CN114827381A - Strong robustness image steganography method and system based on condition standardization flow model - Google Patents
Strong robustness image steganography method and system based on condition standardization flow model Download PDFInfo
- Publication number
- CN114827381A CN114827381A CN202210754766.3A CN202210754766A CN114827381A CN 114827381 A CN114827381 A CN 114827381A CN 202210754766 A CN202210754766 A CN 202210754766A CN 114827381 A CN114827381 A CN 114827381A
- Authority
- CN
- China
- Prior art keywords
- image
- flow model
- distribution
- conditional
- container
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 66
- 238000009826 distribution Methods 0.000 claims abstract description 79
- 238000013507 mapping Methods 0.000 claims abstract description 32
- 230000008569 process Effects 0.000 claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000009466 transformation Effects 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000004088 simulation Methods 0.000 claims abstract description 12
- 230000006835 compression Effects 0.000 claims description 27
- 238000007906 compression Methods 0.000 claims description 27
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 19
- 230000002441 reversible effect Effects 0.000 claims description 18
- 230000008878 coupling Effects 0.000 claims description 14
- 238000010168 coupling process Methods 0.000 claims description 14
- 238000005859 coupling reaction Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 7
- 238000013139 quantization Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000009795 derivation Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 6
- 230000002829 reductive effect Effects 0.000 abstract description 3
- 230000005540 biological transmission Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 238000013461 design Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 102100025172 Calpain-1 catalytic subunit Human genes 0.000 description 2
- 101710124171 Calpain-1 catalytic subunit Proteins 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 230000001143 conditioned effect Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 230000000670 limiting effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012952 Resampling Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/0042—Fragile watermarking, e.g. so as to detect tampering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
- G06T1/0057—Compression invariant watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0021—Image watermarking
- G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
- G06T1/0064—Geometric transfor invariant watermarking, e.g. affine transform invariant
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/02—Affine transformations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
- H04N1/32272—Encryption or ciphering
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/32267—Methods relating to embedding, encoding, decoding, detection or retrieval operations combined with processing of the image
- H04N1/32277—Compression
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/32—Circuits or arrangements for control or supervision between transmitter and receiver or between image input and image output device, e.g. between a still-image camera and its memory or between a still-image camera and a printer device
- H04N1/32101—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title
- H04N1/32144—Display, printing, storage or transmission of additional information, e.g. ID code, date and time or title embedded in the image data, i.e. enclosed or integrated in the image, e.g. watermark, super-imposed logo or stamp
- H04N1/32149—Methods relating to embedding, encoding, decoding, detection or retrieval operations
- H04N1/3232—Robust embedding or watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0065—Extraction of an embedded watermark; Reliable detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2201/00—General purpose image data processing
- G06T2201/005—Image watermarking
- G06T2201/0201—Image watermarking whereby only tamper or origin are detected and no embedding takes place
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a strong robustness image steganography method based on a conditional standardized flow model, which comprises the following steps: s1, constructing a flow model unit: carrying out distribution transformation on the input host image and the hidden image to convert the host image and the hidden image into high-frequency redundant information and a container image; s2, constructing a distribution mapping module: modeling the distribution of the high-frequency redundant information under the condition dependence of the container image, and mapping the distribution to the standard normal distribution to implicitly store valuable hidden information; s3, constructing a distortion simulation unit: simulating the influence of various distortion interferences in the model training process; s4, constructing an image enhancement module: carrying out primary enhancement denoising processing on the interfered container image; and S5, constructing a condition modulation module: and modulating the network parameters of the flow model unit by taking the interference strength and the type as conditions. The method of the invention solves the defects and shortcomings of the current image steganography in robustness, restoration quality and steganography capacity, and solves the problem that the performance of the prior steganography based on learning is greatly reduced when the steganography is subjected to distortion interference.
Description
Technical Field
The invention relates to the fields of image watermarking, information steganography and the like, in particular to a strong robustness image steganography method and system based on a standardized stream model.
Background
Steganography is a widely studied topic that aims to hide audio, images, and hyperlinks etc. messages in an undiscovered manner into a container. Image steganography takes the hidden image and the host image as input to generate a container image. In its reverse process, only receivers with a specific revealing network have the possibility to reconstruct hidden information from the container image, which is visually identical to the host image [1] . Corresponding to steganography, steganalysis techniques typically distinguish between a container and a host image by color, frequency, and other characteristics [2] . Therefore, the hidden image should be hidden in the invisible field of the container image. It is also valuable to embed as much confidential data as possible in the application into the host image, which is evaluated as payload capacity.
Image steganography aims to maintain hiding power while considering security and imperceptibility to steganalysis. Existing steganographic schemes fail to balance imperceptibility with high payload capacity [3] . Conventional methods convert the hidden message in the spatial or adaptive domain, achieving a capacity below 1bpp per pixel. The hidden data is typically embedded in fewer significant bits or indistinguishable parts, thereby limiting the amount of hidden information capacity. Recent learning-based steganography approaches strive to exploit the latent capabilities of concealment [4] . Most of them treat preprocessing, hiding and displaying as separate modules and design specific networks with independent parameters to handle them.
Recent work attempts have been to standardize flow models [5] Introducing image processing problems such as denoising, resampling and the like to show thatThe flow model has great potential with respect to self-encoders (VAEs), generation countermeasure networks (GANs), and other generation models. In the image steganography task, the hiding and displaying process can be viewed as a pair of inverse problems. Therefore, a flow-based reversible neural network is mathematically suitable for this task.
The main problems of the prior art are as follows: since previous image steganography emphasizes volume and invisibility rather than robustness, and in practice ignores noise and compression disturbances, they are generally sensitive to disturbances in the container media propagation process. The normalized flow model tends to be more susceptible to intermediate distortions due to the dependence on the inherent property of the reversible bijective transformation. The prior art methods, once they apply slight noise or lossy compression on the container, reveal that the concealment is almost unrecognizable, as is the host image at the receiving end. The difficulty in solving the above problems and defects is: even if the network is specifically fine-tuned for predefined noise or JPEG compression levels, the reconstruction quality and generalization are still limited.
The significance of solving the problems and the defects is as follows: the invention aims to construct an image steganography model with strong robustness, which can effectively inhibit the interference of distortion such as noise, compression and the like in the media transmission process on a container image, enhance the restoration quality and the universality of the image steganography under various complex conditions, promote the steganography information capacity under limited conditions and contribute to the popularization and the application of the image steganography.
Disclosure of Invention
The invention provides a strong robustness image steganography method and a system based on a conditional standardized flow model, wherein a reversible neural network based on the conditional flow model is used for steganography of a hidden image into a container image, and the method is robust to noise, compression and other interference, so that the defects and the defects of the current image steganography in robustness, restoration quality and steganography capacity are overcome, and the problem that the performance of the steganography based on learning is greatly reduced when the steganography is subjected to distortion interference is solved.
The technical scheme of the invention is as follows: according to one aspect of the invention, a strong robustness image steganography method based on a condition standardization flow model is provided, which comprises the following steps: s1, constructing a flow model unit: carrying out distribution transformation on the input host image and the hidden image to convert the host image and the hidden image into high-frequency redundant information and a container image; s2, constructing a distribution mapping module: under the inspiration of the condition standardization flow, the distribution of high-frequency redundant information under the condition dependence of container images is modeled, and the high-frequency redundant information is mapped to the standard normal distribution to implicitly store valuable hidden information; s3, constructing a distortion simulation unit: simulating the influence of various distortion interferences in the model training process; s4, constructing an image enhancement module: carrying out primary enhancement denoising processing on the interfered container image; and S5, constructing a condition modulation module: and modulating the network parameters of the flow model unit by taking the interference strength and the type as conditions.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model, in step S1, one or more hidden graphs and host graphs are used as forward inputs, a neural network constructed based on the flow model is used for forward mapping to a specific distribution, and a part of the output is a container graph carrying hidden graph information, and a part of the output is a high-frequency redundant distribution; by increasing the number of branch channels of the stream model, the stream model can be modulated by external conditions, and a plurality of hidden images can be easily hidden in one container.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model, in step S2, the distribution mapping module is constructed based on the strong and reversible conditional normalized flow model in which the affine coupling part is merged into the container map as the condition information.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model, in step S3, the differentiable operation is adopted to simulate the effects of gaussian noise, poisson noise and various distortion interferences of JPEG compression.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model, in step S3, a differentiable analog unit for JPEG compression is adopted, a JPEG compression operation of replacing the corresponding quality factor with the differentiable analog unit in a training process, and quantization is replaced with fourier transform in a derivation process to approximate.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model, in step S4, the structure of the network of the image enhancement module is convolution, normalization, and structure of cascade of ReLU activation functions, residual learning is used at the output of the network inside the image enhancement module, for each convolutional layer, a convolutional kernel of 3 × 3 size is used, the step size is set to 1, the receptive field is expanded layer by layer, the number of layers is set to 17, and the number of convolutional kernels of each convolutional layer is set to 64.
Preferably, in the strong robustness image steganography method based on the conditional normalized flow model described above, in step S5, given a distortion level and a type, network parameters of flow model elements in a conditional modulation module will vary from distortion to distortion, the conditional modulation module is disposed on an affine coupling layer of the flow model elements, a noise level or JPEG quality factor QF is input into a fully connected layer, weights are generated by a conditional guided network, and the weights modulate the coupling parameters in the network of flow model elements by affine transformation, thereby controlling feature transformation of the flow model elements.
According to another aspect of the present invention, there is provided a strong robustness image steganography system based on a conditional normalized flow model, comprising: the system comprises a flow model unit, a distribution mapping module, a distortion simulation unit, an image enhancement module and a condition modulation module, wherein the flow model unit is used for carrying out distribution transformation on an input host image and a hidden image and converting the host image and the hidden image into high-frequency redundant information and a container image; the distribution mapping module is used for modeling the distribution of the high-frequency redundant information under the condition dependence of the container image under the inspiration of the condition standardized stream, and mapping the distribution to the standard normal distribution to implicitly store valuable hidden information; the distortion simulation unit is used for simulating the influence of various distortion interferences in the model training process; the image enhancement module is used for carrying out primary enhancement denoising processing on the interfered container image; and the condition modulation module is used for modulating the network parameters of the flow model unit by taking the interference intensity and the type as conditions so as to adapt to different distortion conditions for parameter adjustment.
According to the technical scheme, the method has the beneficial effects that the conditional standardized stream model is introduced to construct the steganography system, and the strong generation capacity of the steganography system allows the higher reduction quality to be still maintained under the high steganography capacity. In a lossless or distorted environment, the method has excellent robustness, and simultaneously maintains the imperceptibility and high-capacity characteristics of steganography, the high robustness of the method can expand the steganography to wider application fields such as face change detection and gray level coloring, GIF and short videos can be compressed into snapshots which are robust to loss compression, the snapshots/thumbnails are visually acceptable and can be used as previews under low data conditions, and video frames can be reconstructed and restored from the snapshots/thumbnails.
Drawings
For a better understanding and appreciation of the concepts, principles of operation, and effects of the invention, reference will now be made in detail to the present embodiments of the invention, examples of which are illustrated in the accompanying drawings.
FIG. 1 is a flow diagram of a strong robustness image steganography method based on a conditional normalized flow model of the present invention;
FIG. 2 is an overall framework diagram of the strong robustness image steganography system of the present invention based on a conditional normalized flow model;
FIG. 3 is a schematic diagram of the distribution mapping module of the present invention predicting redundancy distributions;
FIG. 4 is a block diagram of the flow model modulation based on the interference strength in the present invention;
FIG. 5 is a table showing the output results of steganography under typical interference in embodiment 1 of the present invention;
FIG. 6 is a schematic flow chart of steganography in a real scene according to embodiment 2 of the present invention;
fig. 7 is a schematic flowchart of a tamper detection application according to embodiment 3 of the present invention.
Detailed Description
In order to make the objects, technical means and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific examples. These examples are merely illustrative and not restrictive of the invention.
The main object of the present invention is to design a universal and robust framework for image steganography at various distortions. As shown in fig. 2, it will be severalA hidden imageWith the host imageEmbedding in an information container imageAt the receiving end, the container map can be obtained from the disturbed container mapRestore the hidden pictureAnd a host imageAnd the method has the capability of resisting image distortion. For training stability, the framework of the invention directly learns hidden imagesHost imageAnd container imageCan contain information of a plurality of hidden images while maintaining the same as the host imageThe same appearance.
With reference to fig. 1 and fig. 2, the strong robustness image steganography method based on the conditional normalized flow model of the present invention includes the following steps:
s1, constructing a flow model unit: and carrying out distribution transformation on the input host image and the hidden image, and converting the host image and the hidden image into high-frequency redundant information and a container image.
In FIG. 2, one or more hidden pictures are shownAnd a host imageAs the forward input of the flow model unit, the neural network forward mapping constructed based on the flow model is adopted as the specific distribution, one part of the output is a container graph y carrying hidden graph information, and the other part is high-frequency redundancy distribution h r . And taking the characteristics obtained by the characteristic extractor in the container diagram as auxiliary conditions, and transforming the high-frequency redundant distribution into approximate Gaussian distribution by a distribution mapping (CANP) module. During model training, Gaussian noise, Poisson noise, JPEG compression and the like which are frequently encountered in media transmission are added to a container graph in a simulation mode. Receiver versus disturbed container mapFirstly, denoising and deblurring processing are carried out in an image enhancement module to obtain an enhanced container image, the container image is input as one part of a flow model unit (reverse), the other part of the container image is input with reconstructed high-frequency distribution which is obtained by Gaussian distribution through a reverse distribution mapping module, and a reconstructed host image and a reconstructed hidden image are output in the reverse process of the flow model unit. This step can easily hide a plurality of hidden images into one container by increasing the number of branch channels of the flow model to increase the number of hidden images. Therefore, the scheme of the invention has higher steganographic information capacity, so that the scheme becomes an advanced image steganographic technology.
The standardized flow model is naturally and intuitively applicable to the image steganography task due to its reversibility. The forward mapping is a process of hiding information and the reverse mapping is a process of restoring information, both of which are ideally reversible processes with shared parameters and should be considered as forward and backward processes of the standardized flow to achieve end-to-end optimization. The flow model has two main features: inference functionThe logarithm determinant of (2) is simple to calculate; corresponding inverse functionAnd the solution is easy.
In fig. 3, in the k-th block of the flow model unit,is split intoAnd. They will then pass through an affine coupling, in whichAndconstructed from dense blocks with ReLU activation, inResulting in scaling and skewing. During the operation of each step of the flow model, two parts of information are calculated in the following mode:
it is clear that the above affine coupling layer is mathematically invertible and has a triangular jacobian whose logarithmic determinant is easy to calculate.
In the overall design of the system, various constraints in the steganography system are met by designing a strict and comprehensive loss function. First, the present invention requires a restored host imageAnd a hidden imageShould be as close as possible to the originally input host imageAnd a hidden image. The invention herein usesTo minimize the average distance between each pair of restored images and the original image. At the same time, the image reconstruction module in step S3 is required to be able to reconstruct the container image from the distorted container map as far as possible, for which purpose the image enhancement loss is set. Since the distribution of the data can be explicitly modeled in the standardized flow model, the distribution mapping module of step S2 expects to output the distributionClose to the standard normal distribution, and are distributed with high frequency redundancy by the constraintAnd distribution of container mapsDecoupling by distributed lossesControl byCross entropy above to describe the distance between distributions. Loss of container mapThe system requires a container imageSpatially and frequency domain with the host imageApproximately the same, the present invention further applies Fast Fourier Transform (FFT) extractionThe high frequency components in the container map and the host map are emphasized to be as close as possible. The embedded image display, container invisibility, distortion enhancement and noise distribution distances are as follows:
the overall system penalty function takes into account the above four components: embedded image display, container invisibility, distortion enhancement and noise distribution distance, the overall system loss function is as follows:
in a specific implementation, settingS2, constructing a distribution mapping module: under the inspiration of the conditional standardized stream, the distribution of high-frequency redundant information under the condition dependence of container images is modeled and is mapped to the standard normal distribution so as to implicitly store valuable hidden information.
And modeling the distribution of the high-frequency redundant information under the condition dependence of the container image through a conditional flow model, mapping the forward-mapped high-frequency redundant distribution into standard normal distribution, and implicitly storing valuable hidden information. The conditional normalized flow model is introduced in step S2, and the basic information for display is implicitly retained by adjusting the high-frequency distribution conditioned on the container image so that the front flow model unit does not directly output a normal distribution but outputs a high-frequency redundant distribution associated with the image.
The previous normalized flow models generally convert the input image into a target image and a gaussian noise distribution. However, due to the limited network depth and mapping capability of the flow model, directly using the gaussian distribution as the target output may result in non-ideal results, and high requirements are put on the conversion capability of the flow model.
To solve this problem, the present invention finds that the high-frequency redundant information of the image is useful for the steganographic reconstruction process, and a better decoupling process should be obtained, as shown in fig. 3. In the conditioned stream introduced by the present invention, a high frequency output is assumedDependent on the container. After training, the forward process will compress the input host and hidden image pairsAnd convert it into a container imageAnd high frequency redundancyWherein the containerIs restricted to be close to the host imageAnd at the same time contain a compound fromInformation of. Ideally, the present invention contemplates model generation and real host imagesIdentical container. This can be expressed as a dirac delta function in the equationAnd further approximated by a gaussian distribution as:
this process is called a distributed mapping module, in which high frequency redundancy is usedBy container drawingAs a conditional feature, projected to a standard normal distribution by distribution mapping. In the overall forward process, the system inputs image pairsDecomposition into container images following a simple distributionAnd high frequency redundancy distributionThen transformed into a container mapAnd standard Gaussian distribution. At the receiving end, given the received containerAnd from a standard Gaussian distributionRandom sample obtained by middle samplingThe distribution mapping can be reconstructedAnd then generated in back propagation。
Specifically, as shown in FIG. 3, high frequency redundancy distributionAfter the treatment of activating normalization, the channel exchange is carried out in 1-by-1 convolution, then the whole affine modulation is carried out through an affine embedding layer, and then the channel is divided intoAndthe two parts enter into the flow model operation. Branching of input dataAnd images from containersCondition feature extracted fromMerged and then used as a convolutional neural networkOf a convolutional networkThe coefficients used for affine transform multiplication and addition in the flow model will be generated. High frequency redundancy distribution of inputs via stacking of multiple layer flow modelsCan be changed into toIs a condition-dependent Gaussian distribution. The invention is based on a powerful and still reversible condition standardization flow model, wherein a container graph is mainly fused into an affine coupling part as condition information, and a distribution mapping module, called CANP for short, is constructed. Since each stream block starts with a permutation operation such as convolution,andwill all receiveThe conditional impact of (a), expressed mathematically as:
s3, constructing a distortion simulation unit: and simulating the influence of various distortion interferences in the model training process.
The container diagram is generally interfered by noise, JPEG compression and the like in the medium transmission process, and differentiable operation is adopted to simulate the influence of various distortion interferences such as Gaussian noise, Poisson noise, JPEG compression and the like in the model training process. In the step S3, end-to-end training of the model is realized by simulating common distortion types such as noise and JPEG compression, and due to the addition of distortion simulation, the system does not have complete symmetry, so that the network parameters of forward and reverse mapping of the flow model are not completely equal during training.
Tolerance to JPEG compression and noise is a fundamental problem in high robustness steganography. Noise is easily modeled and jointly trained in deep model training, whereas JPEG is not. The JPEG processing flow consists of four main steps: color space transform, Discrete Cosine Transform (DCT), quantization and entropy coding. In practice, quantization is a lossy and non-differentiable step in JPEG compression. Thus, JPEG is not suitable for direct end-to-end optimization. In order to train the JPEG operation, the invention introduces a differentiable analog unit for JPEG compression, the differentiable analog unit replaces the JPEG compression operation of the corresponding quality factor in the training process, and the quantization is replaced by Fourier transform in the derivation process for approximate processing.
S4, constructing an image enhancement module: and carrying out primary enhancement denoising processing on the interfered container image.
After the reconstruction network receives the container image with the interference, the image enhancement module preliminarily removes the noise and the compressed degradation effect. Step S4 designs a corresponding enhancement module to perform preliminary processing on the container image after the distortion interference. Due to the reversible bijection property of the flow model, the enhanced equal-asymmetry module is not completely reversible in the system, and certain relaxation treatment is carried out on the parameters of the flow model of the system.
As shown in the block diagram of fig. 2, the receiving end of the present invention uses an image enhancement module as a preprocessing unit to remove the effect of image distortion such as noise and JPEG in advance before the stream model. The image enhancement is integrated into the system, and after the receiving end obtains the interfered image, the network preprocesses the distorted image to realize the denoising and JPEG deblocking effect.
The structure of the network of the image enhancement module is a structure of convolution, normalization and ReLU activation function cascade, and residual learning is used for the output of the network inside the image enhancement module. For each convolutional layer, a convolutional kernel of 3 × 3 size is adopted, the step length is set to be 1, the receptive field is expanded layer by layer, the number of layers is set to be 17, and the number of convolutional kernels of each convolutional layer is set to be 64.
In order to adapt the flow model to distortions on the container image, a decoupled training scheme is introduced into the second half of the training process. In the flow-based model, the forward and reverse transfer are theoretically symmetric and equal in parameters. However, the work flow of the invention has irreversible operations of quantization, noise interference, image enhancement network and the like. These non-reversible processes on the intermediate container image require corresponding adjustments to be made to the flow model. The invention provides a scheme that forward and backward parameters are not completely equal to a certain degree at the later stage of model training to expand the tolerance of the flow model, and the relaxation of parameter constraint brings a change space for the forward and backward propagation of the network, so that the original strictly symmetrical flow model has stronger adaptability to complex network transmission conditions.
S5, constructing a condition modulation module: and modulating the network parameters of the flow model unit under the condition of the interference strength and the type so as to adapt to different distortion conditions for parameter adjustment.
Step S5 proposes a conditional modulation module on the standardized stream to modulate the network parameters for parameters of different distortion types and levels, making the present invention a generic controllable model for distortion interference of various types and levels.
In real world applications, it is impractical to train specific network parameters for each type and level of distortion. For general image steganography, the invention enables the network parameters of the steganography system to be dynamically adjusted according to different types and intensities of distortion. Here, the invention proposes a distortion-guided conditional modulation module to control the affine coupling layer to process container images corrupted by gaussian noise or JPEG compression artifacts. In particular, given distortion levels and types: (Representing gaussNoise, QF for JPEG compression), the network parameters of the flow model unit in the conditional modulation module will vary from distortion to distortion. The conditional modulation module is disposed on an affine coupling layer of the flow model element. As shown in FIG. 4, similar to the previous flow model element, the input data is divided into segments after being normalized by activation and convolved by 1 x 1Andan affine transformation coupling is carried out and,the parameters of affine transformation of a channel are transformed by another channelBy convolutional networksAnd a condition modulation module for obtaining the information,the affine transformation of (1) is also such that k becomes k +1 after affine coupling of such a layer. In the conditional modulation module, a noise level or a JPEG quality factor QF is input into the full connection layer FC, the whole conditional guide network is in a residual block structure, and after convoluting Conv1 and activating Act and Conv2, the residual block structure is multiplied by the output of the full connection layer FC to generate a weightThe weights modulate the coupling parameters in the network of flow model elements by affine transformation, thereby controlling the feature transformation of the flow model elements. In this way, the invention can adapt to various distortion conditions through a single model under the coordination of distortion conditions.
The invention discloses a strong robustness image steganography system based on a condition standardization flow model, which comprises the following steps: a flow model unit, a distribution mapping module, a distortion simulation unit, an image enhancement module and a condition modulation module, wherein,
the flow model unit is used for carrying out distribution transformation on the input host image and the hidden image and converting the host image and the hidden image into high-frequency redundant information and a container image;
the distribution mapping module is used for modeling the distribution of the high-frequency redundant information under the condition dependence of the container image under the inspiration of the condition standardized stream, and mapping the distribution to the standard normal distribution to implicitly store valuable hidden information;
the distortion simulation unit is used for simulating the influence of various distortion interferences in the model training process;
the image enhancement module is used for carrying out primary enhancement denoising processing on the interfered container image; and
and the condition modulation module is used for modulating the network parameters of the flow model unit by taking the interference strength and the type as conditions so as to adapt to different distortion conditions for parameter adjustment.
Example 1: image steganography under common transmission distortion interference
As shown in fig. 5, this embodiment is a typical application scenario for which the present invention is directed. The hidden and host images are extracted from the DIV2K dataset, and the container image with hidden information is generated by the invention, and noise and JPEG compression are added on the container image to represent common transmission distortion interference. Model training with BatchSize =16 on NVIDIATeslaV100 graphic computing card, hyper-parameter β 1 Is set to 0.9, beta 2 Set to 0.99 and the learning rate to 0.0001. The performance advantages of the invention (RIIS) and other steganography methods are shown in fig. 5, the same Host map (Host) and hidden image (Secret) are input, Container image (Container) and hidden image (concealed Secret) are output, and from the reconstructed result and error map (Residual), the invention can still reconstruct hidden information reliably under the interference of Noise (Noise σ = 10) and compression (JPEG QF = 90), the reconstructed result has richer texture and clearer edges, and the overall reconstruction error level is also much lower.
Example 2: steganography of image information on real screen
As shown in fig. 6, the method of the present invention can restore a hidden image from a photograph with strong robustness. Given a host graph and a hidden graph, the container graph embedded with hidden information is displayed on a screen or printed on paper, and a picture is taken from a display by a CMOS sensing device, and in the process, the container graph is influenced by various distortion interferences such as lens distortion, moire, sensor noise, motion blur and the like. By means of the strong robustness of the invention, the system can still recover most information of the hidden image from the picture to obtain the restored host picture and the restored hidden picture. This extends the range of steganography applications from digital media to real scenes, helping to achieve the vision of implicitly connecting cyberspace to the real world, and to assist in copyright and integrity protection of digital assets and artwork.
Example 3: detection of face-changing or other image tampering
As shown in fig. 7, the present invention can be used for copyright watermark protection of images on internet media, and detecting common tampering such as face change. For a batch of images needing protection, the invention takes the images as host images, and embeds watermark templates into each host image to obtain a protection processed container map. The container map may be subjected to tampering processes such as face changing and the like, and meanwhile, degradation interference such as noise, compression and the like in network transmission can be received, so that the method can input the recovered container map into a watermark recovery process, compared with an original watermark template, a watermark obtained after reverse reconstruction and tampering, and an unaligned region such as black spots and the like can appear in a corresponding position, and the unaligned region can be positioned through a feature alignment algorithm to detect the tampered and operated region. Experiments show that the detection scheme is effective and accurate under the interference of noise, compression and the like.
From the results of the embodiments 1 to 3, it can be known that the method of the present invention is robust to noise, compression, and other interferences, so as to solve the defects and deficiencies of the current image steganography in robustness, restoration quality, and steganography capacity, and solve the problem that the performance of the prior learning-based steganography is greatly reduced when the steganography is subjected to distortion interference.
The foregoing description is of the preferred embodiment of the concepts and principles of operation in accordance with the invention. The above-described embodiments should not be construed as limiting the scope of the claims, and other embodiments and combinations of implementations according to the inventive concept are within the scope of the invention.
Reference documents:
[1] Ijka B , Pp A , Pj V A , et al. Comprehensive survey of image steganography: Techniques, Evaluations, and trends in future research[J]. Neurocomputing, 2019, 335:299-326.
[2] Barbier J , Mayoura K . Steganalysis of Multi Bit Plane Image Steganography[C]// International Workshop on Digital Watermarking. Springer-Verlag, 2008.
[3] Tsai P , Hu Y C , Yeh H L . Reversible image hiding scheme using predictive coding and histogram shifting[J]. Signal Processing, 2009, 89(6):1129-1143.
[4] Baluja S . Hiding Images Within Images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019:1-1.
[5] Kingma D P , Salimans T , Jozefowicz R , et al. Improving Variational Inference with Inverse Autoregressive Flow[J]. 2016.
Claims (8)
1. the strong robustness image steganography method based on the conditional normalized flow model is characterized by comprising the following steps of:
s1, constructing a flow model unit: carrying out distribution transformation on the input host image and the hidden image to convert the host image and the hidden image into high-frequency redundant information and a container image;
s2, constructing a distribution mapping module: under the inspiration of the condition standardization flow, the distribution of high-frequency redundant information under the condition dependence of container images is modeled, and the high-frequency redundant information is mapped to the standard normal distribution to implicitly store valuable hidden information;
s3, constructing a distortion simulation unit: simulating the influence of various distortion interferences in the model training process;
s4, constructing an image enhancement module: carrying out primary enhancement denoising processing on the interfered container image; and
s5, constructing a condition modulation module: and modulating the network parameters of the flow model unit under the condition of the interference strength and the type.
2. A strong robustness image steganography method based on conditional normalized flow model is characterized in that in step S1, one or more hidden graphs and host graphs are used as forward input, a neural network constructed based on the flow model is used for forward mapping to be a specific distribution, part of the output is a container graph carrying hidden graph information, and part of the output is a high frequency redundancy distribution.
3. A strong robustness image steganography method based on conditional normalized flow model according to claim 1, wherein in step S2, the distribution mapping module is constructed based on a strong and reversible conditional normalized flow model with affine coupling part merged into a container map as condition information.
4. The method for steganography of a strongly robust image based on a conditional normalized flow model according to claim 1, wherein in step S3, the influence of various types of distortion interference of gaussian noise, poisson noise and JPEG compression is simulated by using differentiable operations.
5. A strong robustness image steganography method based on conditional normalized flow model according to claim 1 wherein in step S3, differentiable analog units for JPEG compression are employed, JPEG compression operation of respective quality factors are replaced with the differentiable analog units in training process, quantization is replaced with fourier transform in derivation process for approximation processing.
6. A strong robustness image steganography method based on conditional normalized flow model, as claimed in claim 1, wherein in step S4, the structure of the network of the image enhancement module is convolution, normalization, structure of cascade of ReLU activation functions, the inside of the image enhancement module uses residual learning at the output of the network, for each convolutional layer, a convolution kernel of 3 × 3 size is adopted, the step size is set to 1, the receptive field is expanded layer by layer, the number of layers is set to 17, and the number of convolution kernels of each convolutional layer is set to 64.
7. A strong robustness image steganography method based on conditional normalized flow model, according to claim 1, characterized in that in step S5, given distortion level and type, network parameters of flow model elements in the conditional modulation module will vary from distortion to distortion, the conditional modulation module is deployed on affine coupling layer of the flow model elements, noise level or JPEG quality factor QF is inputted into full connection layer, weight is generated through conditional guided network, weight modulates coupling parameters in network of the flow model elements through affine transformation, thereby controlling feature transformation of the flow model elements.
8. A strong robustness image steganography system based on a conditional normalized flow model is characterized by comprising the following steps: a flow model unit, a distribution mapping module, a distortion simulation unit, an image enhancement module and a condition modulation module, wherein,
the flow model unit is used for carrying out distribution transformation on the input host image and the hidden image and converting the host image and the hidden image into high-frequency redundant information and a container image;
the distribution mapping module is used for modeling the distribution of the high-frequency redundant information under the condition dependence of the container image under the inspiration of the condition standardized stream, and mapping the distribution to the standard normal distribution so as to implicitly store valuable hidden information;
the distortion simulation unit is used for simulating the influence of various distortion interferences in the model training process;
the image enhancement module is used for carrying out primary enhancement denoising processing on the interfered container image; and
and the conditional modulation module is used for modulating the network parameters of the flow model unit by taking the interference intensity and the type as conditions so as to adapt to different distortion conditions for parameter adjustment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210754766.3A CN114827381A (en) | 2022-06-30 | 2022-06-30 | Strong robustness image steganography method and system based on condition standardization flow model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210754766.3A CN114827381A (en) | 2022-06-30 | 2022-06-30 | Strong robustness image steganography method and system based on condition standardization flow model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114827381A true CN114827381A (en) | 2022-07-29 |
Family
ID=82523397
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210754766.3A Pending CN114827381A (en) | 2022-06-30 | 2022-06-30 | Strong robustness image steganography method and system based on condition standardization flow model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114827381A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116543388A (en) * | 2023-07-04 | 2023-08-04 | 深圳大学 | Conditional image generation method and related device based on semantic guidance information |
CN117611422A (en) * | 2024-01-23 | 2024-02-27 | 暨南大学 | Image steganography method based on Moire pattern generation |
CN117939027A (en) * | 2024-03-21 | 2024-04-26 | 南京信息工程大学 | JPEG image steganography method and system based on DCT feature extraction |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112132737A (en) * | 2020-10-12 | 2020-12-25 | 中国人民武装警察部队工程大学 | Reference-free generated image robust steganography method |
US20210192019A1 (en) * | 2019-12-18 | 2021-06-24 | Booz Allen Hamilton Inc. | System and method for digital steganography purification |
CN114140309A (en) * | 2021-12-03 | 2022-03-04 | 中国人民武装警察部队工程大学 | Novel image steganography method and system based on NICE model |
-
2022
- 2022-06-30 CN CN202210754766.3A patent/CN114827381A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210192019A1 (en) * | 2019-12-18 | 2021-06-24 | Booz Allen Hamilton Inc. | System and method for digital steganography purification |
CN112132737A (en) * | 2020-10-12 | 2020-12-25 | 中国人民武装警察部队工程大学 | Reference-free generated image robust steganography method |
CN114140309A (en) * | 2021-12-03 | 2022-03-04 | 中国人民武装警察部队工程大学 | Novel image steganography method and system based on NICE model |
Non-Patent Citations (2)
Title |
---|
KAI ZHANG,ET AL: "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
YOUMIN XU,ET AL: "Robust Invertible Image Steganography", 《PROCEEDINGS OF THE IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR),2022》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116543388A (en) * | 2023-07-04 | 2023-08-04 | 深圳大学 | Conditional image generation method and related device based on semantic guidance information |
CN116543388B (en) * | 2023-07-04 | 2023-10-17 | 深圳大学 | Conditional image generation method and related device based on semantic guidance information |
CN117611422A (en) * | 2024-01-23 | 2024-02-27 | 暨南大学 | Image steganography method based on Moire pattern generation |
CN117611422B (en) * | 2024-01-23 | 2024-05-07 | 暨南大学 | Image steganography method based on Moire pattern generation |
CN117939027A (en) * | 2024-03-21 | 2024-04-26 | 南京信息工程大学 | JPEG image steganography method and system based on DCT feature extraction |
CN117939027B (en) * | 2024-03-21 | 2024-06-07 | 南京信息工程大学 | JPEG image steganography method and system based on DCT feature extraction |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wan et al. | A comprehensive survey on robust image watermarking | |
Emad et al. | A secure image steganography algorithm based on least significant bit and integer wavelet transform | |
Roy et al. | A hybrid domain color image watermarking based on DWT–SVD | |
Amrit et al. | Survey on watermarking methods in the artificial intelligence domain and beyond | |
CN114827381A (en) | Strong robustness image steganography method and system based on condition standardization flow model | |
Roy et al. | An SVD based location specific robust color image watermarking scheme using RDWT and Arnold scrambling | |
Keshavarzian et al. | ROI based robust and secure image watermarking using DWT and Arnold map | |
CN115829819A (en) | Neural network-based image robust reversible information hiding method, device and medium | |
Xu et al. | A compact neural network-based algorithm for robust image watermarking | |
CN113902647A (en) | Image deblurring method based on double closed-loop network | |
CN115908095A (en) | Hierarchical attention feature fusion-based robust image watermarking method and system | |
Vaidya et al. | Imperceptible watermark for a game-theoretic watermarking system | |
Zhou et al. | Geometric correction code‐based robust image watermarking | |
Agarwal et al. | Digital watermarking in the singular vector domain | |
Singh et al. | A secured robust watermarking scheme based on majority voting concept for rightful ownership assertion | |
Mohamed et al. | A Survey on Image Data Hiding Techniques | |
Chawla et al. | A modified secure digital image steganography based on DWT using matrix rotation method | |
Rebahi et al. | Image Watermarking Technique Using Convolutional Autoencoder | |
Ben Jabra et al. | Deep Learning-Based Watermarking Techniques Challenges: A Review of Current and Future Trends | |
Shahi et al. | High Capacity Reversible Steganography on CMY and HSI Color Images Using Image Interpolation | |
Allwadhi et al. | A comprehensive survey of state-of-art techniques in digital watermarking | |
Ayalneh et al. | JPEG copy paste forgery detection using BAG optimized for complex images | |
Singh et al. | Robust data hiding scheme through distinct keypoint selection exploiting modified Bilateral-Laplacian SIFT with encoding pipeline | |
Alaa Sabri et al. | A New Algorithm for a Steganography System | |
You et al. | Two-Stage Watermark Removal Framework for Spread Spectrum Watermarking |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220729 |