CN113129261A - Image tampering detection method based on double-current convolutional neural network - Google Patents

Image tampering detection method based on double-current convolutional neural network Download PDF

Info

Publication number
CN113129261A
CN113129261A CN202110266702.4A CN202110266702A CN113129261A CN 113129261 A CN113129261 A CN 113129261A CN 202110266702 A CN202110266702 A CN 202110266702A CN 113129261 A CN113129261 A CN 113129261A
Authority
CN
China
Prior art keywords
neural network
image
convolutional neural
double
feature
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
Application number
CN202110266702.4A
Other languages
Chinese (zh)
Inventor
毕秀丽
刘延彬
肖斌
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN202110266702.4A priority Critical patent/CN113129261A/en
Publication of CN113129261A publication Critical patent/CN113129261A/en
Pending legal-status Critical Current

Links

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
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/148—Wavelet transforms
    • 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)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Collating Specific Patterns (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an image tampering detection method based on a double-current convolutional neural network, and relates to the technical fields of digital image processing, computer vision, deep learning and the like. The method comprises the following specific steps: 1) collecting and arranging the disclosed tampered image samples; 2) labeling the tampered image samples to obtain labels of the tampered image samples so as to complete construction of the image tampering data set; 3) training a double-current convolutional neural network by using the collected image tampering data set; 4) and testing other tampered images by using the model obtained by training to obtain the final effect. The method can detect the tampered image in reality by using the model obtained by the training of the double-current convolutional neural network, has practical significance and obtains better detection precision.

Description

Image tampering detection method based on double-current convolutional neural network
Technical Field
The invention relates to an image tampering detection method based on a double-current convolutional neural network, and belongs to the technical fields of digital image processing, computer vision, deep learning and the like.
Background
Images are commonly used information carriers in human society. Studies have shown that human-acquired visual image information weighs approximately 80% of human-accepted information. Since the advent of photographic technology, providing photographic evidence has been an effective way to strengthen or disprove claims and mitigate ambiguities. When an event, a report or an instruction is handled in a court, an image may be used to enhance or disprove an assertion to provide information, reprimation or innocence, respectively, to the public to release an individual. But all of this applies only to the condition that the content depicted in the image is authentic and that the content of the modified image after the image has been obtained can be traced back to the first day of photography. Today, with the popularity of digital image acquisition devices and image processing software that is very user friendly, it has become an easy task for both professionals and non-professionals to alter image content. And therefore detection of image tampering, becomes increasingly difficult because the tampered image tends to be visually indistinguishable from the real image. The tampered images which cannot be distinguished accurately transmit a large amount of false information, and great harm is caused to individuals and the society.
Deep learning is a rapid development in recent ten years, and a convolutional neural network is a very representative deep learning method. The convolutional neural network can effectively learn the characteristics required by the task from a disordered sample through a series of operations of convolution, normalization, activation, pooling and the like. Because the convolutional neural network has strong feature learning capability, more and more tasks make major breakthroughs (such as classification tasks, semantic segmentation, target detection and the like) after the convolutional neural network method is used, and therefore, the process of researching whether the convolutional neural network can make a breakthrough in the field of image tampering detection is also a necessary process. However, the current tamper detection method based on the convolutional neural network has the following problems: (1) after the image is tampered, if some post-processing hiding measures (such as integral fuzzy operation or noise addition) are correspondingly carried out, the detection method is invalid; (2) the method is only suitable for certain type of tampering, and the generalization capability is poor; (3) minor tampering is not easily detected; (4) generally, the operation is a multi-stage mixing operation, and the time complexity is high.
Disclosure of Invention
In view of the above problems, the present invention is directed to solving the problems of the existing convolution neural network-based tamper detection method. The invention provides an image tampering detection method based on a double-current convolutional neural network, which is an effective method suitable for image tampering detection based on a convolutional neural network. The method basically solves the problem of the prior tampering detection method based on the convolutional neural network, and can effectively and accurately detect the tampered image under the condition of low time complexity.
In order to achieve the purpose, the invention adopts the technical scheme that: an image tampering detection method based on a double-current convolutional neural network comprises the following steps:
(1) and collecting and arranging the disclosed tampered image number samples.
(2) And marking the collected tampered image samples by using Photoshop, obtaining labels of tampered areas of the tampered image samples, and completing construction of a tampered image data set.
(3) Performing model training on a double-current convolutional neural network by using the tampered image data set obtained after the processing of the step (2), wherein the double-current convolutional neural network comprises an equipment fingerprint flow module, an operation chain fingerprint flow module and a global feature extraction module; the device fingerprint flow module extracts a device fingerprint feature vector of an input image by using a decoder of a U-Net network; the operation chain fingerprint flow module extracts an operation chain fingerprint feature vector of an input image by using three-order discrete wavelet transform; and the global feature extraction module is used for processing by three convolution kernel branches respectively to obtain global feature vectors.
(4) And (4) testing the input image by using the model obtained by training in the step (3) to obtain a final detection result.
Further, the device fingerprint flow module performs convolution operation with convolution kernel size of 3 × 3 and step size of 1 twice on the input image, and then performs convolution operation with convolution kernel size of 3 × 3 and step size of 2 once again to down-sample the feature map; repeating for four times to finally obtain the device fingerprint feature vector with 512 feature maps.
Further, the operation chain fingerprint flow module executes third-order Haar discrete wavelet transform on each color channel of the input image, and obtains the operation chain fingerprint feature vector with 36 feature maps through third-order wavelet decomposition.
Further, in the three branches of the global feature extraction module, the convolution kernel size of each branch is respectively 3 × 3, 5 × 5 and 7 × 7, the step length is 1, convolution operations are respectively performed, feature maps obtained by the three branches are spliced together, and a feature vector with 512 feature maps is obtained through the convolution operation of 1 × 1.
The invention adopts the scheme to have the advantages and beneficial effects as follows:
the invention realizes the task of detecting the tampered image by utilizing the technologies of digital image processing, computer vision, deep learning and the like. The invention is an end-to-end method, and a tampered area can be obtained by inputting a tampered image without any preprocessing and post-processing operation. The invention has the following advantages:
(1) the PyCharm platform is used for training and testing, so that the cost is low;
(2) the method is an end-to-end network, and a tampered area can be obtained by inputting a tampered image without any other operation;
(3) the invention extracts the device fingerprint and the operation chain fingerprint, and the two characteristic vectors have synergistic effect, so the invention can also have better detection effect on some tampered images after post-processing (such as integral fuzzy operation or noise addition).
(4) The invention provides a global feature extraction module to enhance the global view of the network, so that the network can make decisions on the global view, and therefore, the invention is also effective for slightly tampered images or tampered images with a plurality of tampered areas.
(5) The accuracy is high, and the testing accuracy is about 86%;
(6) the method can assist relevant image forensics or image safety work, further reduce adverse effects of tampered images on individuals and the society, has practical significance, and achieves better effect.
Drawings
FIG. 1 is a system flow diagram of the present invention;
FIG. 2 is a tamper image 1 and its label;
FIG. 3 is a predicted result image of FIG. 2 obtained via a dual-stream convolutional neural network;
FIG. 4 is a tamper image 2 and its label;
fig. 5 is a prediction result image obtained by the dual-stream convolutional neural network of fig. 4.
Detailed Description
The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.
The technical scheme of the invention is explained in detail in the following with the accompanying drawings:
as shown in fig. 1, a system flowchart is a tamper detection method based on a dual-stream convolutional neural network, including the following steps:
the first step is as follows: collecting and sorting open samples of tampered images;
the second step is that: labeling the collected tampered image samples by using Photoshop to obtain labels of tampered areas of the tampered image samples and finish the construction of a tampered image data set, so that a double-current convolution neural network is trained;
the third step: training the double-current convolution nerves by using the constructed data set; this network is mainly composed of three parts: the device fingerprint stream, the operation chain fingerprint stream and the global feature extraction module.
In the device fingerprint stream, the invention uses the decoder part of the U-Net network to extract the device fingerprint characteristic vector F of the input images，Fs512 feature maps output by the decoder part of the U-Net network. Specifically, the feature map is downsampled by performing convolution operation twice with a convolution kernel size of 3 × 3 and a step size of 1 on the input image, and then performing convolution operation once again with a convolution kernel size of 3 × 3 and a step size of 2. The above steps are repeated four times, and 512 characteristic maps are finally obtained.
In an operation chain fingerprint stream, the invention uses three-order discrete wavelet transform to extract input imagesIs the operation chain fingerprint feature vector Fc，Fc36 characteristic graphs output by the third-order discrete wavelet transform. Specifically, a third-order Haar discrete wavelet transform is performed on each color channel of the input image, and 36(3 × 12) feature maps (12 (3 × 4) feature maps at each order) are obtained by a third-order wavelet decomposition.
In the global feature extraction module, the invention designs a global feature extraction module based on human visual decision (in order to observe the local part of an object, people approach the object, and in contrast, in order to obtain a global view, people are far away from the object). The module has three branches, each having convolution kernels of 3 × 3, 5 × 5 and 7 × 7 (step size is 1), and these convolution kernels of different sizes represent different positions where a person looks at an object. These three branches produce 3 different global feature maps, specifically, branches with a 3 × 3 convolution kernel are more sensitive to small region tampering, branches with a 5 × 5 convolution kernel are more sensitive to medium-sized tampering regions, and branches with a 7 × 7 convolution kernel are more sensitive to large region or cross-region tampering. Finally, we splice the feature maps obtained from the three branches and obtain 512 feature vectors F with global features by 1 × 1 convolution operationg。
The fourth step: and testing other tampered images (images used in the non-training process) by using the trained network model to obtain a final detection effect, wherein the accuracy rate can reach about 86%.
The specific training process of the double-current convolutional neural network comprises the following steps:
1) sending the training data set obtained in the step 2) into a double-current convolution neural network as input;
2) for a device fingerprint stream, 512 feature maps are finally obtained by passing an input image through a plurality of assemblies consisting of a convolutional layer, a normalization layer and an activation layer.
The convolutional layer is a process of convolving an input image by using k convolution kernels to generate k new feature maps for subsequent processing. Jth output profile for nth layer
Figure BDA0002972296620000041
We can define the convolution operation as:
Figure BDA0002972296620000042
wherein
Figure BDA0002972296620000043
And
Figure BDA0002972296620000044
the convolution kernel and the offset are represented separately,
Figure BDA0002972296620000045
and
Figure BDA0002972296620000046
respectively representing an output characteristic diagram and an input characteristic diagram;
the convolution layer is followed by a normalization layer, batch normalization is used, the batch normalization can accelerate the training speed of the network and avoid overfitting in the training process, and two learnable hyper-parameters (gamma, beta) are introduced to disturb training data and prevent training from shifting. By which the minimum batch B of size m ═ x1,x2,...,xmEvery data item x iniIs converted into yi:
Figure BDA0002972296620000047
Figure BDA0002972296620000048
Wherein EM(xi) And VarM(xi) Respectively represent x in batch BiThe mean and variance of (c), epsilon, are small positive numbers used to avoid a divisor of 0.
The normalization layer is followed by an activation layer that transforms the input feature map by a non-linear mapping. We actually use the Relu activation function, which is defined as:
f(zi)＝max(zi,0)
wherein z isiAs a result of the convolution operation, when zi<0, f (z)i) When z is 0i>0, f (z)i)＝zi；
3) For the operation chain fingerprint flow, each color channel of the input image is subjected to three-order Haar discrete wavelet transform to obtain 36 characteristic maps. It is worth noting that the device fingerprint stream and the operation chain fingerprint stream are trained synchronously;
4) on the basis of the step 3), the device fingerprint feature vector F is usedsAnd operation chain fingerprint feature vector FcSplicing along the channel direction into a feature vector F with 548 feature mapshFinally expressed as Fh＝＜Fs,Fc> (ii). After which we use a 1 × 1 convolution operation to convert FhBecomes a feature vector F having 512 feature mapsh1；
5) On the basis of step 4), Fh1Inputting the data into a global feature extraction module, and respectively obtaining feature vectors F with 512 feature maps through three branches of the global feature module3、F5And F7Then, they are spliced along the channel direction to obtain 1536 eigenvectors F with global characteristicsgFinally F is convolved with 1 × 1gBecomes a feature vector F having 512 feature mapsg1。
6) Based on step 5), we finally map F with global featuresg1And decoding by a decoder of the U-Net network to obtain a final prediction result. The whole double-current convolutional neural network optimizes the predicted result by minimizing a cross entropy loss function through a random gradient descent optimization algorithm, wherein the cross entropy loss function is as follows:
Figure BDA0002972296620000049
wherein p isiAnd miRespectively, the predicted result and the actual label of the ith input image of the network, and N represents the total number of samples.
To verify the effect of the present invention, the following experiment was performed:
a verification experiment is carried out on a computer, and the configuration of the computer comprises an Intel i7-8700 six-core processor, a 16GB memory, a video card nvidia GeForce RTX 2070(8GB), and a platform of Pycharm and Photoshop CS 6.
The experimental method comprises the following steps:
in the experimental process, a plurality of public tampered images are collected and sorted, 80% of the images are used as a training set to train the double-current convolutional neural network, and 20% of the images are used as a test set to test the detection accuracy of the double-current convolutional neural network.
The first step is as follows: all the collected tampered images are manually labeled by using Photoshop CS6 software, namely, a black-and-white image which represents a tampered area of each tampered image is manufactured, wherein white parts represent the tampered areas, and black parts represent non-tampered areas.
The second step is that: and (3) using a PyCharm running program, inputting the training set pictures and the corresponding labels into a double-current convolutional neural network, and performing 100 times of iterative training to obtain a finally trained model.
The third step: and testing the images in the test set by using the trained model and calculating the detection accuracy according to the corresponding labels. Specifically, after the tampered image 1 and the tampered area 2 are input separately (as shown in fig. 2 and 4 (left)), and the trained model is detected, the obtained detection results are shown in fig. 3 and 5, and the predicted tampered area is marked as white, and the non-tampered area is marked as black.
Therefore, after the double-current convolutional neural network is trained, the tampered region of the tampered image can be effectively detected, and the detection accuracy can reach about 86%.
The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.

Claims (9)

1. An image tampering detection method based on a double-current convolutional neural network is characterized by comprising the following steps:
(1) collecting and sorting open samples of tampered images;
(2) labeling the collected tampered image samples by using Photoshop to obtain labels of tampered areas of the tampered image samples and finish the construction of a tampered image data set;
(3) performing model training on a double-current convolutional neural network by using the tampered image data set obtained after the processing of the step (2), wherein the double-current convolutional neural network comprises an equipment fingerprint flow module, an operation chain fingerprint flow module and a global feature extraction module;
(4) and (4) testing the input image by using the model obtained by training in the step (3) to obtain a final detection result.
2. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 1, characterized in that: the device fingerprint flow module extracts a device fingerprint feature vector of an input image by using a decoder of a U-Net network; the operation chain fingerprint flow module extracts an operation chain fingerprint feature vector of an input image by using three-order discrete wavelet transform; and the global feature extraction module is used for processing by three convolution kernel branches respectively to obtain global feature vectors.
3. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 2, characterized in that: the device fingerprint flow module performs convolution operation with convolution kernel size of 3 multiplied by 3 and step length of 1 twice on an input image, and then performs convolution operation with convolution kernel size of 3 multiplied by 3 and step length of 2 once again to sample a feature map down; repeating for four times to finally obtain the device fingerprint feature vector with 512 feature maps.
4. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 2, characterized in that: the operation chain fingerprint flow module executes three-order Haar discrete wavelet transform on each color channel of the input image, and obtains operation chain fingerprint characteristic vectors with 36 characteristic diagrams through three-order wavelet decomposition.
5. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 2, characterized in that: the sizes of convolution kernels of the three branches of the global feature extraction module are respectively 3 x 3, 5 x 5 and 7 x 7, the step lengths are all 1, convolution operation is respectively carried out, feature maps obtained by the three branches are spliced, and feature vectors with 512 feature maps are obtained through the convolution operation of 1 x 1.
6. The image tampering detection method based on the double-current convolutional neural network as claimed in any one of claims 1-5, characterized in that: the step (3) of performing model training on the double-current convolutional neural network comprises the following steps:
1) sending the training data set as input into a double-current convolution neural network;
2) for the device fingerprint stream, the input image passes through a plurality of assemblies consisting of a convolution layer, a normalization layer and an activation layer to finally obtain a device fingerprint feature vector F with 512 feature mapss；
3) For the operation chain fingerprint flow, each color channel of the input image is subjected to three-order Haar discrete wavelet transform to obtain an operation chain fingerprint feature vector F with 36 feature mapsc；
4) Fingerprint feature vector F of equipmentsAnd operation chain fingerprint feature vector FcSplicing along the channel direction into a feature vector F with 548 feature mapshFinally expressed as Fh＝＜Fs,Fc>. thereafter F is convolved using 1 × 1 convolutionhBecomes a feature vector F having 512 feature mapsh1；
5) F is to beh1Inputting the data into a global feature extraction module, and respectively obtaining feature vectors F with 512 feature maps through three branches of the global feature module3、F5And F7Then, they are spliced along the channel direction to obtain 1536 eigenvectors F with global characteristicsgFinally F is convolved with 1 × 1gBecomes a feature vector F having 512 feature mapsg1；
6) Handle Fg1And decoding by a decoder of the U-Net network to obtain a final prediction result.
7. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 6, wherein: the device fingerprint stream of step 2) and the operation chain fingerprint stream of step 3) are trained synchronously.
8. The image tampering detection method based on the double-current convolutional neural network as claimed in claim 6, wherein:
the convolution layer uses k convolution cores to perform convolution on the input image, generates k new feature maps for the subsequent processing, and outputs the jth output feature map of the nth layer
Figure FDA0002972296610000021
The convolution operation is:
Figure FDA0002972296610000022
wherein
Figure FDA0002972296610000023
And
Figure FDA0002972296610000024
the convolution kernel and the offset are represented separately,
Figure FDA0002972296610000025
and
Figure FDA0002972296610000026
respectively representing an output characteristic diagram and an input characteristic diagram;
the convolutional layer is followed by a normalization layer, using batch normalization to introduce two learnable hyper-parameters { γ, β } to ensure that the feature distribution is not corrupted, and a minimum batch B of size m ═ x1,x2,...,xmEvery data item x iniIs converted into yi:
Figure FDA0002972296610000027
Figure FDA0002972296610000028
Wherein EM(xi) And VarM(xi) Respectively represent x in batch BiThe mean and variance of (e), epsilon, are small positive numbers used to avoid divisor 0;
an activation layer follows the normalization layer and transforms the input feature map by a non-linear mapping.
9. The image tampering detection method based on the dual-flow convolutional neural network as claimed in claim 7 or 8, characterized in that: the whole double-current convolutional neural network optimizes the predicted result by minimizing a cross entropy loss function through a random gradient descent optimization algorithm, wherein the cross entropy loss function is as follows:
Figure FDA0002972296610000029
wherein p isiAnd miRespectively, the predicted result and the actual label of the ith input image of the network, and N represents the total number of samples.
CN202110266702.4A 2021-03-11 2021-03-11 Image tampering detection method based on double-current convolutional neural network Pending CN113129261A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110266702.4A CN113129261A (en) 2021-03-11 2021-03-11 Image tampering detection method based on double-current convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110266702.4A CN113129261A (en) 2021-03-11 2021-03-11 Image tampering detection method based on double-current convolutional neural network

Publications (1)

Publication Number Publication Date
CN113129261A true CN113129261A (en) 2021-07-16

Family

ID=76772909

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110266702.4A Pending CN113129261A (en) 2021-03-11 2021-03-11 Image tampering detection method based on double-current convolutional neural network

Country Status (1)

Country Link
CN (1) CN113129261A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113920094A (en) * 2021-10-14 2022-01-11 厦门大学 Image tampering detection technology based on gradient residual U-shaped convolution neural network
CN114764858A (en) * 2022-06-15 2022-07-19 深圳大学 Copy-paste image recognition method, device, computer device and storage medium
CN116343308A (en) * 2023-04-04 2023-06-27 湖南交通工程学院 Fused face image detection method, device, equipment and storage medium
CN117557562A (en) * 2024-01-11 2024-02-13 齐鲁工业大学(山东省科学院) Image tampering detection method and system based on double-flow network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110378254A (en) * 2019-07-03 2019-10-25 中科软科技股份有限公司 Recognition methods, system, electronic equipment and the storage medium of vehicle damage amending image trace
CN112381775A (en) * 2020-11-06 2021-02-19 厦门市美亚柏科信息股份有限公司 Image tampering detection method, terminal device and storage medium

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110378254A (en) * 2019-07-03 2019-10-25 中科软科技股份有限公司 Recognition methods, system, electronic equipment and the storage medium of vehicle damage amending image trace
CN112381775A (en) * 2020-11-06 2021-02-19 厦门市美亚柏科信息股份有限公司 Image tampering detection method, terminal device and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XIULI BI等: ""D-Unet: A Dual-encoder U-Net for Image Splicing Forgery Detection and Localization"", 《HTTPS://ARXIV.ORG/ABS/2012.01821》 *
毕秀丽等: "基于级联卷积神经网络的图像篡改检测算法", 《电子与信息学报》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113920094A (en) * 2021-10-14 2022-01-11 厦门大学 Image tampering detection technology based on gradient residual U-shaped convolution neural network
CN114764858A (en) * 2022-06-15 2022-07-19 深圳大学 Copy-paste image recognition method, device, computer device and storage medium
CN116343308A (en) * 2023-04-04 2023-06-27 湖南交通工程学院 Fused face image detection method, device, equipment and storage medium
CN116343308B (en) * 2023-04-04 2024-02-09 湖南交通工程学院 Fused face image detection method, device, equipment and storage medium
CN117557562A (en) * 2024-01-11 2024-02-13 齐鲁工业大学(山东省科学院) Image tampering detection method and system based on double-flow network
CN117557562B (en) * 2024-01-11 2024-03-22 齐鲁工业大学(山东省科学院) Image tampering detection method and system based on double-flow network

Similar Documents

Publication Publication Date Title
CN113129261A (en) Image tampering detection method based on double-current convolutional neural network
Bayar et al. Design principles of convolutional neural networks for multimedia forensics
Wang et al. FE-YOLOv5: Feature enhancement network based on YOLOv5 for small object detection
Zhao et al. On combining multiscale deep learning features for the classification of hyperspectral remote sensing imagery
Zhang et al. Multi-scale signed recurrence plot based time series classification using inception architectural networks
CN113065578B (en) Image visual semantic segmentation method based on double-path region attention coding and decoding
CN115018824A (en) Colonoscope polyp image segmentation method based on CNN and Transformer fusion
Abdallah et al. Deep learning model based on ResNet-50 for beef quality classification
CN111950457A (en) Oil field safety production image identification method and system
Alshehri A content-based image retrieval method using neural network-based prediction technique
Ansari et al. Review of deep learning techniques for object detection and classification
Uma et al. Copy-move forgery detection of digital images using football game optimization
KR20210039572A (en) Method for classifying stratum using neural network model and device for the same method
Shit et al. An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection
Rakowski et al. Hand shape recognition using very deep convolutional neural networks
Ashrafi et al. Development of image dataset using hand gesture recognition system for progression of sign language translator
Junayed et al. Incept-N: a convolutional neural network based classification approach for predicting nationality from facial features
CN108154107B (en) Method for determining scene category to which remote sensing image belongs
Kuang et al. A spectral-spatial attention aggregation network for hyperspectral imagery classification
Wu et al. A Local-Global Attention Fusion Framework with Tensor Decomposition for Medical Diagnosis
Shanqing et al. A multi-level feature weight fusion model for salient object detection
CN114677332A (en) Depth feature fusion image tampering detection method based on JPEG fingerprints
Ma et al. Frequency Decomposition-Driven Unsupervised Domain Adaptation for Remote Sensing Image Semantic Segmentation
CN114332493A (en) Cross-dimension interactive significance detection model and detection method thereof
CN107317846A (en) Vehicle management system based on cloud platform

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20210716

RJ01 Rejection of invention patent application after publication