CN113609954B - Social network image source identification method and system based on deep learning - Google Patents
Social network image source identification method and system based on deep learning Download PDFInfo
- Publication number
- CN113609954B CN113609954B CN202110876069.0A CN202110876069A CN113609954B CN 113609954 B CN113609954 B CN 113609954B CN 202110876069 A CN202110876069 A CN 202110876069A CN 113609954 B CN113609954 B CN 113609954B
- Authority
- CN
- China
- Prior art keywords
- picture
- training
- camera
- fingerprint
- image
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000013135 deep learning Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000012360 testing method Methods 0.000 claims abstract description 43
- 238000007476 Maximum Likelihood Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 9
- 230000006855 networking Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 239000000654 additive Substances 0.000 claims description 6
- 230000000996 additive effect Effects 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 244000089409 Erythrina poeppigiana Species 0.000 description 1
- 235000009776 Rathbunia alamosensis Nutrition 0.000 description 1
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Images
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a social network image source identification method and system based on deep learning, which adopts a maximum likelihood estimation mode to estimate a fingerprint set of a camera to which a test picture data set belongs, completes the training of a network model according to the noise fingerprint characteristics of a picture and the camera fingerprint characteristics of a camera source thereof, realizes the identification of the social network image source by utilizing the correlation coefficient of a noise picture value and the estimated camera fingerprint, and calculates the identification accuracy. The problem of low accuracy of camera source identification in a traditional single mobile evidence obtaining or social network evidence obtaining method is solved.
Description
Technical Field
The method belongs to the field of deep learning and digital evidence obtaining, and relates to a social network image source identification method and system based on deep learning.
Background
In recent years, with the rise of social networking platforms such as Twitter, facebook, weChat, instagram, sina microblog and the like, online social networking has been widely applied to our daily life, and our life and communication modes are being changed. The smart phone plays an important role in the behavior of publishing and sharing multimedia content (pictures, videos and the like) by a user through a social platform, and provides a medium for fraudulent activities. Therefore, the evidence obtaining method combining the smart phone and the social network platform becomes a research hotspot of digital evidence obtaining, and particularly plays a great role in collecting crime evidence for law enforcement officers.
One common problem with digital forensics is device source identification of images. Camera sensor fingerprints are not only successfully applied for camera recognition and image processing, but also involve camera-based blind image clustering problems. In the online social network, images are shot through a smart phone and uploaded to a social network platform, and the images can be shared with other users after being successfully uploaded so as to be checked and downloaded. Currently, most social networking platforms allow users to take images using smartphones while uploading. Facebook, weChat and other social network platforms contain a large number of photos, which constitute valuable real-time information sources, and are used for matching smart phones, so that the problem that law enforcement personnel are difficult to obtain evidence from mass social network data is solved.
Due to the wide range and heterogeneity of data information on a social network platform and the difficulty of high computational complexity of image forensics algorithms caused by large-scale data sets, few studies have been made to combine image forensics with social network forensics.
Disclosure of Invention
The invention aims to provide a social network image source identification method and system based on deep learning, so as to overcome the defects of the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a social network image source identification method based on deep learning comprises the following steps:
s1, estimating a fingerprint set Kset of a camera to which a test picture data set belongs by adopting a maximum likelihood estimation mode;
s2, randomly choosing a training picture I from the test picture data set, intercepting a sub-picture I 'from any position in the randomly chosen training picture according to a preset size, simultaneously randomly obtaining one l belongs to {0,1}, when l =1, taking a source camera fingerprint image M of the randomly chosen training picture from a fingerprint set Kset, intercepting a sub-picture M' at the same position as the sub-picture intercepted from any position in the training picture from the source camera fingerprint image M, and taking {1, I ', M' } as a combined pair; when label =0, randomly taking out a source camera fingerprint image which is not the training picture I from the fingerprint set Kset, and intercepting a subgraph M 'with the same size as I' from the image, and taking {0, I ', M' } as a combined pair; inputting the acquired pair of pair of images into a training model for training, and repeatedly randomly drawing training images from the test image data set for training until the correlation coefficient between the noise image value output by the training model and the estimated camera fingerprint reaches a set threshold value, thereby completing model training;
and S3, processing the social network image to be verified by adopting the trained model to obtain a noise image value of the social network image to be verified, wherein the camera corresponding to the PRNU value closest to the obtained noise image value is an image source of the social network image to be verified.
Further, the test picture data set is a known camera shot.
Further, estimating a PRNU value of a camera to which the picture data set belongs to be K by adopting a maximum likelihood estimation mode:
W k for testing the noise residual of the k picture in the picture data set, I k The noise picture of the kth picture in the test picture data set is obtained.
Further, the training model adopts a CSI-CNN network structure, and comprises an input layer, a convolutional layer and an output layer.
Further, the input layer is convolved by using 128 × 3 convolution kernels with the size of 3 × 3, and nonlinear output among neurons is realized by using a ReLU activation function; the convolution layer uses a bottleneck residual module of a residual network, sequentially performs convolution through convolution kernels of 1 × 1 × 128, 3 × 3 × 32 and 1 × 1 × 32, and executes batch normalization and a ReLU activation function to output a 128-dimensional feature matrix; the front M layers of the convolutional layers obtain images with additive noise removed through characteristic learning of images in a training set and serve as input of subsequent layers of the convolutional layers, and the subsequent layers of the convolutional layers obtain noise image values of the input images through characteristic learning of the images with the additive noise removed; the output layer outputs an image multiplicative noise fingerprint using a convolution kernel of size 3 x 128.
Further, the threshold value was set to 80%.
Further, calculating a correlation coefficient loss (x, y, l) between the output noise picture and the estimated camera fingerprint by using the acquired combined pair input loss function:
wherein,
when l =1, a larger correlation coefficient ρ (x, y) indicates a smaller loss, indicating that the input x picture comes out of the camera to which the PRNU value y belongs.
Further, dividing the test picture data set into a verification set, a fingerprint estimation set, a training set and a test set according to a proportion;
the fingerprint estimation set is a picture set, and camera fingerprints are estimated according to a maximum likelihood estimation method and a wavelet filtering method;
the training set is used for training the network by using the input training set and the camera fingerprint value, so that the trained network outputs a noise picture of the downloaded picture, and a correlation coefficient can be calculated with the camera fingerprint K to determine a camera source to which the downloaded picture belongs;
the verification set is used for verifying whether the correlation coefficient of the noise image output by the network and the camera fingerprint accords with the actual correlation after the trained network is input.
A social network image source identification system comprises a pre-training module and an identification module;
the pre-training module is used for estimating a fingerprint set Kset of a camera to which the test picture data set belongs, randomly choosing a training picture I from the test picture data set, intercepting a subgraph I 'from any position in the randomly chosen training picture according to a preset size, and simultaneously randomly obtaining one element from {0,1}, when l =1, taking a source camera fingerprint image M of a randomly chosen training picture from the fingerprint set Kset, intercepting a subgraph M' at the same position as an intercepted subgraph at any position in the training picture from the source camera fingerprint image M, and taking {1, I ', M' } as a combined pair; when label =0, randomly taking out a source camera fingerprint image which is not the training picture I from the fingerprint set Kset, and intercepting a subgraph M 'with the same size as I' from the image, and taking {0, I ', M' } as a combined pair; inputting the acquired pair of Pair combination into a training model for training, repeatedly randomly selecting a training picture from a test picture data set for training until a correlation coefficient between a noise picture value output by the training model and an estimated camera fingerprint reaches a set threshold value, and transmitting the trained model to an identification module;
the identification module is used for processing the social network image to be verified according to the trained model to obtain a noise image value of the social network image to be verified, and the camera corresponding to the PRNU value closest to the obtained noise image value is an image source of the social network image to be verified.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention relates to a social network image source identification method based on deep learning, which adopts a maximum likelihood estimation mode to estimate a fingerprint set of a camera to which a test picture data set belongs, completes the training of a network model according to the noise fingerprint characteristics of a picture and the camera fingerprint characteristics of a camera source, realizes the identification of the social network image source by utilizing the correlation coefficient of a noise picture value and the estimated camera fingerprint, and calculates the identification accuracy. The problem of low accuracy of camera source identification in a traditional single mobile evidence obtaining or social network evidence obtaining method is solved.
Furthermore, the CSI-CNN model has higher accuracy in camera source identification of the original picture and various compressed pictures on the social network, and the accuracy is remarkably improved compared with that of a classical traditional algorithm wavelet filtering and a DnCNN camera source identification algorithm.
Furthermore, the method is suitable for detecting the false forged picture information issued by the social network, and the real picture and the false forged picture are identified by extracting the noise fingerprint of the picture issued by an account on the social network and comparing the extracted noise fingerprint.
Drawings
FIG. 1 is a comparison graph of the results of the camera recognition accuracy of the original picture by using the method of the present invention and the Wavelet and DnCNN methods.
FIG. 2 shows the accuracy of picture camera sources on Twitter, wechat and WeiBo three social networking platforms by using the network models CSI-CNN, wavelet and DnCNN of the invention in an Our data set.
FIG. 3 is a comparison graph of camera recognition accuracy rates of images downloaded by a Kaggle data set on Twitter, weiBo and WeChat social network platforms by using the network model CSI-CNN, wavelet and DnCNN methods of the invention.
FIG. 4 is an ROC curve obtained by downloading a picture set on a Twitter social network platform by using the method of the present invention and the Wavelet and DnCNN methods.
FIG. 5 is an ROC curve obtained by downloading a set of pictures on a WeiBo social network platform using the method of the present invention and the Wavelet and DnCNN methods.
FIG. 6 is an ROC curve obtained by downloading a set of pictures on a WeChat social network platform by using the method of the present invention and the Wavelet and DnCNN methods.
Fig. 7 is a flow chart of the method of the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
as shown in fig. 7, a social network image source identification method based on deep learning includes the following steps:
s1, estimating a fingerprint set Kset of a camera to which a test picture data set belongs by adopting a maximum likelihood estimation mode;
specifically, the test picture data set is a picture taken by a known camera, the test picture data set is a picture taken by the known camera, the PRNU value of the picture taken by the known camera is known, and the test picture data set can be divided into a verification set, a fingerprint estimation set, a training set and a test set according to the proportion division, wherein the proportion division is that 1.
And estimating the fingerprint set Kset of the camera to which the picture data set belongs by adopting a mode of maximum likelihood estimation on the fingerprint estimation set divided by the test picture data set. Estimating the PRNU value of a camera to which the picture data set belongs to be K by adopting a maximum likelihood estimation mode:
W k for testing the noise residual, I, of the kth picture in the picture data set k The noise picture of the kth picture in the test picture data set is obtained.
S2, randomly choosing a training picture I from the test picture data set, intercepting a sub-picture I 'from any position in the randomly chosen training picture according to a preset size, simultaneously randomly obtaining one l belongs to {0,1}, when l =1, taking a source camera fingerprint image M of the randomly chosen training picture from a fingerprint set Kset, intercepting a sub-picture M' at the same position as the sub-picture intercepted from any position in the training picture from the source camera fingerprint image M, and taking {1, I ', M' } as a combined pair; when label =0, randomly taking out a source camera fingerprint image which is not the training picture I from the fingerprint set Kset, intercepting a subgraph M 'with the same size as I' from the image, and taking {0, I ', M' } as a combined pair; inputting the acquired pair of combination pairs into a training model for training, and repeatedly randomly selecting training pictures from a test picture data set for training until the correlation coefficient between the noise picture value output by the training model and the estimated camera fingerprint reaches a set threshold value, thereby completing model training;
specifically, the training model adopts a CSI-CNN network structure; the device specifically comprises an input layer, a convolution layer and an output layer, wherein the input layer performs convolution by adopting 128 × 3 convolution kernels with the size of 3 × 3, and nonlinear output among neurons is realized by using a ReLU activation function; the convolution layer uses a Bottleneck Residual Block (bottle Residual Block) of a Residual network, sequentially performs convolution through convolution kernels of 1 × 1 × 128, 3 × 3 × 32 and 1 × 1 × 32, and executes batch normalization and a ReLU activation function to output a 128-dimensional feature matrix; the front M layers of the convolutional layers obtain images with additive noise removed through feature learning of images in the training set and serve as input of subsequent layers of the convolutional layers, and the subsequent layers of the convolutional layers obtain noise image values of the input images through feature learning of the images with the additive noise removed; the output layer outputs the image multiplicative noise fingerprint W using a convolution kernel of size 3 × 3 × 128.
The trained model obtains the noise image value of the input image by adopting a test set test model, and if the correlation coefficient between the obtained noise image value and the test set fingerprint image does not reach the set threshold value, the model is repeatedly trained by adopting the method until the correlation coefficient between the obtained noise image value and the test set fingerprint image reaches the set threshold value; the threshold is set to 80% in the present application.
Calculating a correlation coefficient loss (x, y, l) between the noise picture output by the acquired combined pair input loss function and the estimated camera fingerprint:
wherein,
when l =1, a larger ρ (x, y) correlation coefficient indicates a smaller loss, indicating that the input x picture comes out of the camera to which the PRNU value belongs; when l =0, then ρ (x, y) correlation coefficient is small and loss is also small.
And S3, processing the social network image to be verified by adopting the trained model to obtain a noise image value of the social network image to be verified, wherein the camera corresponding to the PRNU value closest to the obtained noise image value is an image source of the social network image to be verified.
In the imaging process of the camera, the Sensor leaves Sensor Pattern Noise (SPN) in any image taken, which is an inherent feature of the digital camera and mainly consists of Photo Response Non-uniformity (PRNU) and Fixed Pattern Noise (FPN). Even under uniform illumination conditions, due to manufacturing imperfections and non-uniformities in the silicon chips of CCD and CMOS imaging sensors, there will be differences in the output values of the photosites of the same model of sensor, which will produce PRNUs that are unique to the individual sensor, so this application focuses on the uniqueness of the picture PRNU fingerprint.
Example (b): an original picture set F is set, and a picture set which is uploaded to Twiter, wechat and WeiBo and then downloaded is F'. The original picture set is divided into a verification set, a fingerprint estimation set, a training set and a test set according to the proportion of 1.
The fingerprint estimation set is a picture set, and camera fingerprints K are estimated according to a maximum likelihood estimation method and a wavelet filtering method.
The training set is a network trained by using an input original picture and a camera fingerprint K value, so that the trained network outputs a noise picture of a downloaded picture, and a correlation coefficient can be calculated with the camera fingerprint K to determine a camera source to which the downloaded picture belongs; .
The verification set is used for verifying whether the correlation coefficient of the noise image output by the network and the camera fingerprint K accords with the actual correlation after the trained network is input.
The test set is a noise image output by inputting a picture cut from an original picture and with the size of 64 multiplied by 64 and a camera fingerprint K to a trained network, and can calculate a correlation coefficient with the camera fingerprint K to determine a camera source to which a downloaded picture belongs; .
The trained network is applied to a picture set F' downloaded from Twaiter, wechat and WeiBo, and the picture set is divided into a verification set, a fingerprint estimation set, a training set and a test set according to the proportion of 1. The accuracy of the source identification of the network camera trained by the CSI-CNN network structure provided by the invention is more accurate through testing and verification of the trained network.
FIG. 1 shows the accuracy of the Our and Kaggle data sets using the network model CSI-CNN and Wavelet, dnCNN picture camera sources proposed by the present invention. It is obvious from the histogram that the network model provided by the invention has higher recognition rate. Fig. 2 and fig. 3 show the accuracy of image camera sources on Twitter, wechat and WeiBo three social networking platforms by using the network models CSI-CNN, waveet and DnCNN of the present invention in the ourr and Kaggle data sets, respectively. As can be seen from the display results of the histograms in fig. 1, fig. 2 and fig. 3, the CSI-CNN network model proposed by the present invention has higher accuracy in camera source identification of both the original picture and the processed picture downloaded on the social network platform.
Fig. 4, fig. 5, and fig. 6 are ROC curves calculated by the network models CSI-CNN, wavelet, and DnCNN on the picture sets of Twitter, wechat, and WeiBo three social networking platforms, respectively, and it can be seen from the ROC curves that the camera source identification of the network model provided by the present invention on the three platforms has higher accuracy.
The method extracts the characteristics of the original picture data based on the camera fingerprint and the picture noise fingerprint characteristics, further calculates the correlation between the camera fingerprint and the picture noise fingerprint characteristics, trains the network model CSI-CNN, solves the problem that the traditional single mobile evidence obtaining method or social network evidence obtaining method has low accuracy rate on camera source identification, and has higher accuracy rate on identification of the original picture or pictures downloaded on the social network. The image source identification method of the social network based on deep learning solves the problem that the traditional single mobile evidence obtaining or social network evidence obtaining method is low in accuracy rate of camera source identification, and the method provided by the invention has high accuracy rate of identification of original pictures and pictures downloaded on the social network. By combining the PRNU fingerprint with the picture noise fingerprint obtained through deep learning, the accuracy of picture camera source identification on the social network can be greatly improved through the designed loss function and the CSI-CNN network model. And the 64 x 64 pixels randomly cropped from the picture are used as input, and through the expansion of the input data set of the network, enough noise characteristics can be extracted and the overfitting phenomenon can be prevented.
The invention relates to a social network image source identification method based on deep learning, which is characterized in that a CSI-CNN network model is established according to noise fingerprint characteristics of pictures and camera fingerprint characteristics of camera sources of the pictures, noise pictures are output, and the correlation between the noise pictures and a test camera is calculated, so that the identification accuracy is calculated. The model can provide a powerful reference basis for evidence collection of law enforcement officers, and has a very strong application value.
Claims (9)
1. A social network image source identification method based on deep learning is characterized by comprising the following steps:
s1, estimating a fingerprint set Kset of a camera to which a test picture data set belongs by adopting a maximum likelihood estimation mode;
s2, randomly choosing a training picture I from the test picture data set, intercepting a sub-picture I 'from any position in the randomly chosen training picture according to a preset size, simultaneously randomly obtaining one l belongs to {0,1}, when l =1, taking a source camera fingerprint image M of the randomly chosen training picture from a fingerprint set Kset, intercepting a sub-picture M' at the same position as the sub-picture intercepted from any position in the training picture from the source camera fingerprint image M, and taking {1, I ', M' } as a combined pair; when label =0, randomly taking a fingerprint image of the camera not belonging to the training picture I from the fingerprint set Kset, and intercepting a sub-picture M ″ having the same size as I 'from the image, and taking {0, I', M ″ } as a combined pair; inputting the acquired pair of combination pairs into a training model for training, and repeatedly randomly selecting training pictures from a test picture data set for training until the correlation coefficient between the noise picture value output by the training model and the estimated camera fingerprint reaches a set threshold value, thereby completing model training;
and S3, processing the social network image to be verified by adopting the trained model to obtain a noise image value of the social network image to be verified, wherein the camera corresponding to the PRNU value closest to the obtained noise image value is an image source of the social network image to be verified.
2. The deep learning-based social networking image source identification method of claim 1, wherein the test picture data set is a known camera shot picture.
3. The deep learning-based social network image source identification method as claimed in claim 1, wherein a PRNU value of a camera to which the picture data set belongs is estimated as K by means of maximum likelihood estimation:
W k for testing the noise residual, I, of the kth picture in the picture data set k The noise picture of the kth picture in the test picture data set is obtained.
4. The deep learning-based social network image source identification method as claimed in claim 1, wherein the training model adopts a CSI-CNN network structure including an input layer, a convolutional layer and an output layer.
5. The deep learning-based social network image source identification method as claimed in claim 4, wherein the input layer is convolved with 128 × 3 convolution kernels of size 3 × 3, and nonlinear output among neurons is realized by using a ReLU activation function; the convolution layer uses a bottleneck residual module of a residual network, sequentially performs convolution through convolution kernels of 1 × 1 × 128, 3 × 3 × 32 and 1 × 1 × 32, and executes batch normalization and a ReLU activation function to output a 128-dimensional feature matrix; the front M layers of the convolutional layers obtain images with additive noise removed through feature learning of images in the training set and serve as input of subsequent layers of the convolutional layers, and the subsequent layers of the convolutional layers obtain noise image values of the input images through feature learning of the images with the additive noise removed; the output layer outputs an image multiplicative noise fingerprint using a convolution kernel of size 3 x 128.
6. The deep learning based social network image source identification method as claimed in claim 1, wherein the threshold value is set to 80%.
7. The image source identification method for social network based on deep learning of claim 1, wherein the correlation coefficient loss (x, y, l) between the noise picture output by the obtained combined pair input loss function and the estimated camera fingerprint is:
wherein,
when l =1, a larger correlation coefficient ρ (x, y) indicates a smaller loss, indicating that the input x picture comes out of the camera to which the PRNU value y belongs.
8. The image source identification method for the social network based on the deep learning of claim 1, wherein a test picture data set is proportionally divided into a verification set, a fingerprint estimation set, a training set and a test set;
the fingerprint estimation set is used for estimating the camera fingerprint for the picture set according to a maximum likelihood estimation method and a wavelet filtering method;
the training set is a network trained by using the input training set and the camera fingerprint value, so that the trained network outputs a noise picture of the downloaded picture, and a correlation coefficient can be calculated with the camera fingerprint K to determine a camera source to which the downloaded picture belongs;
the verification set is used for verifying whether the correlation coefficient of the noise image output by the network and the camera fingerprint accords with the actual correlation after the trained network is input.
9. A social network image source identification system for use in the social network image source identification method of claim 1, comprising a pre-training module and an identification module;
the pre-training module is used for estimating a fingerprint set Kset of a camera to which the test picture data set belongs, randomly selecting a training picture I from the test picture data set, intercepting a sub-picture I 'from any position in the randomly selected training picture according to a preset size, simultaneously randomly acquiring one l ∈ {0,1}, when l =1, taking a source camera fingerprint image M of the randomly selected training picture from the fingerprint set Kset, intercepting a sub-picture M' at the same position as the sub-picture intercepted at any position in the training picture from the source camera fingerprint image M, and taking {1, I ', M' } as a combination pair; when label =0, randomly taking out the fingerprint image of the camera which is not the training picture I from the fingerprint set Kset, and intercepting a subgraph M 'with the same size as I' from the image, wherein {0, I ', M' } is used as a combined pair; inputting the acquired pair of Pair combination into a training model for training, repeatedly randomly selecting a training picture from a test picture data set for training until a correlation coefficient between a noise picture value output by the training model and an estimated camera fingerprint reaches a set threshold value, and transmitting the trained model to an identification module;
the identification module is used for processing the social network image to be verified according to the trained model to obtain a noise image value of the social network image to be verified, and the camera corresponding to the PRNU value closest to the obtained noise image value is an image source of the social network image to be verified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110876069.0A CN113609954B (en) | 2021-07-30 | 2021-07-30 | Social network image source identification method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110876069.0A CN113609954B (en) | 2021-07-30 | 2021-07-30 | Social network image source identification method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113609954A CN113609954A (en) | 2021-11-05 |
| CN113609954B true CN113609954B (en) | 2022-12-09 |
Family
ID=78306354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110876069.0A Active CN113609954B (en) | 2021-07-30 | 2021-07-30 | Social network image source identification method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113609954B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114240843B (en) * | 2021-11-18 | 2024-09-03 | 支付宝(杭州)信息技术有限公司 | Image detection method, image detection device, computer readable storage medium and electronic device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109658348A (en) * | 2018-11-16 | 2019-04-19 | 天津大学 | The estimation of joint noise and image de-noising method based on deep learning |
| CN110599409B (en) * | 2019-08-01 | 2022-02-22 | 西安理工大学 | Convolutional neural network image denoising method based on multi-scale convolutional groups and parallel |
| CN112418252A (en) * | 2020-12-17 | 2021-02-26 | 绍兴聚量数据技术有限公司 | Source social network identification method based on image-contained social network platform fingerprint |
-
2021
- 2021-07-30 CN CN202110876069.0A patent/CN113609954B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113609954A (en) | 2021-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Cozzolino et al. | Noiseprint: A CNN-based camera model fingerprint | |
| CN109636754B (en) | Extremely-low-illumination image enhancement method based on generation countermeasure network | |
| CN106250877B (en) | Near-infrared face identification method and device | |
| Amerini et al. | Tracing images back to their social network of origin: A CNN-based approach | |
| TWI727548B (en) | Method for face recognition and device thereof | |
| Kang et al. | A context-adaptive SPN predictor for trustworthy source camera identification | |
| CN108875619B (en) | Video processing method and device, electronic equipment and computer readable storage medium | |
| CN108154080B (en) | Method for quickly tracing to source of video equipment | |
| Ahmed et al. | Comparative analysis of a deep convolutional neural network for source camera identification | |
| TWI721786B (en) | Face verification method, device, server and readable storage medium | |
| CN113609954B (en) | Social network image source identification method and system based on deep learning | |
| WO2017177559A1 (en) | Image management method and apparatus | |
| CN111709930A (en) | Pattern noise based picture provenance and tampering identification method | |
| Flor et al. | PRNU-based source camera identification for multimedia forensics | |
| CN111861976B (en) | Digital image source shooting equipment identification method based on hardware fingerprint correlation | |
| Galdi et al. | Exploring new authentication protocols for sensitive data protection on smartphones | |
| Wang et al. | Image source identification using convolutional neural networks in IoT environment | |
| Galdi et al. | Secure user authentication on smartphones via sensor and face recognition on short video clips | |
| WO2022028425A1 (en) | Object recognition method and apparatus, electronic device and storage medium | |
| CN113537088A (en) | Digital image sensor hardware fingerprint matching method and system based on small sample learning | |
| CN108898163A (en) | Information processing method and device, electronic equipment and computer readable storage medium | |
| Nguyen et al. | Face spoofing attack detection based on the behavior of noises | |
| CN111126283A (en) | Rapid in-vivo detection method and system for automatically filtering fuzzy human face | |
| Hadiprakoso | Face anti-spoofing method with blinking eye and hsv texture analysis | |
| Rouhi et al. | Social network forensics through smartphones and shared images |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |