CN111476865A - Image protection method for carrying out image recognition by deep learning neural network - Google Patents
Image protection method for carrying out image recognition by deep learning neural network Download PDFInfo
- Publication number
- CN111476865A CN111476865A CN202010211679.4A CN202010211679A CN111476865A CN 111476865 A CN111476865 A CN 111476865A CN 202010211679 A CN202010211679 A CN 202010211679A CN 111476865 A CN111476865 A CN 111476865A
- Authority
- CN
- China
- Prior art keywords
- image
- deep learning
- photo frame
- size
- input
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/60—Editing figures and text; Combining figures or text
-
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/44—Secrecy systems
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an image protection method for carrying out image recognition facing a deep learning neural network, and relates to the technical field of industrial intelligence information safety. The method comprises the steps of protecting target misleading and protecting non-target misleading, firstly, collecting image data or using the existing data as a training data set, preprocessing data images in the training data set, and converting original images into an input format accepted by a deep learning image classifier model; determining the size of the photo frame according to the size of the original image; and finally, generating a photo frame by using the training data under the determined photo frame size, and putting the preprocessed image into the photo frame to realize the encryption of the image. The user of the encrypted image obtained by the method can easily distinguish the image content, and the method has better user experience.
Description
Technical Field
The invention relates to the technical field of industrial and intelligent information security, in particular to an image protection method for carrying out image recognition facing a deep learning neural network.
Background
Along with the continuous development of scientific technology, the electronic computer technology is more and more widely applied, the distance between people is drawn closer by the arrival of the internet era, a large amount of multimedia information can be easily and quickly transmitted in the network, and the communication between people is facilitated. People experience personal private information while enjoying the convenience of the internet. Particularly in the current data age, the problem is more prominent and the influence range is wider. The development of artificial intelligence enables the life of people to become more transparent, and the privacy infringement means is more concealed and technical, thereby invisibly threatening the data security of people. Particularly, people like to send photos and videos related to the people by various social media or store the photos by cloud disk software and the like, the photos often contain identity information or other sensitive information related to the people, the people often only want to see the photos or the people with specific identities, and the confidentiality of information transmission is very important.
Meanwhile, the research of deep learning also promotes the development of artificial intelligence, and the method is applied to more and more fields, for example, images can be easily identified and detected, and manpower is greatly liberated. As a result, lawbreakers can easily screen out key information from the mass data of the theft, thereby implementing targeted fraudulent theft and the like.
The traditional coping method is to encrypt information, and when viewing the information, a user can see the original information only by decrypting the encrypted information. For image photos, the encryption and decryption of users are too cumbersome, and particularly for mobile end users, the human-computer interaction is not friendly enough. Meanwhile, a user often has a large number of images, and on the premise of not decrypting, the user is difficult to distinguish the images to be searched, and the time and labor are consumed for decrypting all the images, so that a protection method is needed, which can conveniently protect the current image photos, can easily view the content of the current image, and improves the user experience.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide an image protection method for performing image recognition by a deep learning neural network, aiming at the defects of the prior art, and performing image protection for the attack of performing image recognition by using a deep learning model.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: an image protection method for carrying out image recognition facing a deep learning neural network comprises two types of target misleading protection and non-target misleading protection, and comprises the following steps:
step 1, collecting a certain number of images, and marking a data set as a training data set aiming at the content of the images, or using the existing data set as the training data set;
step 2, preprocessing the data images in the training data set, and converting the original images in the training data set into an input format accepted by a deep learning image classifier model;
the preprocessing comprises shape adjustment, size adjustment, data range adjustment and image rotation of the image; the shape of the image is adjusted to adjust the length width proportion of the input image, so that the input data meets the input requirement of the classifier model; the image size adjustment refers to zooming and cutting of the image, and the length and the width of the image are enlarged or reduced, or the image which meets the input size requirement of the deep learning image classifier model is obtained by cutting; the data range adjustment of the image refers to adjusting the pixel value range of the image data, and normalizing the pixel value range to a range suitable for the input of the deep learning image classifier model; the rotation of the image refers to rotating the image by a certain angle, so that the complexity and diversity of input data are increased, and the data quality is improved;
the deep learning image classifier model outputs class labels and the probability of the class labels to which the images belong;
step 3, determining the size of the photo frame according to the size of the original image, namely the pixel width of each edge of the photo frame;
step 4, under the condition of the determined size of the photo frame, generating the photo frame by using the training data, and putting the preprocessed image into the photo frame to realize the encryption of the image;
setting an encrypted image x 'to be composed of an original image and a picture frame, wherein the picture frame is expressed, x' is x · (1-m) + · m, wherein m represents a mask for controlling an image processing area, the size of m is consistent with the size of the encrypted image, and is 1 at a position corresponding to the picture frame, and is 0 at the rest position, so that the original image and the picture frame are synthesized into the encrypted image; the symbol "·" represents pixel-by-pixel multiplication, i.e., pixel values at corresponding positions are multiplied, and the size of the obtained encrypted image is one more frame than that of the original image, i.e., the central area of the image is the original image and the periphery is a photo frame;
the photo frame generated by utilizing the training data is obtained by carrying out gradient descent on an optimization problem, and at the beginning of solving, the photo frame is obtained by randomly sampling in normal distribution, namely the photo frame is initialized to a random number and then is gradually solved through back propagation; for the protection of non-target induction, the photo frame generated by using the training data is solved by the following optimization problem:
min(-loss(f(x·(1-m)+·m)，y))
wherein, loss () is a loss function, f () is a deep learning image classifier model, y is an original label of an image, i.e. a category of the image, and f (x) y argmax (softmax (x));
the target-free induced guard means that for any image input to the deep learning image classifier model, the output image prediction class is no longer the correct class, i.e. f (x') ≠ y;
for image protection with target induction, a photo frame generated by using training data is solved by the following optimization problem:
min loss(f(x·(1-m)+·m)，ytarget)
wherein, ytargetIs the target label to be induced;
protection against object misleading refers to the prediction of an input encrypted image into a particular class by a deep learning image classifier model regardless of the class to which the original image belongs, so that the model considers the input image as non-critical information.
Adopt the produced beneficial effect of above-mentioned technical scheme to lie in: according to the image protection method for image recognition facing the deep learning neural network, the encrypted photo frame can be directly applied only by generating once, so that the calculated amount and the calculating time are greatly reduced, the image protection method has better practicability, after the photo frame is obtained, the image can be completed only by simply splicing images when the image is encrypted, the extremely small calculated amount is occupied, and the image can be decrypted only by removing the photo frame. Meanwhile, the user of the encrypted image obtained by the method can easily distinguish the image content, and better user experience is achieved.
Drawings
Fig. 1 is a flowchart of an image protection method for performing image recognition facing a deep learning neural network according to an embodiment of the present invention;
fig. 2 is a data preprocessing flowchart of an image protection method for performing image recognition by a deep learning neural network according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an encrypted image according to an embodiment of the present invention, where (a) is an original image, (b) is a photo frame, and (c) is an encrypted image.
Detailed Description
The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
In this embodiment, taking a deep learning image classifier model resnet model as an example, the image protection method for performing image recognition by using a deep learning neural network of the present invention is adopted to protect an image that is subjected to image recognition by using the model.
In this embodiment, it is assumed that the state parameters of the model during operation can be obtained, that is, the structure of the resnet model is known, and the parameters of each layer of the resnet model can be obtained, so that when the resnet model predicts a picture, the output of each layer of the model, the finally input category and the probability thereof can be deduced.
In this embodiment, the selected resnet model is denoted by f, and for any preprocessed image x, the resnet model performs a classification task to predict the correct type y of the image x, i.e., f (x) ═ y. This process can be further expressed as: the last but one layer is a logits layer, the output is a 1000-dimensional vector, each dimension number corresponds to the logits of the corresponding category, the 1000-dimensional vector is used as the input of the last layer of softmax layer to obtain a new 1000-dimensional vector corresponding to the probability belonging to each category, and the last renet model outputs the category with the highest probability as the prediction result, namely, f (x) y (logmax (x)).
An image protection method for performing image recognition facing a deep learning neural network, as shown in fig. 1, includes two kinds of protection of target misleading and protection of no target misleading, and includes the following steps:
step 1, collecting a certain number of images, and marking a data set as a training data set aiming at the content of the images, or using the existing data set as the training data set;
in this embodiment, the training data set can be manually constructed in a manner of taking pictures, web crawlers, and the like in real life, and the selected picture needs to conform to the classification field of the resnet model.
The training data set may be obtained from a resnet model, and a representative image corresponding to the model is generated as a data set using known structural parameters of the model.
The training data can also be obtained by directly utilizing the existing network data set, for the resnet model, the training data set is an ImageNet data set, the ImageNet has more than twenty thousand images and more than ten million images so far, and an appropriate number of data sets can be selected as the training set.
Step 2, preprocessing the data images in the training data set, as shown in fig. 2, converting the original images into an input format acceptable by a deep learning image classifier model;
the preprocessing comprises shape adjustment, size adjustment, data range adjustment and image rotation of the image;
the shape of the image is adjusted to adjust the length width proportion of the input image, so that the input data meets the input requirement of the classifier model; for example, the original image aspect ratio is 4: 3, adjustment is needed, and the aspect ratio of the image is adjusted to be 1 by using an interpolation function and other methods: 1. the image size adjustment refers to zooming and cutting of the image, and the length and the width of the image are enlarged or reduced, or the image which meets the input size requirement of the deep learning image classifier model is obtained by cutting; for example, the length and width pixels of the original image are all 300, the size of the image can be adjusted to 224 by using an interpolation method such as bilinear interpolation, cubic spline function and the like, or the image can be directly cut at the center of the image to obtain an image with the size of 224 × 224, and the size of the image conforms to the input of the resnet model.
The data range adjustment of the image refers to adjusting the pixel value range of the image data, and normalizing the pixel value range to a range suitable for the input of the deep learning image classifier model; for example, after the original image is transformed into data, the pixel value interval is [0,255], normalization is required, and the value range of the pixel is adjusted to [ -1,1], i.e., the input interval of resnet. Normalization can be performed using a normal distribution with the mean of the RGB three channels (0.4914,0.4822,0.4465) and the variance (0.2023,0.1994, 0.2010).
The rotation of the image refers to rotating the image by a certain angle, for example, randomly selecting a rotation angle at [ -15,15], 15 refers to rotating the image by 15 degrees to the right, and 15 refers to rotating the image by 15 degrees to the left, and randomly rotating the image can increase the data amount on one hand, and can increase the complexity of image recognition on the other hand, improve the data quality and enhance the generalization of the model.
The deep learning image classifier model outputs class labels and the probability of the class labels to which the images belong;
step 3, determining the size of the photo frame according to the size of the original image, namely the pixel width of each edge of the photo frame;
the protection method only needs the photo frame to encrypt, so that the image can be easily protected, and the larger the photo frame is, the higher the encryption degree of the image is, and meanwhile, the larger the proportion of the photo frame in the image is. In this embodiment, the size of the input image of the resnet model is 224 × 224, and the width of the boundary pixel region where the photo frame is located is 5, that is, the photo frame is located within 5 pixels around the original image, and at this time, the photo frame occupies 8.36% of the area of the encrypted image. When the width of the boundary pixel region where the picture frame is located is 6, the picture frame occupies 9.91% of the original image area.
Step 4, under the condition of the determined size of the photo frame, generating the photo frame by using the training data, and putting the image into the photo frame to realize the encryption of the image;
setting an encrypted image x 'to be composed of an original image and a picture frame, wherein the picture frame is expressed, x' is x · (1-m) + · m, wherein m represents a mask for controlling an image processing area, the size of m is consistent with the size of the encrypted image, and is 1 at a position corresponding to the picture frame, and is 0 at the rest position, so that the original image and the picture frame are synthesized into the encrypted image; the symbol "·" represents pixel-by-pixel multiplication, i.e., pixel values at corresponding positions are multiplied, and the size of the obtained encrypted image is one more frame than that of the original image, i.e., the central area of the image is the original image and the periphery is a photo frame; in this embodiment, the finally obtained encrypted image is shown in fig. 3.
The photo frame generated by utilizing the training data is obtained by carrying out gradient descent on an optimization problem, and at the beginning of solving, the photo frame is obtained by randomly sampling in normal distribution, namely the photo frame is initialized to a random number and then is gradually solved through back propagation;
for the protection of non-target induction, the photo frame generated by using the training data is solved by the following optimization problem:
min(-loss(f(x·(1-m)+·m)，y))
wherein, loss () is a loss function, f () is a deep learning image classifier model, y is an original label of an image, i.e. a category of the image, and f (x) y argmax (softmax (x)); in this embodiment, the loss function is a cross entropy function.
The target-free induced guard means that for any image input to the deep learning image classifier model, the output image prediction class is no longer the correct class, i.e. f (x') ≠ y; for example, for the resnet model originally predicted as category 1, any number other than 1 from 0 to 999 can be predicted after encryption.
The optimization problem increases the loss function value of the encrypted image and the original label y;
for image protection with target induction, a photo frame generated by using training data is solved by the following optimization problem:
min loss(f(x·(1-m)+·m)，ytarget)
wherein, ytargetIs the target label to be induced;
for protection with target misleading, the class of the original image is ignored, the deep learning image classifier model predicts the input encrypted image into a specific class, and the input image is regarded as non-critical information by the model; for example, selecting object class 859, after encryption, regardless of the correct class to which the previous image belongs, the image is considered to be of the object class that makes it unimportant for the molecule to consider the image to be unimportant.
The optimization problem reduces the loss function values between the encrypted image and the target label, making it easier for the encrypted image to be misclassified into target categories.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions and scope of the present invention as defined in the appended claims.
Claims (5)
1. An image protection method for carrying out image recognition facing a deep learning neural network is characterized in that: the method comprises the following steps of target misleading protection and non-target misleading protection:
step 1, collecting a certain number of images, and marking a data set as a training data set aiming at the content of the images, or using the existing data set as the training data set;
step 2, preprocessing the data images in the training data set, and converting the original images in the training data set into an input format accepted by a deep learning image classifier model;
step 3, determining the size of the photo frame according to the size of the original image, namely the pixel width of each edge of the photo frame;
and 4, generating a photo frame by using the training data under the determined photo frame size, and putting the preprocessed image into the photo frame to realize the encryption of the image.
2. The image protection method for image recognition based on the deep learning neural network as claimed in claim 1, wherein:
the goal-free induced protection means that for any image input to the deep learning image classifier model, the output image prediction class is no longer the correct class, i.e. f (x') ≠ y;
the protection with target misleading refers to that the class of the original image is ignored, the deep learning image classifier model predicts the input encrypted image into a specific class, and therefore the model considers the input image as non-critical information.
3. The image protection method for image recognition based on the deep learning neural network as claimed in claim 1, wherein: step 2, the preprocessing comprises shape adjustment, size adjustment, data range adjustment and image rotation of the image; the shape of the image is adjusted to adjust the length width proportion of the input image, so that the input data meets the input requirement of the classifier model; the image size adjustment refers to zooming and cutting of the image, and the length and the width of the image are enlarged or reduced, or the image which meets the input size requirement of the deep learning image classifier model is obtained by cutting; the data range adjustment of the image refers to adjusting the pixel value range of the image data, and normalizing the pixel value range to a range suitable for the input of the deep learning image classifier model; the rotation of the image refers to rotating the image by a certain angle, so that the complexity and diversity of input data are increased, and the data quality is improved;
the deep learning image classifier model outputs class labels to which the images belong and the probability thereof.
4. The image protection method for image recognition based on the deep learning neural network as claimed in claim 1, wherein: the specific method of the step 4 comprises the following steps:
setting an encrypted image x 'to be composed of an original image and a picture frame, wherein the picture frame is expressed, x' is x · (1-m) + · m, wherein m represents a mask for controlling an image processing area, the size of m is consistent with the size of the encrypted image, and is 1 at a position corresponding to the picture frame, and is 0 at the rest position, so that the original image and the picture frame are synthesized into the encrypted image; the symbol "·" represents pixel-by-pixel multiplication, i.e., pixel values at corresponding positions are multiplied, and the size of the obtained encrypted image is one more frame than that of the original image, i.e., the central area of the image is the original image and the periphery is the picture frame.
5. The image protection method for image recognition based on the deep learning neural network as claimed in claim 1, wherein: step 4, generating a photo frame by using the training data, and obtaining the photo frame by performing gradient descent on the optimization problem, wherein the photo frame is obtained by randomly sampling from normal distribution at the beginning of solving, namely the photo frame is initialized to a random number and then is gradually solved by back propagation;
for the protection of non-target induction, the photo frame generated by using the training data is solved by the following optimization problem:
min(-loss(f(x·(1-m)+·m)，y))
wherein, loss () is a loss function, f () is a deep learning image classifier model, y is an original label of an image, i.e. a category of the image, and f (x) y argmax (softmax (x));
for image protection with target induction, a photo frame generated by using training data is solved by the following optimization problem:
min loss(f(x·(1-m)+·m)，ytarget)
wherein, ytargetIs the target label to be induced.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010211679.4A CN111476865B (en) | 2020-03-24 | 2020-03-24 | Image protection method for image recognition based on deep learning neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010211679.4A CN111476865B (en) | 2020-03-24 | 2020-03-24 | Image protection method for image recognition based on deep learning neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111476865A true CN111476865A (en) | 2020-07-31 |
| CN111476865B CN111476865B (en) | 2023-07-07 |
Family
ID=71747713
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010211679.4A Active CN111476865B (en) | 2020-03-24 | 2020-03-24 | Image protection method for image recognition based on deep learning neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111476865B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113808044A (en) * | 2021-09-17 | 2021-12-17 | 北京百度网讯科技有限公司 | Encryption mask determining method, device, equipment and storage medium |
| WO2022152153A1 (en) * | 2021-01-18 | 2022-07-21 | 北京灵汐科技有限公司 | Image processing method and device, key generation method and device, training method, and computer readable medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1808969A (en) * | 2006-01-16 | 2006-07-26 | 钱丽娜 | Password transmission apparatus and method for weak authentication |
| US20060170669A1 (en) * | 2002-08-12 | 2006-08-03 | Walker Jay S | Digital picture frame and method for editing |
| CN107729765A (en) * | 2017-09-30 | 2018-02-23 | 广东欧珀移动通信有限公司 | Management method, device, storage medium and the electronic equipment of multi-medium data |
| CN107770406A (en) * | 2017-10-26 | 2018-03-06 | 北京邮电大学 | Image encryption method and device based on the conversion of multi-parameter fractional order and semi-tensor product |
-
2020
- 2020-03-24 CN CN202010211679.4A patent/CN111476865B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060170669A1 (en) * | 2002-08-12 | 2006-08-03 | Walker Jay S | Digital picture frame and method for editing |
| CN1808969A (en) * | 2006-01-16 | 2006-07-26 | 钱丽娜 | Password transmission apparatus and method for weak authentication |
| CN107729765A (en) * | 2017-09-30 | 2018-02-23 | 广东欧珀移动通信有限公司 | Management method, device, storage medium and the electronic equipment of multi-medium data |
| CN107770406A (en) * | 2017-10-26 | 2018-03-06 | 北京邮电大学 | Image encryption method and device based on the conversion of multi-parameter fractional order and semi-tensor product |
Non-Patent Citations (2)
| Title |
|---|
| 石慧 等: "抗几何攻击图像水印算法研究", 计算机工程与设计, pages 4796 - 4801 * |
| *** 等: "FVEA-H:一种用于H.264 的快速视频加密算法", ***仿真学报, pages 4363 - 4366 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022152153A1 (en) * | 2021-01-18 | 2022-07-21 | 北京灵汐科技有限公司 | Image processing method and device, key generation method and device, training method, and computer readable medium |
| CN113808044A (en) * | 2021-09-17 | 2021-12-17 | 北京百度网讯科技有限公司 | Encryption mask determining method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111476865B (en) | 2023-07-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yan et al. | Secure video retrieval using image query on an untrusted cloud | |
| Younus et al. | Video steganography using knight tour algorithm and LSB method for encrypted data | |
| CN115378574B (en) | Lightweight dynamic image data encryption method and system | |
| CN111476865B (en) | Image protection method for image recognition based on deep learning neural network | |
| Liu et al. | Unauthorized AI cannot recognize me: Reversible adversarial example | |
| CN105975869B (en) | Adaptively selected encryption method based on salient region | |
| Xu et al. | Audio-visual autoencoding for privacy-preserving video streaming | |
| Ye et al. | PRA-TPE: Perfectly recoverable approximate thumbnail-preserving image encryption | |
| Deng et al. | Color image watermarking using regularized extreme learning machine | |
| Duan et al. | StegoPNet: Image steganography with generalization ability based on pyramid pooling module | |
| Cui et al. | Multitask identity-aware image steganography via minimax optimization | |
| Dalal et al. | A secure video steganography scheme using DWT based on object tracking | |
| Li et al. | Coverless image steganography using morphed face recognition based on convolutional neural network | |
| Wen et al. | A hybrid model for natural face de-identiation with adjustable privacy | |
| Pandurangi Ramacharya et al. | Fast partial image encryption with fuzzy logic and chaotic mapping | |
| Darwish et al. | A new colour image copyright protection approach using evolution-based dual watermarking | |
| Niu et al. | A novel generation method for diverse privacy image based on machine learning | |
| Jin et al. | Minimizing Turtle-Shell Matrix Based Stego Image Distortion Using Particle Swarm Optimization. | |
| CN106055632B (en) | Video authentication method based on scene frame fingerprints | |
| Kaur et al. | Adaptive block based steganographic model with dynamic block estimation with fuzzy rules | |
| Seem et al. | Artificial neural network, convolutional neural network visualization, and image security | |
| CN109561236A (en) | Image encryption, recognition methods device, equipment and medium | |
| Shashidhar et al. | Novel Approach for Steganography to Camouflage Digital Information Using Least Significant Bit | |
| Neethu et al. | Securing image posts in social networking sites | |
| Hilal et al. | Forensic Analysis of Images on Online Social Network: A Survey |
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 |