CN112487369A - Frame loss resistant GIF dynamic image copyright authentication method - Google Patents
Frame loss resistant GIF dynamic image copyright authentication method Download PDFInfo
- Publication number
- CN112487369A CN112487369A CN202011481416.1A CN202011481416A CN112487369A CN 112487369 A CN112487369 A CN 112487369A CN 202011481416 A CN202011481416 A CN 202011481416A CN 112487369 A CN112487369 A CN 112487369A
- Authority
- CN
- China
- Prior art keywords
- image
- frame
- gif
- dynamic image
- watermark
- 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 37
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000012217 deletion Methods 0.000 claims abstract description 20
- 230000037430 deletion Effects 0.000 claims abstract description 20
- 238000004088 simulation Methods 0.000 claims abstract description 8
- 238000005457 optimization Methods 0.000 claims abstract 2
- 238000000605 extraction Methods 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 239000006002 Pepper Substances 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000011068 loading method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/10—Protecting distributed programs or content, e.g. vending or licensing of copyrighted material ; Digital rights management [DRM]
- G06F21/16—Program or content traceability, e.g. by watermarking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/30—Creation or generation of source code
- G06F8/31—Programming languages or programming paradigms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Technology Law (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a frame loss resistant GIF dynamic image copyright authentication method. The invention mainly comprises the following steps: on the basis, a watermark-containing image is generated through staged training optimization, and a frame is constructed to remove noise types in a truthful way so as to resist the attack of noise loss of high-strength frame-level information of the dynamic GIF image. The method provides a solution for the robustness watermarking of the GIF dynamic image, improves the image performance by using an anti-network and a staged training mode, generates a high-quality watermarked image, realizes the full-true-frame deletion noise simulation, and has practical application value in the common clipping and splicing operation and malicious image frame deletion attack scene of the dynamic image.
Description
Technical Field
The invention relates to the field of deep learning and information hiding, in particular to a frame loss resistant GIF dynamic image copyright authentication method.
Background
The GIF dynamic image is expressed by means of high fit, the dynamic interest is widely concerned on the social platform, the dynamic creation heat of the whole people is raised, the GIF dynamic image is widely applied to various large social media platforms along with the popularity of the GIF dynamic image, the GIF dynamic image is also gradually applied to the commercial fields of advertisements and the like, and the commercialization of the GIF dynamic image also provides higher requirements for the copyright protection of the GIF dynamic image. Digital Watermarking technology is an important research hotspot in the field of information security, wherein robust Watermarking is more effective means for copyright protection, however, the existing Watermarking research of GIF images mainly focuses on Fragile Watermarking, such as the document "munir.r.a frame watermark Scheme for Authentication of GIF images, international Journal on electric Engineering and information.2017" utilizes random bits based on a chaotic system to process before embedding, and uses EzStego to hide messages in palette images when embedding, but when applied to dynamic GIF images, only repeatedly operates on each frame image. For the aspect of robust watermarking, research on GIF dynamic images is lacked at present.
A generative confrontation network (GAN) was first applied to the field of face generation, and training of a Generator is promoted by mutual game of the confrontation network and the Generator. With the development of network hardware, the generated image is almost the same as the real image, and the generation type countermeasure network is successfully applied in the fields of image synthesis, scene migration, style migration and the like. In recent years, a generative confrontation network gradually leaves open the head in the field of information Hiding, a watermarking method based on a generative confrontation network technology appears, and a new idea is provided for information Hiding, namely documents' Zhu J, Kaplan R, Johnson J, and Fei-Fei L.HiDDeN: mapping data with depth networks. The game training of the countermeasure network of the generating type countermeasure network and the generator promotes the generator to train and generate high-quality images, and the images are matched with the target pursued by watermark research to a certain extent.
The GIF dynamic image is different from a common static image, and the dynamic image has time dimension information, so that the GIF dynamic image is required to resist common image processing types and faces noise attack of frame dimensions. The common clipping and stitching processing operations of GIF motion pictures add difficulty to their robustness. Loss of GIF watermark information may result when subjected to such image processing or deletion of illegal image frames. Therefore, the method should have robustness against frame-level noise attack while generating high-quality watermark-containing images.
Based on the above, the present patent aims to provide a method for generating a high-quality GIF dynamic image watermark, which can extract the watermark completely even if a large number of frame images are lost, thereby realizing omnibearing copyright authentication.
Disclosure of Invention
The invention provides a frame loss resistant GIF dynamic image copyright authentication method, which can resist loss of a large number of frame images while generating high-quality watermark of a GIF dynamic image. The method mainly comprises the following steps: an end-to-end copyright authentication method based on GAN is provided, and the generated watermark-containing image is further optimized in a staged training mode, and on the basis, the full-true construction frame deletes the noise type to resist the noise attack of high-strength frame level information loss.
The specific contents are as follows:
(1) a method for authenticating the rights of GIF dynamic images based on GAN end-to-end is provided: and (3) building a generating type countermeasure network based on a convolutional neural network by using a watermark technology, embedding the watermark image, generating the image containing the watermark and finishing watermark extraction.
The model consists of a preprocessor, a generator, a countermeasure network, a noise layer and a decoder, wherein the preprocessor realizes the preprocessing of the watermark image W and generates a preprocessed watermark image W', wherein W and h are the width and the height of the watermark image W respectively; the generator receives the preprocessed watermark images W' and the preprocessed GIF dynamic images C to generate watermark-containing images S, W, h and n which respectively represent the width, height and frame number of the dynamic images C; inputting an original dynamic image C and a generated watermark-containing image S by an anti-network, outputting information feedback to promote the training of a generator, and accelerating convergence until a watermark-containing image close to the original dynamic image is generated finally; the noise layer is arranged in front of the decoder and used for simulating common noise types, a watermark-containing image S is input, and the watermark-containing image after noise processing is marked as S'; the decoder inputs the watermark-containing image S 'which is processed by noise, and can still extract the watermark image W' more completely. Aiming at the characteristics of the dynamic image, the model is designed based on a three-dimensional convolution neural network, and the cubic form of the three-dimensional convolution kernel is more beneficial to extracting the space-time characteristics of the dynamic image.
In the training process, firstly, the network parameters of the generator are fixed, the watermark-containing dynamic image generated by the generator and the original dynamic image sample are input into the countermeasure network, and supervision training is carried out to improve parameters of the countermeasure network, so that the watermark-containing data generated by the generator can be better distinguished from real data. Then parameters of the confrontation network are fixed, a generator module is trained, and the generator module comprises a preprocessor, a generator and a decoder, so that the water-containing printed image generated by the generator is closer to a real data image, and finally the confrontation network cannot make correct judgment. With the increase of the training times, the countermeasure network and the generator reach balance, the watermark-containing image generated by the generator gradually approaches to a real image, at the moment, the accuracy is reduced to 0.5, and meanwhile, the countermeasure network cannot distinguish and judge whether the image is true or false, so that the visual quality of the watermark-containing image is improved. It is noted that since pattern collapse is likely to occur during the training of the countermeasure network, spectrum normalization (spectrum normalization) is added to the last layer of the countermeasure network to stabilize the training.
In training, the preprocessor, the generator, the noise layer, and the decoder are trained simultaneously, and the challenge network is trained alternately with the generator module. The countermeasure network is essentially similar to a classifier, and the output layer is full-connection layer output, aiming at judging whether the output image is a watermark image or an original dynamic image. Wherein a loss function L is usedadvTraining a countermeasure network in a watermark model, and defining Adv as the countermeasure network:
Ladv(W',C,S)=log(1-Adv((C,W')))+log(Adv(C))=log(1-Adv((S)))+log(Adv(C))
the entire network model is trained using a loss function L:
L(W,W',C,S)=α||W-W”||+β||C-S||+δLadv1,
where α, β, δ are weight assigned, Ladv1=log(1-Adv(S))。
And the noise layer is used for designing common image processing types to simulate noise attack and simulating common plane image processing types. Frame deletion and frame replacement are designed for frame level noise attacks that the dynamic image may suffer. Because the neural network needs to be continuously conductive, the noise type is converted into a tensor form in the network for simulation.
Defining: c ═ C1,c1,...,cmDenotes a set of original moving images in a time dimension, S ═ S1,s1,...,smDenotes a set of watermarked moving images in a time dimension, where m denotes the number of frames of the moving image, and the noise type is denoted as S' ═ F (S), F ∈ F, F ═ F (S)1,f2,f3,...,fzWhere z represents the number of noise types. Czero=0w*hIndicating zero vector fill and equal size to the motion picture frame.
And (4) frame deletion, namely randomly selecting partial frames of the image containing the watermark, replacing the partial frames by utilizing a zero vector with the same size as the plane, and simulating to delete the frame image in the image containing the watermark. Expressed as:
S'={s1,Czero,s3,Czero,...,sm}
and (4) frame replacement, namely randomly selecting any frame of the original watermark image C, replacing part of frames of the watermark image, and aiming at simulating noise interference caused by watermark information loss due to the fact that part of frames in the watermark image are replaced. Expressed as:
S'={s1,cm1,s3,cm2,...,sm},cmn∈C
(2) a GIF dynamic image watermarking scheme capable of resisting high-strength frame information loss noise is provided: through staged training, frame level information loss noise attack simulation is restored, and a scheme for generating high-quality GIF dynamic images with watermarks is provided.
In the end-to-end training of the GAN-based end-to-end GIF dynamic image copyright authentication method, an optimizer trains the whole frame at the same time, and the conditions of long training period, slow convergence due to mutual influence of several modules in the training and the like exist. Based on this, a staged GAN-based GIF moving image watermarking method is proposed. In the end-to-end watermarking method, the parameters of the generator and the decoder are trained simultaneously in an end-to-end mode, and the deviation between the watermark extracted by the decoder and the input watermark is transmitted back through the module. In the neural network training, since the watermark extraction effect needs to be ensured, the loss function is always converged towards the direction of reducing the watermark image quality to ensure the extraction accuracy of the decoder, which affects the visual effect of the generated watermark-containing image to a certain extent. In addition, under the attack of noise, the end-to-end watermarking method is slowly converged in the training process, and even the quality of the generated watermark-containing image is reduced. By training the generator and the decoder in stages, the influence caused by the training of the decoder is isolated, the current stage target of the network is concentrated, the image performance is improved, and meanwhile, the consumption of the training network can be reduced.
In the stage training, stage one, optimize the watermark-containing image generation under the noiseless condition, train the preprocessor, generator, fight against the network. And in the second stage, the preprocessor and the generator in the first stage are kept, the parameters of the antithetical network are unchanged, and a noise layer is loaded for training a decoder. In the second stage, because the training of the generator is completed, the noise attack is not transmitted back to the generator through the network, and the generation of the high-quality watermark-containing image is further ensured. The noise type design space is expanded while the network is isolated, the noise type in the actual life can be completely restored, and the continuous and micro requirement of the network training on the noise type is avoided.
In daily life, dynamic images are often spliced by clipping to compose a better visual effect. Meanwhile, the dynamic image with watermark may suffer from the attack of maliciously deleting the image frame in the transmission process. Under the above-mentioned scenes, the watermark information is lost, thereby increasing the difficulty of extracting the watermark-containing image. In the method based on the staged watermarking, the attack simulation of high-strength frame-level information loss noise is provided, and the generator is isolated by using the network, so that the frame deletion of the watermark-containing image can be constructed, and the direct deletion of the frame image is realized by randomly deleting the watermark-containing frame and simulating the attack of frame deletion noise. During training, watermark extraction capacity is improved through transfer learning, training is started from the condition of no frame loss, then training is started through low-strength frame deletion, and finally transfer learning is carried out through high-strength frame deletion.
Compared with the prior art, the technical scheme at least has the following remarkable effects:
1. the invention provides a GIF dynamic image copyright authentication method based on GAN end-to-end, which is characterized in that a functional module design is carried out based on a model of a convolutional neural network, an end-to-end watermark network structure is realized on three-dimensional convolution, and the copyright authentication of a GIF dynamic image product is realized by optimizing a watermark-containing dynamic image by using a countermeasure network.
2. A GIF dynamic image watermarking scheme capable of resisting high-strength frame information loss noise is provided, through staged training, the quality of images containing watermarks is improved, a noise layer is isolated, a large number of frame image losses are simulated in a truthful way, the simulation of high-strength frame deletion noise attack is realized, and the robustness of common clipping processing or malicious frame noise deletion in an actual scene is obtained.
Drawings
FIG. 1 is a schematic diagram of the training of a GIF dynamic image watermarking scheme capable of resisting high-strength frame information loss noise according to the present invention;
FIG. 2 is a network sub-module diagram of "a GIF dynamic image copyright authentication method based on GAN end-to-end" according to the present invention;
Detailed Description
The invention relates to a dynamic GIF image watermarking method based on a generating type countermeasure network. For convenience of explanation, the embodiment describes a specific implementation of the present invention by taking a GIF image segment with a time frame number of 8 as an example, but those skilled in the art should know that the technical solution of the present application does not limit the format of the GIF image. These embodiments are merely to explain the technical principles of the present invention and are not intended to limit the scope of the present invention.
The method can be implemented according to the following steps, is not limited to any programming language, and in the example, a python programming language is taken as an example, and the model is built on a deep learning platform of the pytorch, and the specific steps are as follows:
the method comprises the following steps: end-to-end watermark network model
Fig. 1 shows a training mode of the GIF dynamic image watermarking scheme of the present invention that can resist high-strength frame information loss noise, and fig. 2 shows a network structure of the GAN end-to-end-based GIF dynamic image copyright authentication method of the present invention, that is, a preprocessor, a generator, a countermeasure network, a noise layer, and a decoder. The input data item comprises a digital watermark image to be embedded and an original carrier image, namely a dynamic GIF image. The data item can be output as the embedded watermark-containing image and the extracted watermark identification image.
Step two: data set preparation.
N GIF moving images are arbitrarily selected from the image database A as carrier images, and the carrier images are cut into GIF images with the format of 256 × 8. Under the condition of considering copyright protection, a trademark image data set is selected as a watermark image database B, N images are randomly selected and processed into images which are cut into the size of 256 × 256. Training sets and test sets were randomly generated at arbitrary scale.
Step three: and (5) watermark model training.
Training the water model in the first step, wherein the specific experimental parameters are as follows: using an Adam optimizer, the initial learning rate was 0.0001 and the momentum parameter was (. beta.) (β)1=0.9,β20.999) due to the memory limit batch _ size is set to 1. The invention discloses a training end-to-end GAN-based GIF dynamic image copyright authentication method, which selects 5 noise interferences of common Gaussian blur, median filtering, salt-pepper noise, frame deletion and frame replacement, wherein the frame deletion and frame replacement randomly selects any 2 frames of 8 frames to delete or replace. Based on the network structure of the invention 2, the stage-by-stage GIF dynamic image watermark scheme training which can resist the loss noise of high-strength frame information is carried out, and the stage 1 finishes the training of a generator, a preprocessor and a decoder under the condition of no noise interference through the stage-by-stage training. On the basis of the stage one, a noise layer is added to realize direct frame deletion noise simulation, a decoder is trained, namely, the network parameters trained in the previous stage are loaded, and an Adam optimizer only trains the decoder. Firstly training a non-set frame deletion type, loading model parameters, directly deleting 2 frames of images, finally transferring to directly deleting 4 frames of images for training, and finishing the noise simulation of the directly deleted (4/8) frames of images.
Step four: and (6) testing.
And testing on the watermark model which is trained in the third step and can resist various noise interferences. The testing process is specifically that only a single image processing type f is adopted in the noise layer by replacing the image processing type of the noise layer1,f2,f3,...,fzThe above tests were carried out separately.
In summary, aiming at the GIF dynamic image, the invention provides a GAN-based copyright authentication method, training is carried out in stages while frame deletion noise is simulated, a robust watermark of the GIF dynamic image is generated, the current functional requirements on the GIF dynamic image are met, and the method has practical application value in common clipping and splicing operation and malicious image frame deletion attack scenes of the dynamic image.
It will be appreciated by persons skilled in the art that the scope of the present invention is not limited to the specific embodiments described. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and it is noted that the technical solutions after the changes or substitutions will fall within the protection scope of the invention.
Claims (3)
1. A frame loss resistant GIF dynamic image copyright authentication method is characterized by comprising the following steps:
the invention provides a watermarking method of a GIF dynamic image capable of resisting frame loss, which can still finish copyright owner authentication when the GIF dynamic image loses a large amount of frame image information, and mainly comprises the following steps:
on the basis, a watermark-containing image is generated through staged training optimization, and a frame is constructed to remove noise types in a truthful way so as to resist the attack of noise loss of high-strength frame-level information of the dynamic GIF image.
2. The GAN-based frame loss resistant GIF dynamic image watermarking method as claimed in claim 1, wherein an end-to-end generation-based countermeasure network-based GIF dynamic image copyright authentication method is proposed, which specifically comprises:
watermark embedding and extraction of the GIF dynamic image are completed through a preprocessor, a generator, a countermeasure network, a decoder, the countermeasure network and a noise layer module. The confrontation network and the generator carry out confrontation training, the generator training is promoted, the generated watermark-containing image is optimized, meanwhile, in the design of the confrontation network, spectrum normalization is added in the last layer of network, and the confrontation network is stabilized in the training process.
3. The GIF dynamic image watermarking scheme capable of resisting high-strength frame information loss noise according to claim 2, wherein a staged training mode and a full true frame deletion noise simulation scheme are proposed for the watermarking method, and specifically include:
the method is characterized in that a generator is trained in a first stage in a staged mode, parameters are fixed, and then a decoder is trained in a second stage. The encoder and the decoder are trained in stages, the influence caused by the training of the decoder is isolated, the current stage target of the network is concentrated, and the high-strength frame deletion is simulated really.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011481416.1A CN112487369A (en) | 2020-12-16 | 2020-12-16 | Frame loss resistant GIF dynamic image copyright authentication method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011481416.1A CN112487369A (en) | 2020-12-16 | 2020-12-16 | Frame loss resistant GIF dynamic image copyright authentication method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112487369A true CN112487369A (en) | 2021-03-12 |
Family
ID=74917208
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011481416.1A Pending CN112487369A (en) | 2020-12-16 | 2020-12-16 | Frame loss resistant GIF dynamic image copyright authentication method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112487369A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113689318A (en) * | 2021-07-30 | 2021-11-23 | 南京信息工程大学 | Deep semi-fragile watermarking method for image authentication and defense against samples |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101668170A (en) * | 2009-09-23 | 2010-03-10 | 中山大学 | Digital television program copyright protecting method for resisting time synchronization attacks |
CN109345441A (en) * | 2018-10-19 | 2019-02-15 | 上海唯识律简信息科技有限公司 | A kind of de-watermarked method and system of image based on generation confrontation network |
CN110276708A (en) * | 2019-05-08 | 2019-09-24 | 济南浪潮高新科技投资发展有限公司 | A kind of image digital watermark generation and identification system and method based on GAN network |
CN110727928A (en) * | 2019-10-12 | 2020-01-24 | 湘潭大学 | 3D video copyright comprehensive protection method based on deep reinforcement learning optimization |
US20200226407A1 (en) * | 2019-01-16 | 2020-07-16 | Rok Mobile International Ltd. | Delivery of digital content customized using images of objects |
CN111598761A (en) * | 2020-04-17 | 2020-08-28 | 中山大学 | Anti-printing shot image digital watermarking method based on image noise reduction |
CN111681155A (en) * | 2020-06-09 | 2020-09-18 | 湖南大学 | GIF dynamic image watermarking method based on deep learning |
CN111696046A (en) * | 2019-03-13 | 2020-09-22 | 北京奇虎科技有限公司 | Watermark removing method and device based on generating type countermeasure network |
-
2020
- 2020-12-16 CN CN202011481416.1A patent/CN112487369A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101668170A (en) * | 2009-09-23 | 2010-03-10 | 中山大学 | Digital television program copyright protecting method for resisting time synchronization attacks |
CN109345441A (en) * | 2018-10-19 | 2019-02-15 | 上海唯识律简信息科技有限公司 | A kind of de-watermarked method and system of image based on generation confrontation network |
US20200226407A1 (en) * | 2019-01-16 | 2020-07-16 | Rok Mobile International Ltd. | Delivery of digital content customized using images of objects |
CN111696046A (en) * | 2019-03-13 | 2020-09-22 | 北京奇虎科技有限公司 | Watermark removing method and device based on generating type countermeasure network |
CN110276708A (en) * | 2019-05-08 | 2019-09-24 | 济南浪潮高新科技投资发展有限公司 | A kind of image digital watermark generation and identification system and method based on GAN network |
CN110727928A (en) * | 2019-10-12 | 2020-01-24 | 湘潭大学 | 3D video copyright comprehensive protection method based on deep reinforcement learning optimization |
CN111598761A (en) * | 2020-04-17 | 2020-08-28 | 中山大学 | Anti-printing shot image digital watermarking method based on image noise reduction |
CN111681155A (en) * | 2020-06-09 | 2020-09-18 | 湖南大学 | GIF dynamic image watermarking method based on deep learning |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113689318A (en) * | 2021-07-30 | 2021-11-23 | 南京信息工程大学 | Deep semi-fragile watermarking method for image authentication and defense against samples |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lydia et al. | Application of discrete transforms with selective coefficients for blind image watermarking | |
CN111598761B (en) | Anti-printing shooting image digital watermarking method based on image noise reduction | |
CN113222800B (en) | Robust image watermark embedding and extracting method and system based on deep learning | |
CN108230278B (en) | Image raindrop removing method based on generation countermeasure network | |
CN111681155B (en) | GIF dynamic image watermarking method based on deep learning | |
Wei et al. | Generative steganography network | |
Wang et al. | HidingGAN: High capacity information hiding with generative adversarial network | |
Li et al. | A generative steganography method based on WGAN-GP | |
CN116645260B (en) | Digital watermark attack method based on conditional diffusion model | |
Fang et al. | Encoded feature enhancement in watermarking network for distortion in real scenes | |
CN115640609A (en) | Feature privacy protection method and device | |
CN112487369A (en) | Frame loss resistant GIF dynamic image copyright authentication method | |
Chen et al. | Patch selection denoiser: An effective approach defending against one-pixel attacks | |
KR20200094938A (en) | Data imbalance solution method using Generative adversarial network | |
Panagiotou et al. | Procedural 3D terrain generation using generative adversarial networks | |
You et al. | Image generation network for covert transmission in online social network | |
Huang et al. | Steganography embedding cost learning with generative multi-adversarial network | |
Lin et al. | A high-capacity distortion-free information hiding algorithm for 3D polygon models | |
CN113554545A (en) | Model watermarking method for image processing model copyright protection | |
CN114119330B (en) | Robust digital watermark embedding and extracting method based on neural network | |
CN117156152A (en) | Model training method, encoding method, decoding method and equipment | |
CN115496843A (en) | Local realistic-writing cartoon style migration system and method based on GAN | |
Divya et al. | Steganographic data hiding using modified APSO | |
CN111353394A (en) | Video behavior identification method based on three-dimensional alternate update network | |
Pillai | Fake colorized and morphed image detection using convolutional 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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210312 |