CN111259427A - Image processing method and device based on neural network and storage medium - Google Patents
Image processing method and device based on neural network and storage medium Download PDFInfo
- Publication number
- CN111259427A CN111259427A CN202010070776.6A CN202010070776A CN111259427A CN 111259427 A CN111259427 A CN 111259427A CN 202010070776 A CN202010070776 A CN 202010070776A CN 111259427 A CN111259427 A CN 111259427A
- Authority
- CN
- China
- Prior art keywords
- image
- noise
- neural network
- encrypted
- network
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 117
- 238000003672 processing method Methods 0.000 title claims abstract description 11
- 238000000034 method Methods 0.000 claims abstract description 47
- 238000012545 processing Methods 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims description 80
- 239000011159 matrix material Substances 0.000 claims description 65
- 230000006870 function Effects 0.000 claims description 62
- 238000004590 computer program Methods 0.000 claims description 10
- 230000000007 visual effect Effects 0.000 abstract description 14
- 238000003062 neural network model Methods 0.000 description 26
- 238000010586 diagram Methods 0.000 description 22
- 238000013473 artificial intelligence Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000003745 diagnosis Methods 0.000 description 5
- 230000005540 biological transmission Effects 0.000 description 4
- 230000002441 reversible effect Effects 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000008014 freezing Effects 0.000 description 2
- 238000007710 freezing Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 1
- 229910052802 copper Inorganic materials 0.000 description 1
- 239000010949 copper Substances 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2221/00—Indexing scheme relating to security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F2221/21—Indexing scheme relating to G06F21/00 and subgroups addressing additional information or applications relating to security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F2221/2107—File encryption
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Security & Cryptography (AREA)
- Computational Linguistics (AREA)
- Bioethics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Hardware Design (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The present disclosure relates to an image processing method, apparatus and storage medium based on a neural network, wherein the method comprises: obtaining encryption noise according to the first image, a preset secret key and a noise generator; obtaining a second image according to the first image and the encrypted noise; and obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network. In the embodiment of the disclosure, the encrypted neural network is only effective for the image encrypted by the preset key, and cannot be directly applied to the original image under the condition that the structure and parameters of the neural network are stolen, so that the safety of image processing is improved; meanwhile, the encrypted image still keeps the visual effect of the original image, and after the original image is encrypted and distributed to equipment carrying the encrypted neural network, the encrypted neural network can be used for prediction, and the correctness of a prediction result can be further verified according to the encrypted image.
Description
Technical Field
The present disclosure relates to the field of computer image processing technologies, and in particular, to an image processing method and apparatus based on a neural network, and a storage medium.
Background
Artificial Intelligence (AI) technology, represented by a neural network, is applied to a variety of scenes such as medical diagnosis, face recognition, automatic driving, and the like. In order to improve the security of image processing, the neural network model can be encrypted, so that the neural network model cannot be directly used even if being stolen.
In the related art, the original data is encrypted first, and then training or prediction of a neural network model is performed using the encrypted data. However, in the scheme, the visual effect of the image itself can be changed in the encryption process, and meanwhile, because the data encryption and the neural network prediction are integrated, if the encrypted data are distributed to other equipment for prediction, the image received by the equipment does not have the original visual effect, and the correctness of the prediction result cannot be verified and checked.
Disclosure of Invention
In view of the above, the present disclosure provides an image processing method and apparatus based on a neural network, and a storage medium.
According to an aspect of the present disclosure, there is provided a neural network-based image processing method, the method including:
obtaining encryption noise according to the first image, a preset secret key and a noise generator;
obtaining a second image according to the first image and the encrypted noise;
and obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network.
In one possible implementation, the method further includes:
and obtaining the first image according to the second image and the encryption noise.
In a possible implementation manner, the obtaining of the encrypted noise according to the first image, the preset key, and the noise generator includes:
obtaining the preset secret key according to at least one designated position selected from the first noise matrix; wherein the first noise matrix is a matrix of the same size as the first image;
assigning at least one vector output by the noise generator to the at least one designated position respectively to obtain a second noise matrix, wherein the values of other pixel positions except the at least one designated position are 0;
and taking the second noise matrix as the encryption noise.
In a possible implementation manner, before assigning the at least one vector output by the noise generator to the at least one designated position respectively, the method further includes:
selecting the values of the positions of other pixel points except the at least one appointed position from the first noise matrix;
and obtaining at least one vector input into the noise generator according to the values of the positions of other pixel points except the at least one designated position.
In one possible implementation, the method further includes:
training according to the first sample image and the second sample image to obtain the encrypted neural network;
the encrypted neural network is a Generative Adaptive Network (GAN) obtained by Generative Adaptive Networks (GANs) between a noise generation network and an image discrimination network.
In one possible implementation, the method further includes:
training according to the first sample image to obtain an original neural network;
the second sample image is a sample image obtained by encrypting the first sample image after a noise matrix is obtained according to the noise generation network;
the training according to the first sample image and the second sample image to obtain the encrypted neural network includes:
obtaining a fourth loss function value according to a first loss function value of the first sample image on the original neural network, a second loss function value of the first sample image on the encrypted neural network, a third loss function value of the second sample image on the encrypted neural network and a 2 norm of a noise matrix obtained by the noise generation network, and training the noise generation network through back propagation of the fourth loss function value;
obtaining a fifth loss function according to a first discrimination output value obtained by inputting the first sample image into the image discrimination network and a second discrimination output value obtained by inputting the second sample image into the image discrimination network, and training the image discrimination network through the back propagation of the fifth loss function value;
if the noise generation network and the image discrimination network reach convergence through the generation countermeasure training, ending the generation countermeasure training, taking the noise generation network as the noise generator, and taking the image discrimination network as a discriminator;
and training the encrypted neural network to converge according to the noise generator and the discriminator.
According to another aspect of the present disclosure, there is provided a neural network-based image processing apparatus including:
the encryption noise generation module is used for obtaining encryption noise according to the first image, a preset secret key and the noise generator;
the image encryption module is used for obtaining a second image according to the first image and the encryption noise;
and the prediction module is used for obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network.
In one possible implementation, the apparatus further includes:
and the image decryption module is used for obtaining the first image according to the second image and the encryption noise.
In one possible implementation, the encryption noise generation module includes:
the preset key generating unit is used for obtaining the preset key according to at least one appointed position selected from the first noise matrix; wherein the first noise matrix is a matrix of the same size as the first image;
the second noise matrix generating unit is used for assigning at least one vector output by the noise generator to the at least one designated position respectively to obtain a second noise matrix, and the values of other pixel positions except the at least one designated position are 0;
and an encryption noise generation unit configured to use the second noise matrix as the encryption noise.
In a possible implementation manner, before assigning the at least one vector output by the noise generator to the at least one designated position respectively, the method further includes:
selecting the values of the positions of other pixel points except the at least one appointed position from the first noise matrix;
and obtaining at least one vector input into the noise generator according to the values of the positions of other pixel points except the at least one designated position.
In one possible implementation, the apparatus further includes:
the training module is used for training according to the first sample image and the second sample image to obtain the encrypted neural network;
the encrypted neural network is a GAN obtained by the generation countermeasure training between the noise generation network and the image discrimination network.
In a possible implementation manner, the training module is further configured to perform training according to the first sample image to obtain an original neural network;
the second sample image is a sample image obtained by encrypting the first sample image after a noise matrix is obtained according to the noise generation network;
the training module comprises:
a noise generation network training unit, configured to obtain a fourth loss function value according to a first loss function value of the first sample image on the original neural network, a second loss function value of the first sample image on the encrypted neural network, a third loss function value of the second sample image on the encrypted neural network, and a 2-norm of a noise matrix obtained by the noise generation network, and train the noise generation network through back propagation of the fourth loss function value;
the image discrimination network training unit is used for obtaining a fifth loss function according to a first discrimination output value obtained by inputting the first sample image into the image discrimination network and a second discrimination output value obtained by inputting the second sample image into the image discrimination network, and training the image discrimination network through the back propagation of the fifth loss function value;
an encrypted neural network generation unit configured to end the generation countermeasure training if the noise generation network and the image discrimination network converge through the generation countermeasure training, use the noise generation network as the noise generator, and use the image discrimination network as the discriminator; and training the encrypted neural network to converge according to the noise generator and the discriminator.
According to another aspect of the present disclosure, there is provided a neural network-based image processing apparatus including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the above method.
According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.
In the embodiment of the disclosure, an original image is encrypted through a preset secret key and a noise generator, and a prediction result is obtained by using the encrypted image and an encrypted neural network; the encrypted neural network is only effective for the image encrypted by the preset key, and cannot be directly applied to the original image under the condition that the structure and parameters of the neural network are stolen, so that the safety of image processing is improved; meanwhile, the encrypted image still keeps the visual effect of the original image, and after the original image is encrypted and distributed to equipment carrying the encrypted neural network, the encrypted neural network can be used for prediction, and the correctness of a prediction result can be further verified according to the encrypted image.
Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.
FIG. 1 shows a flow diagram of a neural network-based image processing method according to an embodiment of the present disclosure;
FIG. 2 illustrates a schematic diagram of prediction by a trained encrypted neural network according to an embodiment of the present disclosure;
FIG. 3 shows a schematic diagram of encrypting an image according to an embodiment of the present disclosure;
FIG. 4 shows a schematic diagram of training by an encrypted neural network according to an embodiment of the present disclosure;
FIG. 5 shows a schematic flow diagram for training by training an encrypted neural network according to an embodiment of the present disclosure;
FIG. 6 illustrates a block diagram of a neural network-based image processing apparatus according to an embodiment of the present disclosure;
fig. 7 illustrates a block diagram of an apparatus for image processing with a neural network, according to an embodiment of the present disclosure.
Detailed Description
Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.
The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.
Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.
In recent years, an artificial intelligence technology represented by a neural network has undergone a rapid development, and the technology has been applied to many scenes such as medical diagnosis, face recognition, automatic driving, and the like, particularly in the field of image processing. In these applications, the most central part is the neural network model, so encrypting the neural network model can prevent the neural network model from being directly used even if stolen, thereby improving the security of image processing.
In the related technology, the original data is encrypted firstly, and then the encrypted data is used for training or predicting a neural network model; in this scheme, the encryption process changes the visual effect of the image itself and the encryption method is irreversible, i.e. the original image cannot be restored from the encrypted image. Meanwhile, in the scheme, data encryption and neural network prediction are integrated, and if the encrypted data are distributed to equipment carrying the neural network prediction for prediction, the image received by the equipment does not have the original visual effect; in some application scenarios, such as the field of medical imaging, the encrypted image can only be used for model prediction and cannot be used for diagnosis by a doctor through observation, so that the correctness of the model prediction result cannot be verified and checked.
Therefore, the present disclosure provides an image processing scheme based on a neural network, which encrypts an original image through a preset key and a noise generator, and obtains a prediction result by using the encrypted image and the encrypted neural network; the encrypted neural network is only effective for the image encrypted by the preset key, and cannot be directly applied to the original image under the condition that the structure and parameters of the neural network are stolen, so that the safety of image processing is improved; meanwhile, the encrypted image still keeps the visual effect of the original image, and after the original image is encrypted and distributed to equipment carrying the encrypted neural network, the encrypted neural network can be used for prediction, and the correctness of a prediction result can be further verified according to the encrypted image.
Fig. 1 shows a flowchart of a neural network-based image processing method according to an embodiment of the present disclosure. As shown in fig. 1, the method may include the steps of:
and step 30, obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network.
In the embodiment of the disclosure, in the process of predicting the first image, the preset key and the noise generator may be used to generate the encrypted noise, and the encrypted noise is used to encrypt the first image, so as to obtain the second image, and then the trained encrypted neural network may be used to predict the second image, so as to obtain the related prediction result of image encryption. The encrypted neural network can obtain a correct prediction result only when the second image is input, and can give an incorrect prediction result when the first image is input, so that the encryption of the neural network model is realized. In addition, since the preset key is produced and grasped by the holder of the encrypted neural network, even if information such as the structure and parameters of the encrypted neural network is stolen, the thief cannot obtain the key, so that the first image cannot be encrypted, and further cannot use the encrypted neural network, thereby ensuring the security of image processing.
It should be noted that, in the embodiment of the present disclosure, since the image processing method can maintain the visual effect of the image while ensuring the prediction effect of the encrypted neural network, after the first image is encrypted to obtain the second image, and the second image is distributed to the device on which the encrypted neural network is mounted, the second image can be predicted by using the corresponding encrypted neural network, and the visual effect of the second image itself will not be damaged. Therefore, in the embodiment of the present disclosure, the encryption of the first image to obtain the second image (i.e., steps 10 and 20) and the prediction of the second image by using the encrypted neural network (i.e., step 30) may be implemented together or separately, which is not limited thereto; illustratively, encrypting the first image to obtain the second image may be implemented by the first device, and at the same time, predicting the second image may be implemented by the second device carrying the corresponding encrypted neural network. Different preset keys can be set for encrypted neural networks deployed in different devices, so that the safety of the neural network model is further ensured.
For example, fig. 2 is a schematic diagram illustrating prediction by a trained encrypted neural network according to an embodiment of the present disclosure, and as shown in fig. 2, an original image (i.e., a first image) is first encrypted by using a preset key and a noise generator to obtain an encrypted image (i.e., a second image). The encrypted image is then input to a trained encrypted neural network, so that a correct prediction result can be obtained. Meanwhile, as shown by a dotted line in fig. 2, if an original image to be predicted is directly input to the trained encrypted neural network, an erroneous prediction result may be given. Therefore, even if the encrypted neural network model is stolen, the original image cannot be encrypted due to the absence of the preset key, and a correct prediction result cannot be obtained.
In a possible implementation manner, the obtaining of the encrypted noise according to the first image, the preset key, and the noise generator includes: obtaining the preset secret key according to at least one designated position selected from the first noise matrix; wherein the first noise matrix is a matrix of the same size as the first image; assigning at least one vector output by the noise generator to the at least one designated position respectively to obtain a second noise matrix, wherein the values of other pixel positions except the at least one designated position are 0; and taking the second noise matrix as the encryption noise.
Illustratively, the size of the first image is H × W, and the size of the first noise matrix is H × W, N specific locations are selected on the first noise matrix, and the N locations are a preset key. The noise generator outputs a vector with the length of N, and the value of each component in the vector is respectively assigned to N designated positions selected in the first noise matrix; meanwhile, values of positions other than the N positions in the first noise matrix are assigned to 0, and thus, a second noise matrix is generated. The specific numerical value of N and the specific position at the designated position may be set according to actual requirements, which is not limited in the embodiments of the present disclosure, for example, the value of N and the specific position at the designated position may be set according to the number of rows and columns (H and W) of the first noise matrix, and the numerical value of N may be consistent with the number of rows H of the first noise matrix; the specific locations may be distributed over the rows in the first noise matrix, i.e. a specific location is selected in each row.
In a possible implementation manner, before assigning the at least one vector output by the noise generator to the at least one designated position respectively, the method further includes: selecting the values of the positions of other pixel points except the at least one appointed position from the first noise matrix; and obtaining at least one vector input into the noise generator according to the values of the positions of other pixel points except the at least one designated position.
Illustratively, according to the above embodiment, the values of all pixel points remaining in the first image except the N selected specific positions on the first noise matrix are extracted and arranged as a vector with a length H × W-N; however, the vector may be input to a noise generator to obtain the encrypted noise in the manner described in the above embodiment, and the first image may be superimposed with the encrypted noise to obtain the second image.
For example, fig. 3 shows a schematic diagram of encrypting an image according to an embodiment of the present disclosure, as shown in fig. 3, the size of an original image (i.e., a first image) is 4 × 4, a matrix with the same size as the original image is first generated, 4 specific positions are selected on the matrix, and the 4 positions can be regarded as a preset key. The values of all the pixels remaining except these 4 positions are then extracted and arranged as a first vector of length 12. Next, the vector is input into a noise generator, a second vector with a length of 4 is output, and the value of each element in the obtained new vector is respectively assigned to 4 selected positions in the noise matrix, and the values of the positions in the noise matrix except the 4 positions are assigned to 0, so that a noise matrix (i.e. encryption noise) is generated. Further, the noise matrix and the original image matrix may be added to obtain a new matrix, which is the encrypted image matrix (i.e., the second image).
In one possible implementation, the method further includes: and obtaining the first image according to the second image and the encryption noise.
For example, when the second image needs to be converted into the first image, the noise matrix may be generated by using a corresponding process with reference to the way of performing image encryption in the above embodiment, and then the noise matrix may be subtracted from the second image to obtain the first image; therefore, the image encryption flow in the embodiment of the disclosure is reversible, that is, the original image can be restored from the encrypted image, thereby avoiding the original data loss possibly caused by the encryption process.
Based on this, the embodiment of the present disclosure further provides a training method of the neural network encrypted in the above embodiment.
In one possible implementation, the method further includes: training according to the first sample image and the second sample image to obtain the encrypted neural network; the encrypted neural network is a GAN obtained by the generation countermeasure training between the noise generation network and the image discrimination network.
The GAN is a deep learning model, is composed of a noise generation network and an image discrimination network, and generates required output by mutual game. The noise generation network is used for generating a noise matrix from the first sample image, and the image discrimination network judges whether the input image is the first sample image or the second sample image. Thus, the noise generation network can generate noise as small as possible to confuse the image discrimination network, thereby minimizing the difference between the second sample image and the first sample image. After the GAN is finished, fixing each parameter in the GAN, continuing to train the encrypted neural network until convergence, and after the training is finished, the noise generator is used for encrypting the first image in the prediction process of the above embodiment, and the encrypted neural network is used for predicting the second image. Therefore, the GAN and a corresponding training strategy are used for realizing model encryption, and a noise generator and an encrypted neural network which meet requirements are trained by selecting a proper loss function, so that the difference between an original image and an encrypted image is very small, and correct and wrong prediction results can be respectively given after the original image and the encrypted image are input into the encrypted neural network, thereby playing the role of improving the safety of image processing.
For example, fig. 4 shows a schematic diagram of training an encrypted neural network according to an embodiment of the present disclosure, as shown in fig. 4, in the training phase, a preset key may be first generated in the manner described in the above embodiment, the preset key is grasped by an owner of the neural network model, and an original image (i.e., a first sample image) is encrypted by using the preset key to generate an encrypted image (i.e., a second sample image), and then the neural network model is trained by using the original image and the encrypted image at the same time; when the noise generation network is trained, the encryption noise is reduced as much as possible by selecting a proper loss function, meanwhile, the neural network model gives a result which is correct as much as possible on the encrypted image and a result which is wrong as much as possible on the original image, and the trained encrypted neural network is stored after the training process is finished.
In one possible implementation, the method further includes: training according to the first sample image to obtain an original neural network; the second sample image is a sample image obtained by encrypting the first sample image after a noise matrix is obtained according to the noise generation network;
the training according to the first sample image and the second sample image to obtain the encrypted neural network includes: obtaining a fourth loss function value according to a first loss function value of the first sample image on the original neural network, a second loss function value of the first sample image on the encrypted neural network, a third loss function value of the second sample image on the encrypted neural network and a 2 norm of a noise matrix obtained by the noise generation network, and training the noise generation network through back propagation of the fourth loss function value; obtaining a fifth loss function according to a first discrimination output value obtained by inputting the first sample image into the image discrimination network and a second discrimination output value obtained by inputting the second sample image into the image discrimination network, and training the image discrimination network through the back propagation of the fifth loss function value; if the noise generation network and the image discrimination network reach convergence through the generation countermeasure training, ending the generation countermeasure training, taking the noise generation network as the noise generator, and taking the image discrimination network as a discriminator; and training the encrypted neural network to converge according to the noise generator and the discriminator.
It should be noted that, in the embodiment of the present disclosure, the encrypted noise is generated by GAN, and since a term representing the magnitude of the noise is added to the fourth loss function for training the noise generation network, the noise generation network encrypts the first sample image with the noise as small as possible, so that the second sample image hardly generates a visual difference with respect to the first sample image, and the accuracy of the prediction result can be further verified according to the second sample image.
For example, fig. 5 shows a schematic flow chart of training an encrypted neural network according to an embodiment of the present disclosure, and as shown in fig. 5, first, an original neural network model is trained using an original image (i.e., a first sample image). Then, a GAN (including a noise generation network for generating a noise matrix from the original image and an image discrimination network for discriminating whether the input image is the original image or the encrypted image) is constructed. When the noise generation network is trained, freezing parameters of the image discrimination network, calculating by using corresponding output results obtained by the original image respectively through the noise generation network, the image discrimination network and the neural network model to obtain a fourth loss function, and updating the parameters of the noise generation network by the fourth loss function through an optimizer, thereby achieving the purpose of training the noise generation network; correspondingly, when the image discrimination network is trained, the parameters of the noise generation network are frozen, a fifth loss function is calculated by using corresponding output results obtained by the original image and the encrypted image respectively through the network, and the parameters of the image discrimination network are updated by the fifth loss function through an optimizer, so that the aim of training the image discrimination network is fulfilled; and repeatedly and alternately training the noise generation network and the image discrimination network until convergence. In this way, the noise generation network can generate noise as small as possible to confuse the image discrimination network by generating the countermeasure training, so that the difference between the encrypted image and the original image is as small as possible; in the training process, the loss function (i.e., the fifth loss function) when the training image discriminates the network is selected as:
Ld＝yn+(1-yc)
wherein y isc,ynRespectively representing the output values of the input image as the original image and the encrypted image.
The loss function when training the noise-generating network (i.e., the fourth loss function) is chosen as:
Lg＝(1-yn)-|L(dc,mc)–L(dc,mn)|+|L(dc,mc)-L(dn,mn)|+Lp
wherein, L (d)c,mc) The loss function value of the original image on the original neural network model (i.e. the first loss function value), L (d)c,mn) For the loss function value of the original image on the encrypted neural network (i.e. the second loss function value), L (d)n,mn) For the loss function value (i.e. the third loss function value), L, of the encrypted image on the encrypted neural networkpA 2-norm representing a noise matrix produced by the noise generating network; wherein L ispRepresenting the intensity of the added noise, the loss function can make the noise generated when training the noise generation network as small as possible, thereby maintaining the noise as possibleVisual impact of the original image.
And finally, after the GAN converges, freezing parameters of the noise generator and the discriminator, and continuing to train the encrypted neural network until the GAN converges to obtain the trained encrypted neural network.
It should be noted that, although the above embodiments are described as examples of an image processing method based on a neural network, those skilled in the art can understand that the present disclosure should not be limited thereto. In fact, the user can flexibly set each implementation mode according to personal preference and/or actual application scene, as long as the technical scheme of the disclosure is met.
In this way, the reversible image encryption mode controlled by the key in the embodiment of the disclosure is realized by combining the image encryption and the corresponding model training strategy, rather than simply using the encrypted image training model; because the item representing the noise magnitude is added into the loss function of the training noise generator, the model can encrypt the image with the noise as small as possible, the encrypted image hardly generates visual difference relative to the original image, for some application scenes such as medical AI products, the image encryption part and the neural network model can be separated, namely, the image distributed to the equipment carrying the neural network model is the encrypted image, and the encrypted image still keeps the visual effect of the original image, so the image can be used for predicting the neural network model and can also be directly observed by a doctor to make diagnosis, thereby facilitating the doctor to further verify the prediction result given by the neural network model. Meanwhile, as the model encryption method is controlled by the preset key, different encryption methods can be set for different equipment by changing the preset key, and even if the whole neural network model is stolen, the model can not be used without the corresponding key, so that the safety of image processing is effectively protected, and the method is suitable for the encryption requirements of images in the fields of medical AI and the like.
Fig. 6 illustrates a block diagram of an image processing apparatus based on a neural network according to an embodiment of the present disclosure. As shown in fig. 6, the apparatus may include: an encryption noise generation module 41 for obtaining encryption noise according to the first image, the preset key and the noise generator; an image encryption module 42, configured to obtain a second image according to the first image and the encryption noise; and the prediction module 43 is configured to obtain a prediction result related to image encryption through the second image and the encrypted neural network.
In one possible implementation, the apparatus further includes: and the image decryption module is used for obtaining the first image according to the second image and the encryption noise.
In one possible implementation, the encryption noise generation module includes: the preset key generating unit is used for obtaining the preset key according to at least one appointed position selected from the first noise matrix; wherein the first noise matrix is a matrix of the same size as the first image; the second noise matrix generating unit is used for assigning at least one vector output by the noise generator to the at least one designated position respectively to obtain a second noise matrix, and the values of other pixel positions except the at least one designated position are 0; and an encryption noise generation unit configured to use the second noise matrix as the encryption noise.
In a possible implementation manner, before assigning the at least one vector output by the noise generator to the at least one designated position respectively, the method further includes:
selecting the values of the positions of other pixel points except the at least one appointed position from the first noise matrix;
and obtaining at least one vector input into the noise generator according to the values of the positions of other pixel points except the at least one designated position.
In one possible implementation, the apparatus further includes: the training module is used for training according to the first sample image and the second sample image to obtain the encrypted neural network; the encrypted neural network is a GAN obtained by the generation countermeasure training between the noise generation network and the image discrimination network.
In a possible implementation manner, the training module is further configured to perform training according to the first sample image to obtain an original neural network; the second sample image is a sample image obtained by encrypting the first sample image after a noise matrix is obtained according to the noise generation network; the training module comprises: a noise generation network training unit, configured to obtain a fourth loss function value according to a first loss function value of the first sample image on the original neural network, a second loss function value of the first sample image on the encrypted neural network, a third loss function value of the second sample image on the encrypted neural network, and a 2-norm of a noise matrix obtained by the noise generation network, and train the noise generation network through back propagation of the fourth loss function value; the image discrimination network training unit is used for obtaining a fifth loss function according to a first discrimination output value obtained by inputting the first sample image into the image discrimination network and a second discrimination output value obtained by inputting the second sample image into the image discrimination network, and training the image discrimination network through the back propagation of the fifth loss function value; an encrypted neural network generation unit configured to end the generation countermeasure training if the noise generation network and the image discrimination network converge through the generation countermeasure training, use the noise generation network as the noise generator, and use the image discrimination network as the discriminator; and training the encrypted neural network to converge according to the noise generator and the discriminator.
It should be noted that, although the above embodiments are described as examples of a neural network-based image processing apparatus, those skilled in the art can understand that the present disclosure should not be limited thereto. In fact, the user can flexibly set each implementation mode according to personal preference and/or actual application scene, as long as the technical scheme of the disclosure is met.
In this way, the reversible image encryption mode controlled by the key in the embodiment of the disclosure is realized by combining the image encryption and the corresponding model training strategy, rather than simply using the encrypted image training model; because the item representing the noise magnitude is added into the loss function of the training noise generator, the model can encrypt the image with the noise as small as possible, the encrypted image hardly generates visual difference relative to the original image, for some application scenes such as medical AI products, the image encryption part and the neural network model can be separated, namely, the image distributed to the equipment carrying the neural network model is the encrypted image, and the encrypted image still keeps the visual effect of the original image, so the image can be used for predicting the neural network model and can also be directly observed by a doctor to make diagnosis, thereby facilitating the doctor to further verify the prediction result given by the neural network model. Meanwhile, as the model encryption method is controlled by the preset key, different encryption methods can be set for different equipment by changing the preset key, and even if the whole neural network model is stolen, the model can not be used without the corresponding key, so that the safety of image processing is effectively protected, and the method is suitable for the encryption requirements of images in the fields of medical AI and the like.
Fig. 7 shows a block diagram of an apparatus 1900 for neural network based image processing according to an embodiment of the present disclosure. For example, the apparatus 1900 may be provided as a server. Referring to fig. 7, the device 1900 includes a processing component 1922 further including one or more processors and memory resources, represented by memory 1932, for storing instructions, e.g., applications, executable by the processing component 1922. The application programs stored in memory 1932 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1922 is configured to execute instructions to perform the above-described method.
The device 1900 may also include a power component 1926 configured to perform power management of the device 1900, a wired or wireless network interface 1950 configured to connect the device 1900 to a network, and an input/output (I/O) interface 1958. The device 1900 may operate based on an operating system stored in memory 1932, such as Windows Server, MacOS XTM, UnixTM, LinuxTM, FreeBSDTM, or the like.
In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1932, is also provided that includes computer program instructions executable by the processing component 1922 of the apparatus 1900 to perform the above-described methods.
The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.
The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.
The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.
The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).
Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.
These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.
Claims (10)
1. An image processing method based on a neural network, the method comprising:
obtaining encryption noise according to the first image, a preset secret key and a noise generator;
obtaining a second image according to the first image and the encrypted noise;
and obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network.
2. The method of claim 1, further comprising:
and obtaining the first image according to the second image and the encryption noise.
3. The method of claim 1, wherein obtaining the encrypted noise from the first image, the predetermined key, and the noise generator comprises:
obtaining the preset secret key according to at least one designated position selected from the first noise matrix; wherein the first noise matrix is a matrix of the same size as the first image;
assigning at least one vector output by the noise generator to the at least one designated position respectively to obtain a second noise matrix, wherein the values of other pixel positions except the at least one designated position are 0;
and taking the second noise matrix as the encryption noise.
4. The method of claim 2, wherein before assigning the at least one vector output by the noise generator to the at least one designated location, respectively, the method further comprises:
selecting the values of the positions of other pixel points except the at least one appointed position from the first noise matrix;
and obtaining at least one vector input into the noise generator according to the values of the positions of other pixel points except the at least one designated position.
5. The method according to any one of claims 1 to 4, further comprising:
training according to the first sample image and the second sample image to obtain the encrypted neural network;
the encrypted neural network is a generative confrontation network GAN obtained through generative confrontation training between the noise generation network and the image discrimination network.
6. The method of claim 5, further comprising:
training according to the first sample image to obtain an original neural network;
the second sample image is a sample image obtained by encrypting the first sample image after a noise matrix is obtained according to the noise generation network;
the training according to the first sample image and the second sample image to obtain the encrypted neural network includes:
obtaining a fourth loss function value according to a first loss function value of the first sample image on the original neural network, a second loss function value of the first sample image on the encrypted neural network, a third loss function value of the second sample image on the encrypted neural network and a 2 norm of a noise matrix obtained by the noise generation network, and training the noise generation network through back propagation of the fourth loss function value;
obtaining a fifth loss function according to a first discrimination output value obtained by inputting the first sample image into the image discrimination network and a second discrimination output value obtained by inputting the second sample image into the image discrimination network, and training the image discrimination network through the back propagation of the fifth loss function value;
if the noise generation network and the image discrimination network reach convergence through the generation countermeasure training, ending the generation countermeasure training, taking the noise generation network as the noise generator, and taking the image discrimination network as a discriminator;
and training the encrypted neural network to converge according to the noise generator and the discriminator.
7. An image processing apparatus based on a neural network, comprising:
the encryption noise generation module is used for obtaining encryption noise according to the first image, a preset secret key and the noise generator;
the image encryption module is used for obtaining a second image according to the first image and the encryption noise;
and the prediction module is used for obtaining a relevant prediction result of image encryption through the second image and the encrypted neural network.
8. The apparatus of claim 7, further comprising:
the training module is used for training according to the first sample image and the second sample image to obtain the encrypted neural network;
the encrypted neural network is a generative confrontation network GAN obtained through generative confrontation training between the noise generation network and the image discrimination network.
9. An image processing apparatus based on a neural network, comprising:
a processor;
a memory for storing processor-executable instructions;
wherein the processor is configured to implement the method of any one of claims 1 to 6 when executing the memory-stored executable instructions.
10. A non-transitory computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the method of any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010070776.6A CN111259427B (en) | 2020-01-21 | 2020-01-21 | Image processing method and device based on neural network and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010070776.6A CN111259427B (en) | 2020-01-21 | 2020-01-21 | Image processing method and device based on neural network and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111259427A true CN111259427A (en) | 2020-06-09 |
| CN111259427B CN111259427B (en) | 2020-11-06 |
Family
ID=70952482
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010070776.6A Active CN111259427B (en) | 2020-01-21 | 2020-01-21 | Image processing method and device based on neural network and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111259427B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563275A (en) * | 2020-07-14 | 2020-08-21 | 中国人民解放军国防科技大学 | Data desensitization method based on generation countermeasure network |
| CN112163229A (en) * | 2020-09-14 | 2021-01-01 | 长春理工大学 | Learning type digital image encryption and decryption method based on loop generation countermeasure network |
| CN112395635A (en) * | 2021-01-18 | 2021-02-23 | 北京灵汐科技有限公司 | Image processing method, device, secret key generating method, device, training method and device, and computer readable medium |
| CN112733173A (en) * | 2021-01-18 | 2021-04-30 | 北京灵汐科技有限公司 | Image processing method, device, secret key generating method, device, training method and device, and computer readable medium |
| CN113407968A (en) * | 2021-06-29 | 2021-09-17 | 平安国际智慧城市科技股份有限公司 | Encryption method, device, equipment and storage medium of target detection model |
| CN113408653A (en) * | 2021-07-12 | 2021-09-17 | 广东电网有限责任公司 | Identification method for adaptively reducing complex light and shadow interference and related device |
| 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 (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108615073A (en) * | 2018-04-28 | 2018-10-02 | 北京京东金融科技控股有限公司 | Image processing method and device, computer readable storage medium, electronic equipment |
| CN108629193A (en) * | 2018-04-26 | 2018-10-09 | 成都大象分形智能科技有限公司 | A kind of encryption protection system and method for artificial nerve network model |
| CN108776836A (en) * | 2018-06-08 | 2018-11-09 | 电子科技大学 | A kind of training of the secret protection neural network based on VHE and prediction technique |
| CN109636741A (en) * | 2018-11-14 | 2019-04-16 | 建湖云飞数据科技有限公司 | A kind of image denoising processing method |
| CN109657156A (en) * | 2019-01-22 | 2019-04-19 | 杭州师范大学 | A kind of personalized recommendation method generating confrontation network based on circulation |
| CN109685068A (en) * | 2018-12-27 | 2019-04-26 | 上海七牛信息技术有限公司 | A kind of image processing method and system based on generation confrontation neural network |
| CN110071798A (en) * | 2019-03-21 | 2019-07-30 | 深圳大学 | A kind of equivalent key acquisition methods, device and computer readable storage medium |
| US20190295073A1 (en) * | 2018-03-22 | 2019-09-26 | Via Science, Inc. | Secure data processing transactions |
| CN110312053A (en) * | 2019-05-10 | 2019-10-08 | 南通理工学院 | Based on sublevel Fourier spectral decomposition and the optical encryption method for mixing random mask |
| CN110490128A (en) * | 2019-08-16 | 2019-11-22 | 南京邮电大学 | A kind of hand-written recognition method based on encryption neural network |
| CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
| CN110633650A (en) * | 2019-08-22 | 2019-12-31 | 首都师范大学 | Convolutional neural network face recognition method and device based on privacy protection |
-
2020
- 2020-01-21 CN CN202010070776.6A patent/CN111259427B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190295073A1 (en) * | 2018-03-22 | 2019-09-26 | Via Science, Inc. | Secure data processing transactions |
| CN108629193A (en) * | 2018-04-26 | 2018-10-09 | 成都大象分形智能科技有限公司 | A kind of encryption protection system and method for artificial nerve network model |
| CN108615073A (en) * | 2018-04-28 | 2018-10-02 | 北京京东金融科技控股有限公司 | Image processing method and device, computer readable storage medium, electronic equipment |
| CN108776836A (en) * | 2018-06-08 | 2018-11-09 | 电子科技大学 | A kind of training of the secret protection neural network based on VHE and prediction technique |
| CN109636741A (en) * | 2018-11-14 | 2019-04-16 | 建湖云飞数据科技有限公司 | A kind of image denoising processing method |
| CN109685068A (en) * | 2018-12-27 | 2019-04-26 | 上海七牛信息技术有限公司 | A kind of image processing method and system based on generation confrontation neural network |
| CN109657156A (en) * | 2019-01-22 | 2019-04-19 | 杭州师范大学 | A kind of personalized recommendation method generating confrontation network based on circulation |
| CN110071798A (en) * | 2019-03-21 | 2019-07-30 | 深圳大学 | A kind of equivalent key acquisition methods, device and computer readable storage medium |
| CN110312053A (en) * | 2019-05-10 | 2019-10-08 | 南通理工学院 | Based on sublevel Fourier spectral decomposition and the optical encryption method for mixing random mask |
| CN110490128A (en) * | 2019-08-16 | 2019-11-22 | 南京邮电大学 | A kind of hand-written recognition method based on encryption neural network |
| CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
| CN110633650A (en) * | 2019-08-22 | 2019-12-31 | 首都师范大学 | Convolutional neural network face recognition method and device based on privacy protection |
Non-Patent Citations (2)
| Title |
|---|
| JINGYI LIU: "A digital image encryption algorithm based on bit-planes and an improved logistic map", 《MULTIMEDIA TOOLS AND APPLICATIONS》 * |
| 唐贤伦: "基于条件深度卷积生成对抗网络的图像识别方法", 《自动化学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111563275A (en) * | 2020-07-14 | 2020-08-21 | 中国人民解放军国防科技大学 | Data desensitization method based on generation countermeasure network |
| CN112163229A (en) * | 2020-09-14 | 2021-01-01 | 长春理工大学 | Learning type digital image encryption and decryption method based on loop generation countermeasure network |
| CN112163229B (en) * | 2020-09-14 | 2023-04-28 | 长春理工大学 | Learning type digital image encryption and decryption method based on cyclic generation countermeasure network |
| CN112395635A (en) * | 2021-01-18 | 2021-02-23 | 北京灵汐科技有限公司 | Image processing method, device, secret key generating method, device, training method and device, and computer readable medium |
| CN112733173A (en) * | 2021-01-18 | 2021-04-30 | 北京灵汐科技有限公司 | Image processing method, device, secret key generating method, device, training method and device, and computer readable medium |
| CN112395635B (en) * | 2021-01-18 | 2021-05-04 | 北京灵汐科技有限公司 | Image processing method, device, secret key generating method, device, training method and device, and computer readable 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 |
| CN113407968A (en) * | 2021-06-29 | 2021-09-17 | 平安国际智慧城市科技股份有限公司 | Encryption method, device, equipment and storage medium of target detection model |
| CN113408653A (en) * | 2021-07-12 | 2021-09-17 | 广东电网有限责任公司 | Identification method for adaptively reducing complex light and shadow interference and related device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111259427B (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111259427B (en) | Image processing method and device based on neural network and storage medium | |
| US11902413B2 (en) | Secure machine learning analytics using homomorphic encryption | |
| CN109033854B (en) | Model-based prediction method and device | |
| CN110414567B (en) | Data processing method and device and electronic equipment | |
| CN113239404A (en) | Federal learning method based on differential privacy and chaotic encryption | |
| CN111125727A (en) | Confusion circuit generation method, prediction result determination method, device and electronic equipment | |
| CN112787971B (en) | Construction method of side channel attack model, password attack equipment and computer storage medium | |
| US20190005233A1 (en) | Secure code optimization method and system | |
| CN112101531B (en) | Neural network model training method, device and system based on privacy protection | |
| CN111935179B (en) | Model training method and device based on trusted execution environment | |
| CN114186256B (en) | Training method, device, equipment and storage medium of neural network model | |
| CN111131658B (en) | Image steganography method, device, electronic equipment and medium | |
| US11823060B2 (en) | Method and system for performing deterministic data processing through artificial intelligence | |
| CN111191267A (en) | Model data processing method, device and equipment | |
| CN111813544A (en) | Processing method and device of computing task, scheduling and management system and medium | |
| CN115150063A (en) | Model encryption method and device and electronic equipment | |
| CN113448545B (en) | Method, apparatus, storage medium and program product for machine learning model servitization | |
| US11829468B2 (en) | Neural network confidentiality | |
| US20240007287A1 (en) | Machine learning device, machine learning system, and machine learning method | |
| US10223249B2 (en) | Test methodology for detection of unwanted cryptographic key destruction | |
| CN115021985B (en) | Logistic regression model training method and system without third party participation | |
| CN114817937A (en) | Keyboard encryption method, device, storage medium and computer program product | |
| US20240154802A1 (en) | Model protection method and apparatus | |
| Kumar et al. | Integrating machine learning algorithms with an advanced encryption scheme: enhancing data security and privacy | |
| CN116049840B (en) | Data protection method, device, related equipment and system |
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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: An image processing method, device and storage medium based on Neural Network Effective date of registration: 20220715 Granted publication date: 20201106 Pledgee: Su Qiwen Pledgor: BEIJING ANDE YIZHI TECHNOLOGY Co.,Ltd. Registration number: Y2022990000432 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Granted publication date: 20201106 Pledgee: Su Qiwen Pledgor: BEIJING ANDE YIZHI TECHNOLOGY Co.,Ltd. Registration number: Y2022990000432 |