CN110071798A - A kind of equivalent key acquisition methods, device and computer readable storage medium - Google Patents
A kind of equivalent key acquisition methods, device and computer readable storage medium Download PDFInfo
- Publication number
- CN110071798A CN110071798A CN201910216928.6A CN201910216928A CN110071798A CN 110071798 A CN110071798 A CN 110071798A CN 201910216928 A CN201910216928 A CN 201910216928A CN 110071798 A CN110071798 A CN 110071798A
- Authority
- CN
- China
- Prior art keywords
- neural network
- data set
- training
- equivalent key
- network model
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/80—Optical aspects relating to the use of optical transmission for specific applications, not provided for in groups H04B10/03 - H04B10/70, e.g. optical power feeding or optical transmission through water
- H04B10/85—Protection from unauthorised access, e.g. eavesdrop protection
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/08—Key distribution or management, e.g. generation, sharing or updating, of cryptographic keys or passwords
- H04L9/0861—Generation of secret information including derivation or calculation of cryptographic keys or passwords
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L2209/00—Additional information or applications relating to cryptographic mechanisms or cryptographic arrangements for secret or secure communication H04L9/00
- H04L2209/12—Details relating to cryptographic hardware or logic circuitry
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Security & Cryptography (AREA)
- Theoretical Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Electromagnetism (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of equivalent key acquisition methods, device and computer readable storage mediums, by obtaining preset training data set；The training data set includes the combination of multiple plaintext images with corresponding ciphertext sequence；Neural network is trained based on the training data set, obtains the neural network model of training completion；The neural network model is determined as equivalent key；The equivalent key is used to carry out safety analysis to optical encryption system.Implementation through the invention, a series of known ciphertexts-are trained to being placed in neural network structure in plain text, the mapping relations between ciphertext and plaintext to obtain optical encryption system, using this as equivalent key, improve the efficiency and accuracy of safety analysis, and without whether additional random scrambling or other secondary cryptographic operations is carried out to ciphertext sequence concerning encryption system, extend applicable scene.
Description
Technical field
The present invention relates to optics art of cryptography more particularly to a kind of equivalent key acquisition methods, device and computer
Readable storage medium storing program for executing.
Background technique
Optical image encryption is a kind of novel encryption technology different from traditional mathematics encryption, encrypts skill with traditional mathematics
Art is compared, and optical image encryption technology has various dimensions, large capacity, high robust and natural parallel data processing capacity
Advantage, so that the encryption technology is got the attention in recent years and sustainable development.
However, first concern point should be the safety of the system, it is necessary to by password as a kind of cryptographic system
System carries out stringent safety analysis to confirm whether it is reliable.Currently, being carried out in the related technology to optical encryption system
When safety analysis, it usually needs dependent on the geometric parameter and dependency structure of optical encryption system, by plaintext attack side
Formula obtain system key, attack efficiency and accuracy are lower, and which be only applicable to encryption system not to ciphertext sequence into
The scene of row additional random scrambling or other secondary cryptographic operations is applicable in scene and more limits to.
Summary of the invention
The main purpose of the embodiment of the present invention is to provide a kind of equivalent key acquisition methods, device and computer-readable deposits
Storage media is at least able to solve and adds in the related technology dependent on the geometric parameter of optical encryption system and dependency structure progress optics
The safety analysis of close system, analysis efficiency and accuracy are lower, are applicable in the problem of scene is more limited to.
To achieve the above object, first aspect of the embodiment of the present invention provides a kind of equivalent key acquisition methods, this method
Include:
Obtain preset training data set；The training data set includes multiple plaintext images and corresponding ciphertext sequence
Combination；
Neural network is trained based on the training data set, obtains the neural network model of training completion；
The neural network model is determined as equivalent key；The equivalent key is for pacifying optical encryption system
Full property analysis.
To achieve the above object, second aspect of the embodiment of the present invention provides a kind of equivalent key acquisition device, the device
Include:
Module is obtained, for obtaining preset training data set；The training data set includes multiple plaintext images
With the combination of corresponding ciphertext sequence；
Training module obtains the mind of training completion for being trained based on the training data set to neural network
Through network model；
Determining module, for the neural network model to be determined as equivalent key；The equivalent key is used for optics
Encryption system carries out safety analysis.
To achieve the above object, the third aspect of the embodiment of the present invention provides a kind of electronic device, which includes:
Processor, memory and communication bus；
The communication bus is for realizing the connection communication between the processor and memory；
The processor is above-mentioned any one to realize for executing one or more program stored in the memory
The step of kind equivalent key acquisition methods.
To achieve the above object, fourth aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the meter
Calculation machine readable storage medium storing program for executing is stored with one or more program, and one or more of programs can be by one or more
It manages device to execute, the step of to realize any one of the above equivalent key acquisition methods.
Equivalent key acquisition methods, device and the computer readable storage medium provided according to embodiments of the present invention, passes through
Obtain preset training data set；The training data set includes the combination of multiple plaintext images with corresponding ciphertext sequence；
Neural network is trained based on the training data set, obtains the neural network model of training completion；By the nerve
Network model is determined as equivalent key；The equivalent key is used to carry out safety analysis to optical encryption system.By this hair
A series of known ciphertexts-are trained to being placed in neural network structure, to obtain optical encryption by bright implementation in plain text
Mapping relations between the ciphertext and plaintext of system improve the efficiency of safety analysis and accurate using this as equivalent key
Property, and without additional random scrambling or other secondary cryptographic operations whether is carried out to ciphertext sequence concerning encryption system, it extends
Applicable scene.
Other features of the invention and corresponding effect are described in the aft section of specification, and should be appreciated that
At least partly effect is apparent from from the record in description of the invention.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those skilled in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is the basic procedure schematic diagram for the equivalent key acquisition methods that first embodiment of the invention provides；
Fig. 2 is the optical encryption schematic diagram that first embodiment of the invention provides；
Fig. 3 is the training schematic diagram for the deep neural network model that first embodiment of the invention provides；
Fig. 4 is the schematic diagram for abandon to neural network Regularization that first embodiment of the invention provides；
Fig. 5 is the structural schematic diagram for the equivalent key acquisition device that second embodiment of the invention provides；
Fig. 6 is the structural schematic diagram for the electronic device that third embodiment of the invention provides.
Specific embodiment
In order to make the invention's purpose, features and advantages of the invention more obvious and easy to understand, below in conjunction with the present invention
Attached drawing in embodiment, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described reality
Applying example is only a part of the embodiment of the present invention, and not all embodiments.Based on the embodiments of the present invention, those skilled in the art
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
First embodiment:
In order to solve to carry out optical encryption dependent on the geometric parameter of optical encryption system and dependency structure in the related technology
The safety analysis of system, analysis efficiency and accuracy are lower, are applicable in the technical issues of scene is more limited to, and the present embodiment proposes
A kind of equivalent key acquisition methods, as shown in Figure 1 the basic procedure for equivalent key acquisition methods provided in this embodiment shows
It is intended to, the equivalent key acquisition methods that the present embodiment proposes include the following steps:
Step 101 obtains preset training data set；Training data set includes multiple plaintext images and corresponding ciphertext
The combination of sequence.
Specifically, the present invention is based on the basic norm Kerckhoffs principles that cryptoanalysis field is generally acknowledged, using selection
Plaintext attack scheme, i.e. hypothesis attacker can select a series of given plaintexts and be able to know that its is corresponding close
Text.It should be noted that neural network is trained under the frame of supervised learning, to need to construct training in the present embodiment
Data acquisition system, to train neural network based on multiple training datas in training data set.The training number of the present embodiment
According to being that " plaintext-ciphertext " is right, each specific plaintext image corresponds to a ciphertext sequence.
Optionally, obtaining preset training data set includes: to load random-phase marks in spatial light modulator,
Light source Jing Guo spatial light modulator is encoded；By coding after light source respectively to multiple plaintext images to be encrypted into
Row encryption, obtains ciphertext sequence corresponding to each plaintext image；By the combination of all plaintext images and corresponding ciphertext sequence
It is configured to training data set.
Specifically, communicating pair, when being communicated, first communication party sends a private information encryption to second and leads to
Letter side, the two share a key Key, the one-dimensional sequence { S which is made of N number of numberi(i=1 ..., N).
Before encryption, first communication party is to gather { SiIn N number of element be that seed generates N number of random-phase marksAssuming that { SiAnd random-phase marksBetween relationship be specific, and be only communicate it is double
Fang Suozhi, phase value are evenly distributed on [0,2 π] namely each random-phase marksBoth correspond to { SiOne
A key components.It should be noted that a kind of preferred embodiment as the present embodiment, optical encryption system can be meter
Calculate the optical encryption system of ghost imaging.It is illustrated in figure 2 optical encryption schematic diagram provided in this embodiment, by laser
(Laser) beam of laser emitted passes through spatial light modulator (Spatial Light Modulator, SLM), while will be N number of
Random-phase marks (Random Phase Mask, RPM), which are successively loaded, encodes light source on SLM, the light beam after coding
After the fresnel diffraction for carrying out a distance Z, it is irradiated to plaintext to be encrypted (Plaintext) image T (x, y), then coverlet
Pixel bucket detector (Bucket Detector, BD) detection acquisition, and record its corresponding intensity value.For N number of difference
Phase mask, repetitive operation n times record N number of intensity value BiAn one-dimensional sequence is formed as ciphertext (Ciphertext).It is logical
Chang Eryan, N value are bigger, and quality reconstruction is better.
Step 102 is trained neural network based on training data set, obtains the neural network mould of training completion
Type.
Specifically, based on constructed training data set in the present embodiment, using certain under specific training environment
Optimization algorithm carry out neural network model training, wherein training when learning rate and frequency of training according to actual needs and
It is fixed, it does not limit uniquely herein.In the training process, the ciphertext sequence in training data set is the defeated of neural network model
Enter, and output constraint of the corresponding plaintext image as neural network model, by iterating over training data set
All " plaintext-ciphertexts " is right, determines each training parameter in neural network model, and then obtains the neural network of training completion
Model.
Optionally, neural network includes deep neural network DNN, convolutional neural networks CNN, in Recognition with Recurrent Neural Network RNN
Any one.
Specifically, in practical applications, different types of neural network can be selected according to different usage scenarios, to instruct
The neural network model for practicing the safety analysis for optical encryption system, since there is DNN preferable one-dimensional data to handle energy
Power preferably selects DNN to train neural network model in the present embodiment.
Optionally, neural network be deep neural network when, deep neural network include: an input layer, three it is hidden
Layer, an adjustment layer and an output layer are hidden, the neuron between adjacent layer is connected using full connection mode.
It is illustrated in figure 3 the training schematic diagram of deep neural network model provided in this embodiment.It should be noted that this
Input layer (Input Layer), three hidden layer (Hidden can be determined according to the length of single ciphertext sequence in embodiment
Layer), in output layer (Output Layer) and adjustment layer (Reshaping Layer) independent neuron number, for example:
The size of plaintext image is 28*28, then corresponding ciphertext sequence is 1*784, and the number of our above-mentioned independent neurons will be set
It is set to 784.In a training process, we input the ciphertext sequence of a 1*784 in input layer first, then, by 1*784
Sequence successively pass through input layer, three hidden layers and output layer, and 1 × 784 sequence is adjusted to 28* using an adjustment layer
The output image of 28 pixels.
Optionally, neural network is trained based on training data set, obtains the neural network model of training completion
Include: to input ciphertext sequence in M deep neural network model, passes sequentially through input layer, three hidden layers, output layer and tune
After flood training, the reality output plaintext image of the M times repetitive exercise is obtained, M takes the positive integer more than or equal to 1；Using default
Loss function, reality output plaintext image is compared with the original plaintext image in training data set；It is tied comparing
When fruit meets the preset condition of convergence, M deep neural network model is determined as to the deep neural network model of training completion；
When comparison result is unsatisfactory for the preset condition of convergence, continue the M+1 times repetitive exercise, until meeting the condition of convergence.
Specifically, training data is being input to deep neural network above-mentioned in the present embodiment please continue to refer to Fig. 3,
After training to deep neural network model, output image is compared with original plaintext image, utilizes loss function (Loss
Function) differentiate output image whether meet demand, if conditions are not met, then needing to continue in training data set
" plaintext-ciphertext " to the parameters continued in iteration optimization deep neural network model, until the model can will be in training set
Any ciphertext be converted to the output image met the requirements, i.e., corresponding plaintext image output image similar enough.According to
The training process of above-mentioned neural network successively to training set in sample be trained, complete training set in all samples (i.e.
" plaintext-ciphertext " to) training after, whole process is repeated as many times repeatedly in the present embodiment by as primary complete training process
Generation optimization reaches set the number of iterations or meets the preset standard of loss function (condition of convergence), that is, completes whole
The training process of a deep neural network model.
Optionally, loss function is average absolute value error function, is indicated are as follows:
Wherein, yi' indicate correspond to original plaintext image true value in i-th of element, yiIt indicates to correspond to practical defeated
I-th of element in the output valve of plaintext image out, Δ indicate the average absolute value error of output valve and true value.
Specifically, selecting average absolute value error (Mean Absolute Error, MAE) function in the present embodiment is damage
Function is lost, it is more steady than mean square error (Mean Squared Error, MSE) function because it is not necessarily to square operation.
It should be noted that each neuron is by activation primitive to input in deep neural network model
Information function, and then transmitted by weight connection, and obtain corresponding nonlinear response output.It, can be in the present embodiment
Select sigmoid function for activation primitive, function statement are as follows: yk+1=sigmoid (BiWk+bk), wherein k indicates the number of plies, BiTable
Show the input of some neuron of kth layer, yk+1Indicate the output response of some neuron of kth layer, WkIndicate some nerve of kth layer
The weight of member, bkIndicate that the biasing of some neuron of kth layer, weight and biasing are referred to as the training parameter of neural network.
It should also be noted that output figure is judged in the present embodiment according to the loss function in deep neural network model
Seem it is no meet the requirements, need to optimize network parameter if being unsatisfactory for, the present invention using the adaptive moment estimate
(Adaptive Moment Estimation, Adam) algorithm, alternatively referred to as stochastic gradient descent optimization algorithm, to calculate loss
The gradient that function is closed in entire training dataset is adjusted each using the single order moments estimation and second moment estimation self-adaptive of gradient
The learning rate of network parameter, this method have different parameters different autoadapted learning rates, so that calculation amount obtains greatly
Width reduces, and accelerates the training convergence of deep neural network.
Optionally, be trained based on training data set to neural network includes: to be marked using discarding regularization method
The preset neuron of random drop in each layer of quasi- neural network reduces neuron connection amount；It is right based on training data set
Neural network after discarding neuron is trained.
It is illustrated in figure 4 the schematic diagram provided in this embodiment for abandon to neural network Regularization.Specifically,
In the training process of neural network model, training set is excessive and frequency of training excessively may result in nerve network system and go out
Existing over-fitting.In order to avoid such case, is further used in the present invention and abandon regularization (Dropout
Regularization) technology, for each layer of standard neural network (Standard Neural Net), random drop one
A little units, main function are exactly temporarily to lose the company between a part of neuron in neural metwork training learning process at random
The case where connecing, reducing neuron scale, and then prevent over-fitting.
Optionally, after the neural network model for obtaining training completion, further includes: obtain preset test data set
It closes；Test data set includes the combination of multiple plaintext images with corresponding ciphertext sequence；By the ciphertext sequence in test data set
Column are input to the neural network model of training completion, obtain the plaintext image of test output；Will test output plaintext image with
Plaintext image in test data set carries out relatedness computation；When the degree of correlation is greater than preset relevance threshold, will train
The neural network model of completion is determined as effective neural network model.
Specifically, in the present embodiment after neural network model is completed in training, also using in the verifying of test data set
The validity of neural network model is stated, i.e., by the ciphertext sequence inputting in test data set to trained neural network mould
Type determines its validity by comparing the test plaintext image of its output and the correlation of original plaintext image, in test data
When the degree of correlation between initial data is greater than preset threshold, determine that the neural network model trained is effective, correct mould
Then effective neural network model is determined as equivalent key by type, otherwise, then illustrate that trained neural network model building is deposited
In mistake, need to restart to construct neural network model.
Neural network model is determined as equivalent key by step 103；Equivalent key is for pacifying optical encryption system
Full property analysis.
Specifically, the neural network model in the present embodiment is the equivalent key that can be considered optical encryption system, it is equivalent close
Key refers to the key that can be used for restoring ciphertext.It should be noted that equivalent key acquisition methods of the invention are without knowing meter
The geometric parameter and dependency structure for calculating optical encryption system, regardless of encryption system whether to ciphertext sequence carried out it is additional with
Machine scramble or other common secondary cryptographic operations.Meanwhile if carrying out down-sampled processing to ciphertext in training (reduces training consumption
When, improve attack efficiency), attack method is still effective；If introducing a small amount of noise when cracking subsequent ciphertext, attack method also according to
So effectively.
The equivalent key acquisition methods provided according to embodiments of the present invention, by obtaining preset training data set；Institute
State the combination that training data set includes multiple plaintext images with corresponding ciphertext sequence；Based on the training data set to nerve
Network is trained, and obtains the neural network model of training completion；The neural network model is determined as equivalent key；It is described
Equivalent key is used to carry out safety analysis to optical encryption system.A series of implementation through the invention, by known ciphertexts-
It is trained in plain text to being placed in neural network structure, so that the mapping obtained between the ciphertext and plaintext of optical encryption system is closed
System regard this as equivalent key, improves the efficiency and accuracy of safety analysis, and nothing concerning encryption system whether to close
Literary sequence carries out additional random scrambling or other secondary cryptographic operations, extends applicable scene.
Second embodiment:
In order to solve to carry out optical encryption dependent on the geometric parameter of optical encryption system and dependency structure in the related technology
The safety analysis of system, analysis efficiency and accuracy are lower, are applicable in the technical issues of scene is more limited to, and the present embodiment is shown
A kind of equivalent key acquisition device, specifically refers to Fig. 5, the equivalent key acquisition device of the present embodiment includes:
Module 501 is obtained, for obtaining preset training data set；Training data set include multiple plaintext images with
The combination of corresponding ciphertext sequence；
Training module 502 obtains the nerve of training completion for being trained based on training data set to neural network
Network model；
Determining module 503, for neural network model to be determined as equivalent key；Equivalent key is used for optical encryption system
System carries out safety analysis.
Specifically, including multiple training datas in training data set, the training data of the present embodiment is " plaintext-close
Right, each the specific plaintext image of text ", corresponding ciphertext sequence.In the training process, the ciphertext in training data set
Sequence is the input of neural network model, and output constraint of the corresponding plaintext image as neural network model, is passed through
All " plaintext-ciphertexts " for iterating over training data set is right, determines each training parameter in neural network model, in turn
Obtain the neural network model of training completion.Neural network model in the present embodiment can be considered the equivalent of optical encryption system
Key, equivalent key refer to the key that can be used for restoring ciphertext.It should be noted that one kind as the present embodiment is preferred
Embodiment, optical encryption system can be the optical encryption system for calculating ghost imaging.
In some embodiments of the present embodiment, obtains module 501 and be specifically used for loading random-phase marks in sky
Between on optical modulator, the light source Jing Guo spatial light modulator is encoded；By the light source after coding respectively to multiple to be added
Close plaintext image is encrypted, and ciphertext sequence corresponding to each plaintext image is obtained；By all plaintext images with it is corresponding
The combination of ciphertext sequence is configured to training data set.
In some embodiments of the present embodiment, equivalent key acquisition device further include: test module, it is pre- for obtaining
If test data set；Test data set includes the combination of multiple plaintext images with corresponding ciphertext sequence；By test data
The neural network model that ciphertext sequence inputting in set is completed to training, obtains the plaintext image of test output；It will test defeated
The plaintext image in plaintext image and test data set out carries out relatedness computation；It is greater than the preset degree of correlation in the degree of correlation
When threshold value, the neural network model that training is completed is determined as effective neural network model.
In some embodiments of the present embodiment, neural network includes deep neural network DNN, convolutional neural networks
Any one in CNN, Recognition with Recurrent Neural Network RNN.
Further, when neural network is deep neural network, deep neural network includes: an input layer, three
Hidden layer, an adjustment layer and an output layer, the neuron between adjacent layer are connected using full connection mode.
In some embodiments of the present embodiment, training module 502 is specifically used for defeated in M deep neural network model
Enter ciphertext sequence, after passing sequentially through input layer, three hidden layers, output layer and adjustment layer training, obtains the M times repetitive exercise
Reality output plaintext image, M takes the positive integer more than or equal to 1；Using preset loss function, by reality output plaintext image
It is compared with the original plaintext image in training data set；When comparison result meets the preset condition of convergence, by M depth
Degree neural network model is determined as the deep neural network model of training completion；The preset condition of convergence is unsatisfactory in comparison result
When, continue the M+1 times repetitive exercise, until meeting the condition of convergence.
Further, in some embodiments of the present embodiment, loss function is average absolute value error function, is indicated
Are as follows:
Wherein, y'iIndicate to correspond to i-th of element in the true value of original plaintext image, yiIt indicates to correspond to practical defeated
I-th of element in the output valve of plaintext image out, Δ indicate the average absolute value error of output valve and true value.
In some embodiments of the present embodiment, training module 502 is specifically used for marking using discarding regularization method
The preset neuron of random drop in each layer of quasi- neural network reduces neuron connection amount；It is right based on training data set
Neural network after discarding neuron is trained, and obtains the neural network model of training completion.
It is provided in this embodiment equivalent to should be noted that the equivalent key acquisition methods in previous embodiment can be based on
Key acquisition device realizes that those of ordinary skill in the art can be clearly understood that, for convenience and simplicity of description, this
The specific work process of equivalent key acquisition device described in embodiment, can be with reference to the correspondence in preceding method embodiment
Process, details are not described herein.
Using equivalent key acquisition device provided in this embodiment, by obtaining preset training data set；The instruction
Practice the combination that data acquisition system includes multiple plaintext images with corresponding ciphertext sequence；Based on the training data set to neural network
It is trained, obtains the neural network model of training completion；The neural network model is determined as equivalent key；It is described equivalent
Key is used to carry out safety analysis to optical encryption system.A series of implementation through the invention, by known ciphertext-plaintexts
It is trained to being placed in neural network structure, so that the mapping relations between the ciphertext and plaintext of optical encryption system are obtained,
Using this as equivalent key, improve the efficiency and accuracy of safety analysis, and without concerning encryption system whether to ciphertext
Sequence carries out additional random scrambling or other secondary cryptographic operations, extends applicable scene.
3rd embodiment:
A kind of electronic device is present embodiments provided, it is shown in Figure 6 comprising processor 601, memory 602 and logical
Believe bus 603, in which: communication bus 603 is for realizing the connection communication between processor 601 and memory 602；Processor
601 for executing one or more computer program stored in memory 602, equivalent in above-described embodiment one to realize
At least one step in key acquisition method.
The present embodiment additionally provides a kind of computer readable storage medium, which, which is included in, is used for
Store any method or skill of information (such as computer readable instructions, data structure, computer program module or other data)
The volatibility implemented in art or non-volatile, removable or non-removable medium.Computer readable storage medium includes but not
It is limited to RAM (Random Access Memory, random access memory), ROM (Read-Only Memory, read-only storage
Device), EEPROM (Electrically Erasable Programmable read only memory, band electric erazable programmable
Read-only memory), flash memory or other memory technologies, (Compact Disc Read-Only Memory, CD is only by CD-ROM
Read memory), digital versatile disc (DVD) or other optical disc storages, magnetic holder, tape, disk storage or other magnetic memory apparatus,
Or any other medium that can be used for storing desired information and can be accessed by a computer.
Computer readable storage medium in the present embodiment can be used for storing one or more computer program, storage
One or more computer program can be executed by processor, with realize the method in above-described embodiment one at least one step
Suddenly.
The present embodiment additionally provides a kind of computer program, which can be distributed in computer-readable medium
On, by can computing device execute, to realize at least one step of the method in above-described embodiment one；And in certain situations
Under, at least one shown or described step can be executed using the described sequence of above-described embodiment is different from.
The present embodiment additionally provides a kind of computer program product, including computer readable device, the computer-readable dress
It sets and is stored with computer program as shown above.The computer readable device may include calculating as shown above in the present embodiment
Machine readable storage medium storing program for executing.
As it can be seen that those skilled in the art should be understood that whole or certain steps in method disclosed hereinabove, be
Functional module/unit in system, device may be implemented as the software (computer program code that can be can be performed with computing device
To realize), firmware, hardware and its combination appropriate.In hardware embodiment, the functional module that refers in the above description/
Division between unit not necessarily corresponds to the division of physical assemblies；For example, a physical assemblies can have multiple functions, or
One function of person or step can be executed by several physical assemblies cooperations.Certain physical assemblies or all physical assemblies can be by realities
It applies as by processor, such as the software that central processing unit, digital signal processor or microprocessor execute, or is implemented as hard
Part, or it is implemented as integrated circuit, such as specific integrated circuit.
In addition, known to a person of ordinary skill in the art be, communication media generally comprises computer-readable instruction, data knot
Other data in the modulated data signal of structure, computer program module or such as carrier wave or other transmission mechanisms etc, and
It and may include any information delivery media.So the present invention is not limited to any specific hardware and softwares to combine.
The above content is combining specific embodiment to be further described to made by the embodiment of the present invention, cannot recognize
Fixed specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs,
Without departing from the inventive concept of the premise, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to the present invention
Protection scope.
Claims (10)
1. a kind of equivalent key acquisition methods characterized by comprising
Obtain preset training data set；The training data set includes the group of multiple plaintext images with corresponding ciphertext sequence
It closes；
Neural network is trained based on the training data set, obtains the neural network model of training completion；
The neural network model is determined as equivalent key；The equivalent key is used to carry out safety to optical encryption system
Analysis.
2. equivalent key acquisition methods as described in claim 1, which is characterized in that described to obtain preset training data set
Include:
By random-phase marks load in spatial light modulator, the light source Jing Guo the spatial light modulator is encoded；
Multiple plaintext images to be encrypted are encrypted respectively by the light source after coding, it is right to obtain each plaintext image institute
The ciphertext sequence answered；
All plaintext images are configured to training data set with the combination of the corresponding ciphertext sequence.
3. equivalent key acquisition methods as described in claim 1, which is characterized in that obtaining the neural network mould of training completion
After type, further includes:
Obtain preset test data set；The test data set includes the group of multiple plaintext images with corresponding ciphertext sequence
It closes；
The neural network model that ciphertext sequence inputting in the test data set is completed to the training, obtains testing defeated
Plaintext image out；
Plaintext image in the plaintext image and the test data set of the test output is subjected to relatedness computation；
When the degree of correlation is greater than preset relevance threshold, the neural network model that the training is completed is determined as effectively
Neural network model；
It is described the neural network model is determined as equivalent key to include:
Effective neural network model is determined as equivalent key.
4. equivalent key acquisition methods as described in claim 1, which is characterized in that the neural network be depth nerve net
When network, the deep neural network includes: an input layer, three hidden layers, an adjustment layer and an output layer, adjacent
Neuron between layer is connected using full connection mode.
5. equivalent key acquisition methods as claimed in claim 4, which is characterized in that described to be based on the training data set pair
Neural network is trained, and the neural network model for obtaining training completion includes:
The ciphertext sequence is inputted in M deep neural network model, passes sequentially through the input layer, three hidden layers, output
After layer and adjustment layer training, the reality output plaintext image of the M times repetitive exercise is obtained；
Using preset loss function, by the original plaintext figure in the reality output plaintext image and the training data set
As being compared；
When comparison result meets the preset condition of convergence, the M deep neural network model is determined as training completion
Deep neural network model；
When comparison result is unsatisfactory for the preset condition of convergence, continue the M+1 times repetitive exercise, until meeting the convergence
Condition.
6. equivalent key acquisition methods as claimed in claim 5, which is characterized in that the loss function is average absolute value mistake
Difference function indicates are as follows:
Wherein, y 'iIndicate to correspond to i-th of element in the true value of the original plaintext image, yiIt indicates to correspond to the reality
Border exports i-th of element in the output valve of plaintext image, and Δ indicates the average absolute value of the output valve Yu the true value
Error.
7. equivalent key acquisition methods as described in claim 1, which is characterized in that described to be based on the training data set pair
Neural network, which is trained, includes:
Using regularization method preset neuron of random drop in each layer of standard neural network is abandoned, neuron is reduced
Connection amount；
Based on the training data set, the neural network after discarding neuron is trained.
8. a kind of equivalent key acquisition device characterized by comprising
Module is obtained, for obtaining preset training data set；The training data set include multiple plaintext images with it is right
Answer the combination of ciphertext sequence；
Training module obtains the nerve net of training completion for being trained based on the training data set to neural network
Network model；
Determining module, for the neural network model to be determined as equivalent key；The equivalent key is used for optical encryption
System carries out safety analysis.
9. a kind of electronic device characterized by comprising processor, memory and communication bus；
The communication bus is for realizing the connection communication between the processor and memory；
The processor is for executing one or more program stored in the memory, to realize such as claim 1 to 7
Any one of described in equivalent key acquisition methods the step of.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage have one or
Multiple programs, one or more of programs can be executed by one or more processor, to realize such as claim 1 to 7
Any one of described in equivalent key acquisition methods the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910216928.6A CN110071798B (en) | 2019-03-21 | 2019-03-21 | Equivalent key obtaining method and device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910216928.6A CN110071798B (en) | 2019-03-21 | 2019-03-21 | Equivalent key obtaining method and device and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110071798A true CN110071798A (en) | 2019-07-30 |
| CN110071798B CN110071798B (en) | 2022-03-04 |
Family
ID=67366428
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910216928.6A Active CN110071798B (en) | 2019-03-21 | 2019-03-21 | Equivalent key obtaining method and device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110071798B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674941A (en) * | 2019-09-25 | 2020-01-10 | 南开大学 | Data encryption transmission method and system based on neural network |
| CN111259427A (en) * | 2020-01-21 | 2020-06-09 | 北京安德医智科技有限公司 | Image processing method and device based on neural network and storage medium |
| CN111709867A (en) * | 2020-06-10 | 2020-09-25 | 四川大学 | Novel full convolution network-based equal modulus vector decomposition image encryption analysis method |
| 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 |
| CN112802145A (en) * | 2021-01-27 | 2021-05-14 | 四川大学 | Color calculation ghost imaging method based on deep learning |
| CN113726979A (en) * | 2021-07-31 | 2021-11-30 | 浪潮电子信息产业股份有限公司 | Picture encryption method, decryption method, encryption system and related devices |
| CN116032636A (en) * | 2023-01-06 | 2023-04-28 | 南京通力峰达软件科技有限公司 | Internet of vehicles data encryption method and system based on neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160352520A1 (en) * | 2013-10-29 | 2016-12-01 | Jory Schwach | Encryption using biometric image-based key |
| CN106411510A (en) * | 2016-10-28 | 2017-02-15 | 深圳大学 | Method and apparatus for obtaining equivalent key of random phase coding-based optical encryption system |
| CN107659398A (en) * | 2017-09-28 | 2018-02-02 | 四川长虹电器股份有限公司 | Suitable for Android symmetric encryption method |
| CN108921282A (en) * | 2018-05-16 | 2018-11-30 | 深圳大学 | A kind of construction method and device of deep neural network model |
-
2019
- 2019-03-21 CN CN201910216928.6A patent/CN110071798B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160352520A1 (en) * | 2013-10-29 | 2016-12-01 | Jory Schwach | Encryption using biometric image-based key |
| CN106411510A (en) * | 2016-10-28 | 2017-02-15 | 深圳大学 | Method and apparatus for obtaining equivalent key of random phase coding-based optical encryption system |
| CN107659398A (en) * | 2017-09-28 | 2018-02-02 | 四川长虹电器股份有限公司 | Suitable for Android symmetric encryption method |
| CN108921282A (en) * | 2018-05-16 | 2018-11-30 | 深圳大学 | A kind of construction method and device of deep neural network model |
Non-Patent Citations (2)
| Title |
|---|
| 文洁: "MSE与MAE对机器学习性能优化的作用比较", 《信息与电脑(理论版)》 * |
| 金升箭: "《深度学习 基于MATLAB的设计实例》", 30 April 2018 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110674941A (en) * | 2019-09-25 | 2020-01-10 | 南开大学 | Data encryption transmission method and system based on neural network |
| CN111259427A (en) * | 2020-01-21 | 2020-06-09 | 北京安德医智科技有限公司 | Image processing method and device based on neural network and storage medium |
| CN111259427B (en) * | 2020-01-21 | 2020-11-06 | 北京安德医智科技有限公司 | Image processing method and device based on neural network and storage medium |
| CN111709867A (en) * | 2020-06-10 | 2020-09-25 | 四川大学 | Novel full convolution network-based equal modulus vector decomposition image encryption analysis method |
| CN111709867B (en) * | 2020-06-10 | 2022-11-25 | 四川大学 | Novel full convolution network-based equal-modulus vector decomposition image encryption analysis method |
| 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 |
| CN112802145A (en) * | 2021-01-27 | 2021-05-14 | 四川大学 | Color calculation ghost imaging method based on deep learning |
| CN113726979A (en) * | 2021-07-31 | 2021-11-30 | 浪潮电子信息产业股份有限公司 | Picture encryption method, decryption method, encryption system and related devices |
| CN113726979B (en) * | 2021-07-31 | 2024-04-26 | 浪潮电子信息产业股份有限公司 | Picture encryption method, picture decryption method, picture encryption system and related devices |
| CN116032636A (en) * | 2023-01-06 | 2023-04-28 | 南京通力峰达软件科技有限公司 | Internet of vehicles data encryption method and system based on neural network |
| CN116032636B (en) * | 2023-01-06 | 2023-10-20 | 南京通力峰达软件科技有限公司 | Internet of vehicles data encryption method based on neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110071798B (en) | 2022-03-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110071798A (en) | A kind of equivalent key acquisition methods, device and computer readable storage medium | |
| CN109214973B (en) | Method for generating countermeasure security carrier aiming at steganalysis neural network | |
| US20160012330A1 (en) | Neural network and method of neural network training | |
| CN112862001B (en) | Privacy protection method and system for decentralizing data modeling under federal learning | |
| US12033233B2 (en) | Image steganography utilizing adversarial perturbations | |
| CN115860112B (en) | Model inversion method-based countermeasure sample defense method and equipment | |
| CN113285797B (en) | Multi-image encryption method for optical rotation domain based on compressed sensing and deep learning | |
| CN111681154A (en) | Color image steganography distortion function design method based on generation countermeasure network | |
| Fang et al. | Gifd: A generative gradient inversion method with feature domain optimization | |
| Gu et al. | Federated deep learning with bayesian privacy | |
| CN118070107B (en) | Deep learning-oriented network anomaly detection method, device, storage medium and equipment | |
| Yaras et al. | Randomized histogram matching: A simple augmentation for unsupervised domain adaptation in overhead imagery | |
| CN117350373B (en) | Personalized federal aggregation algorithm based on local self-attention mechanism | |
| CN116743934B (en) | Equal resolution image hiding and encrypting method based on deep learning and ghost imaging | |
| CN113792632A (en) | Finger vein identification method, system and storage medium based on multi-party cooperation | |
| CN111951954A (en) | Body health state detection method and device, readable storage medium and terminal equipment | |
| CN116341004B (en) | Longitudinal federal learning privacy leakage detection method based on feature embedding analysis | |
| Liu et al. | Structure aware visual cryptography | |
| CN111275603B (en) | Security image steganography method based on style conversion and electronic device | |
| CN115374863A (en) | Sample generation method, sample generation device, storage medium and equipment | |
| Hu et al. | Research on encrypted face recognition algorithm based on new combined chaotic map and neural network | |
| Škorić | On the entropy of keys derived from laser speckle; statistical properties of Gabor-transformed speckle | |
| Hu et al. | The recovery scheme of computer-generated holography encryption–hiding images based on deep learning | |
| Zhou et al. | Optical encryption using a sparse-data-driven framework | |
| Hualong et al. | Research on double encryption of ghost imaging by SegNet deep neural network |
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 |