including initial learning rate, learning rate decay interval, batch size and the like. And measuring the performance of the model and the generalization capability of the model by using the accuracy of the model on a verification set separated in advance. And stopping training when the model loss is converged and the accuracy is not improved any more, and taking the converged model as a finally used reproduction detection model.
In step S4, it is determined that the photo to be detected is copied if the output probability of the photo is higher than or equal to the threshold, and different thresholds are set according to different usage scenarios during usage, for example, when the copy detector is used in store qualification screening service, the recall rate and the accuracy rate of a normal picture will be more emphasized, so the threshold is set to 50% in the present application. When the copying detector is used in the sales visit audit business, the recall rate and the precision rate of the copied photos are more emphasized, so that the threshold value is set to be 90%. Different thresholds are set in different application scenes, so that the application is more flexible and efficient.
According to the method for detecting the reprint of the shop photo, the reprint detector with high accuracy can be obtained by training the model or finely adjusting the existing model when in need, whether the shop photo belongs to the reprint photo or not can be automatically identified, efficiency and accuracy are greatly improved compared with manual review, the problems existing in the business of shop qualification review and sales visit review of enterprises can be effectively solved, and the method is a good innovation scheme.
The innovation point of the application is that:
1. for the reproduction detection data without the source data set, the reproduction detection special data set is obtained by summarizing and finding the reproduction characteristics and labeling the characteristics according to the reproduction characteristics, wherein the characteristics at least comprise at least one of fuzzy characteristic labeling, moire phenomenon characteristic labeling, black edge characteristic labeling, watermark characteristic labeling and mobile phone frame characteristic standards;
2. special preprocessing is carried out aiming at the copying characteristics, the preprocessing comprises graying the picture, changing the light and shade degree and normalizing, and specific values of parameters in graying, the light and shade degree and normalization are obtained through experimental analysis according to the unique copying characteristics;
3. constructing a new neural network aiming at the reproduction characteristics, wherein the number of neurons and other related parameters are specific parameters obtained according to reproduction experiment analysis;
4. different thresholds are used for different scenarios.
This application another aspect provides a door face photograph reproduction detection device corresponding to above-mentioned method, and the device mainly includes the reproduction detector, and:
the threshold setting module is used for determining a detection threshold according to the source of the photo to be detected;
and the detection and filtering module is used for detecting and filtering the copied photos according to the copying probability and the detection threshold value.
The copying detector is obtained based on deep learning, a new neural network model is built according to the description and aiming at the characteristics of the copied photos, the built data set is used for training the model, and the face copying detector is obtained after the model converges.
In other alternative embodiments, the features of the photos are extracted by a traditional algorithm, and the extracted features are input into classifiers such as XGB, SVM and the like for classification.
In other aspects of the present application, a computer device includes a processor, a memory, and a computer program stored on the memory and executable on the processor, the processor executing the computer program for implementing the method for detecting a portrait duplication.
In other aspects of the application, a readable storage medium stores a computer program which, when executed by a processor, is adapted to implement a method of detecting a portrait impression as described above.
In particular, according to embodiments of the present application, the processes described above with reference to the flow diagrams may be implemented as a computer software program, in particular a computer program installed on a mobile phone terminal, which is capable of interacting with a server. For example, embodiments of the present application include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated by the flow chart. The computer storage media of the present application may be computer-readable signal media or computer-readable storage media or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The modules or units described in the embodiments of the present application may be implemented by software or hardware. The modules or units described may also be provided in a processor, the names of which in some cases do not constitute a limitation of the module or unit itself.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present application should be covered within the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A method for detecting the reproduction of a door picture is characterized by comprising the following steps:
carrying out feature labeling on the obtained photo with the copying characteristic, wherein the feature labeling at least comprises at least one of fuzzy feature labeling, moire phenomenon feature labeling, black edge feature labeling, watermark feature labeling and mobile phone frame feature standards;
dividing the photos subjected to feature labeling into a training set, a test set and a verification set;
carrying out graying processing on the picture;
training the training set through a built neural network model, testing through the test set, and verifying through the verification set until the model is converged to obtain a door face photograph copying detector;
outputting the reproduction probability of the photo to be detected through the door face photograph reproduction detector;
determining a detection threshold value according to the source of the photo to be detected;
and detecting and filtering the copied photos according to the copying probability and the detection threshold value.
2. The method for detecting a faceprint reproduction as claimed in claim 1, wherein the method comprises the steps of: 1: a scale of 1 divides the photographs into a training set, a test set, and a validation set.
3. The method for detecting the copying of the door face as claimed in claim 1, wherein the picture is grayed according to the conversion formula G ═ pR + qG + wB, where G represents a single channel of a grayscale map, RGB represents three channels of a color map, respectively, p is set to 0.340 to 0.344, q is set to 0.505 to 0.509, w is set to 0.149 to 0.153, and p + q + w is 1.
4. The method for detecting a portrait reproduction of claim 1, further comprising, after graying the photograph:
and carrying out data augmentation on the photo by changing the brightness of the photo.
5. The method for detecting a faceprint reproduction as claimed in claim 4, wherein said changing the brightness of the photograph comprises:
increasing the picture brightness to 1.5 times the original brightness and/or decreasing the picture brightness to 0.7 times the original brightness.
6. The method for detecting a portrait reproduction of claim 1, further comprising, after graying the photograph: all photographs were normalized for size.
7. The method for detecting the reproduction of a door photograph as claimed in claim 1, wherein the neural network model is constructed by:
the system comprises a first convolution layer, two basic block modules and two fully-connected layers, wherein each basic block module is formed by 4 channels in parallel, the first channel comprises 3 convolution layers of 3 x 3, the second channel comprises 2 convolution layers of 3 x 3, the third channel comprises a convolution layer of 3 x 3, the fourth channel is an identity module, and a pooling layer is added behind the first convolution layer and the two basic block modules.
8. The method for detecting the copying of the door face as claimed in claim 7, wherein in the constructed neural network model, a dropout layer for neuron inactivation is connected after the first full-link layer, and a relu function is adopted as an activation function in the neural network model.
9. The method for detecting faceprint reproduction as claimed in claim 8, wherein a batch normalization layer is provided before the activation function.
10. The method for detecting a faceprint duplication as claimed in claim 1, wherein determining a detection threshold based on the source of the photo to be detected comprises:
and if the source of the photo to be detected is the store qualification inspection business, setting the detection threshold value to be 50%. And if the source of the photo to be detected is a sale visit audit business, setting the detection threshold value to be 90%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011145416.4A CN112258481A (en) | 2020-10-23 | 2020-10-23 | Portal photo reproduction detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011145416.4A CN112258481A (en) | 2020-10-23 | 2020-10-23 | Portal photo reproduction detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112258481A true CN112258481A (en) | 2021-01-22 |
Family
ID=74264977
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011145416.4A Pending CN112258481A (en) | 2020-10-23 | 2020-10-23 | Portal photo reproduction detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112258481A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114693629A (en) * | 2022-03-25 | 2022-07-01 | 北京城市网邻信息技术有限公司 | Image recognition method and device, electronic equipment and readable medium |
| CN117333762A (en) * | 2023-12-02 | 2024-01-02 | 深圳爱莫科技有限公司 | Image reproduction identification method based on multi-feature fusion |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460649A (en) * | 2017-02-22 | 2018-08-28 | 阿里巴巴集团控股有限公司 | A kind of image-recognizing method and device |
| CN109815960A (en) * | 2018-12-21 | 2019-05-28 | 深圳壹账通智能科技有限公司 | Reproduction image-recognizing method, device, equipment and medium based on deep learning |
| CN110348511A (en) * | 2019-07-08 | 2019-10-18 | 创新奇智(青岛)科技有限公司 | A kind of picture reproduction detection method, system and electronic equipment |
| CN110569716A (en) * | 2019-07-26 | 2019-12-13 | 浙江工业大学 | Goods shelf image copying detection method |
| CN110717450A (en) * | 2019-10-09 | 2020-01-21 | 深圳大学 | Training method and detection method for automatically identifying copied image of original document |
| CN111008651A (en) * | 2019-11-13 | 2020-04-14 | 科大国创软件股份有限公司 | Image reproduction detection method based on multi-feature fusion |
| CN111191568A (en) * | 2019-12-26 | 2020-05-22 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for identifying copied image |
| CN111275685A (en) * | 2020-01-20 | 2020-06-12 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for identifying copied image of identity document |
| WO2020147398A1 (en) * | 2019-01-17 | 2020-07-23 | 平安科技（深圳）有限公司 | Reproduced image detection method and device, computer device and storage medium |
| WO2020147445A1 (en) * | 2019-01-16 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Rephotographed image recognition method and apparatus, computer device, and computer-readable storage medium |
| CN111476268A (en) * | 2020-03-04 | 2020-07-31 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for training reproduction recognition model and image recognition |
-
2020
- 2020-10-23 CN CN202011145416.4A patent/CN112258481A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460649A (en) * | 2017-02-22 | 2018-08-28 | 阿里巴巴集团控股有限公司 | A kind of image-recognizing method and device |
| CN109815960A (en) * | 2018-12-21 | 2019-05-28 | 深圳壹账通智能科技有限公司 | Reproduction image-recognizing method, device, equipment and medium based on deep learning |
| WO2020147445A1 (en) * | 2019-01-16 | 2020-07-23 | 深圳壹账通智能科技有限公司 | Rephotographed image recognition method and apparatus, computer device, and computer-readable storage medium |
| WO2020147398A1 (en) * | 2019-01-17 | 2020-07-23 | 平安科技（深圳）有限公司 | Reproduced image detection method and device, computer device and storage medium |
| CN110348511A (en) * | 2019-07-08 | 2019-10-18 | 创新奇智(青岛)科技有限公司 | A kind of picture reproduction detection method, system and electronic equipment |
| CN110569716A (en) * | 2019-07-26 | 2019-12-13 | 浙江工业大学 | Goods shelf image copying detection method |
| CN110717450A (en) * | 2019-10-09 | 2020-01-21 | 深圳大学 | Training method and detection method for automatically identifying copied image of original document |
| CN111008651A (en) * | 2019-11-13 | 2020-04-14 | 科大国创软件股份有限公司 | Image reproduction detection method based on multi-feature fusion |
| CN111191568A (en) * | 2019-12-26 | 2020-05-22 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for identifying copied image |
| CN111275685A (en) * | 2020-01-20 | 2020-06-12 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for identifying copied image of identity document |
| CN111476268A (en) * | 2020-03-04 | 2020-07-31 | 中国平安人寿保险股份有限公司 | Method, device, equipment and medium for training reproduction recognition model and image recognition |
Non-Patent Citations (2)
| Title |
|---|
| YAJIE XING 等: ""Coupling Two-Stream RGB-D Semantic Segmentation Network by Idempotent Mappings"", 2019 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING (ICIP) * |
| 叶宇光: ""基于深度残差网络的图像识别技术研究"", 《韶关学院学报》, vol. 41, no. 6 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114693629A (en) * | 2022-03-25 | 2022-07-01 | 北京城市网邻信息技术有限公司 | Image recognition method and device, electronic equipment and readable medium |
| CN117333762A (en) * | 2023-12-02 | 2024-01-02 | 深圳爱莫科技有限公司 | Image reproduction identification method based on multi-feature fusion |
| CN117333762B (en) * | 2023-12-02 | 2024-03-15 | 深圳爱莫科技有限公司 | Image reproduction identification method based on multi-feature fusion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zheng et al. | A survey on image tampering and its detection in real-world photos | |
| Rahmouni et al. | Distinguishing computer graphics from natural images using convolution neural networks | |
| Jia et al. | Saliency-based deep convolutional neural network for no-reference image quality assessment | |
| Hsu et al. | Detecting image splicing using geometry invariants and camera characteristics consistency | |
| CN109446990B (en) | Method and apparatus for generating information | |
| CN111368764B (en) | False video detection method based on computer vision and deep learning algorithm | |
| CN112150450B (en) | Image tampering detection method and device based on dual-channel U-Net model | |
| CN112257709A (en) | Signboard photo auditing method and device, electronic equipment and readable storage medium | |
| CN110503099B (en) | Information identification method based on deep learning and related equipment | |
| CN112258481A (en) | Portal photo reproduction detection method | |
| JP2001325593A (en) | Duplicate picture detecting method in automatic albuming system | |
| Al-Dhabi et al. | Deepfake video detection by combining convolutional neural network (cnn) and recurrent neural network (rnn) | |
| KR20150128510A (en) | Apparatus and method for liveness test, and apparatus and method for image processing | |
| TWI803243B (en) | Method for expanding images, computer device and storage medium | |
| Zhao et al. | Source camera identification via low dimensional PRNU features | |
| CN117975577A (en) | Deep forgery detection method and system based on facial dynamic integration | |
| Guo et al. | WDXI: The dataset of X-ray image for weld defects | |
| CN116383791A (en) | Customer identity authentication method, device, electronic equipment and medium | |
| Rai et al. | An end-to-end real-time face identification and attendance system using convolutional neural networks | |
| Li et al. | An improved PCB defect detector based on feature pyramid networks | |
| CN113268870A (en) | Monte Carlo-based image recognition reliability evaluation method under outdoor environment condition | |
| CN112818774A (en) | Living body detection method and device | |
| CN112418020A (en) | Attention mechanism-based YOLOv3 illegal billboard intelligent detection method | |
| Wang et al. | Deep-learning-based block similarity evaluation for image forensics | |
| CN112199976A (en) | Certificate picture generation method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |