CN111696046A - Watermark removing method and device based on generating type countermeasure network - Google Patents
Watermark removing method and device based on generating type countermeasure network Download PDFInfo
- Publication number
- CN111696046A CN111696046A CN201910190722.0A CN201910190722A CN111696046A CN 111696046 A CN111696046 A CN 111696046A CN 201910190722 A CN201910190722 A CN 201910190722A CN 111696046 A CN111696046 A CN 111696046A
- Authority
- CN
- China
- Prior art keywords
- watermark
- image
- generator
- loss function
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000012549 training Methods 0.000 claims abstract description 90
- 238000009826 distribution Methods 0.000 claims abstract description 11
- 230000006870 function Effects 0.000 claims description 96
- 238000004590 computer program Methods 0.000 claims description 13
- 238000003860 storage Methods 0.000 claims description 13
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 abstract description 14
- 230000000694 effects Effects 0.000 abstract description 13
- 238000010586 diagram Methods 0.000 description 9
- 238000010606 normalization Methods 0.000 description 7
- 238000005457 optimization Methods 0.000 description 7
- 238000007781 pre-processing Methods 0.000 description 7
- 239000011800 void material Substances 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000011423 initialization method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/77—Retouching; Inpainting; Scratch removal
-
- 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
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The invention provides a watermark removing method and device based on a generative countermeasure network. The generative confrontation network comprises a generator, and the method comprises the following steps: acquiring a training sample image; inputting the training sample image into a generator to generate a corresponding watermark-free image; calculating a generator loss function, and training a convolution kernel parameter to be learned according to the generator loss function; repeating the steps for multiple times, and when the loss function of the generator is converged, storing the convolution kernel parameters obtained by training and the generator as a trained watermark removal network model; and inputting the watermark-containing image into a watermark removal network model to obtain a corresponding watermark-removed image. The scheme provided by the invention can better simulate the real distribution of the image, can not damage the original pixels of the image while effectively removing the watermark, avoids the loss of image information in the watermark removing process, improves the watermark removing effect, and has better generalization capability.
Description
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to a watermark removal method and apparatus based on a generative countermeasure network.
Background
Watermarking is a widely used way of protecting copyright information of multimedia data such as images, videos, etc. However, some watermarks of a malicious marketing nature can affect the enjoyment of the image. Therefore, how to remove these marketing watermarks becomes a relatively important technical requirement. The traditional watermark removing method comprises sharpening, wiener filtering, wavelet transformation and the like, but the methods have larger use limitations, the appearance and the mode of the watermark need to be known when the watermark is removed, the watermark is not completely removed and the original image information is lost, and along with the development of computers and the continuous research of deep learning in recent years, a more effective deep learning method is used for solving the traditional research problem, so that the method is more effective and feasible. The generative countermeasure network is a deep learning model and is widely applied to a plurality of fields such as data generation and image restoration.
Disclosure of Invention
The present invention proposes a watermark removal method and apparatus based on a generative countermeasure network to overcome the above problems or at least partially solve the above problems.
According to an aspect of the present invention, there is provided a watermark removal method based on a generative countermeasure network, the generative countermeasure network including a generator, the method including:
acquiring a training sample image;
inputting the training sample image into the generator to generate a corresponding waterless image;
calculating a generator loss function, and training a convolution kernel parameter to be learned according to the generator loss function;
repeating the steps for multiple times, and when the loss function of the generator is converged, storing the convolution kernel parameters obtained by training and the generator as a trained watermark removal network model;
and inputting the image containing the watermark into the watermark removing network model to obtain a corresponding image with the watermark removed.
Optionally, before the obtaining of the training sample, the method includes:
and carrying out parameter initialization on the convolution kernel to be learned, and giving an initial value to the convolution kernel to be learned.
Optionally, the performing parameter initialization on the convolution kernel to be learned includes:
and initializing the convolution kernel parameters to be learned by using a Gaussian distribution with a mean value of 0 and a variance of 0.01.
Optionally, the inputting the training sample image into the generator to generate a corresponding watermark-free image includes:
and inputting the water-containing print sample image into the generator, and sequentially performing convolution and deconvolution operations on the water-containing print sample image and the initial value of the convolution kernel parameter to output a corresponding water-free print image.
Optionally, the generative countermeasure network further comprises a discriminator, and the method further comprises:
the image after the watermark is removed and outputted by the watermark removing network model and the original image without the watermark are inputted into a discriminator together;
judging whether the image after the watermark removal output by the watermark removal network model meets the requirements or not;
and feeding back the judgment result to a watermark removal network model, and optimizing the watermark removal network model based on the judgment result.
Optionally, the discriminator comprises a global discriminator and/or a local discriminator, and the method further comprises:
calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator;
and adjusting the convolution kernel parameters to be learned according to the loss function.
Optionally, the determining whether the image output by the watermark removal network model after removing the watermark meets the requirement, feeding back a determination result to the generator, and optimizing the watermark removal network model based on the determination result includes:
respectively extracting the image without the watermark and the corresponding overall image characteristics of the image without the watermark, and judging whether the overall image of the image without the watermark is a real image or not according to the extracted overall image characteristics; and/or
Respectively extracting local characteristics of a corresponding watermark removal area image part in the image without the watermark and a corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
and feeding back the judgment result to the global discriminator and/or the local discriminator, and optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
Optionally, before the acquiring the training sample image, the method further includes:
collecting a large number of images without watermarks and watermark images;
taking the image without the watermark as an original image, and adding the watermark image to the original image with random transparency to synthesize a watermark-containing image;
and generating a training sample image set by using the watermark-containing image as a training sample.
Optionally, before the inputting the training sample image into the generator to generate the corresponding watermark-free image, the method further includes:
and carrying out scale normalization preprocessing on the input training sample image.
Optionally, the generator is a full convolution network with an encode-decode structure, and includes an encoder and a decoder;
the encoder comprises a first convolution module, a second convolution module and a third convolution module, and the decoder comprises a first deconvolution module, a second deconvolution module and a third deconvolution module;
each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module.
Based on the same inventive concept as the above method, there is also provided a watermark removing apparatus based on a generative countermeasure network, the generative countermeasure network including a generator, the apparatus including:
the acquisition module is used for acquiring a training sample image;
the generating module is used for inputting the training sample image into the generator to generate a corresponding watermark-free image;
the first calculation module is used for calculating the generator loss function and training the convolution kernel parameters to be learned according to the generator loss function;
the storage module is used for repeating the steps for multiple times, and when the loss function of the generator is converged, the convolution kernel parameters obtained by training and the generator are stored as a trained watermark removal network model;
and the watermark removing module is used for inputting the watermark-containing image into the watermark removing network model to obtain a corresponding image with the watermark removed.
Optionally, the apparatus further comprises:
and the initialization module is used for carrying out parameter initialization on the convolution kernel to be learned before the training sample is obtained and endowing an initial value to the convolution kernel to be learned.
Optionally, the initialization module is further configured to initialize the convolution kernel parameter to be learned by using a gaussian distribution with a mean value of 0 and a variance of 0.01.
Optionally, the generating module is further configured to input the image containing the watermark sample into the generator, and sequentially perform convolution and deconvolution operations on the image containing the watermark sample and the initial value of the convolution kernel parameter to output a corresponding watermark-free image.
Optionally, the generative countermeasure network further includes a discriminator, and the apparatus further includes:
the input module is used for inputting the image after the watermark is removed and output by the watermark removal network model and the original image without the watermark into the discriminator together;
the judging module is used for judging whether the image after the watermark is removed and output by the watermark removing network model meets the requirements or not;
and the optimizing module is used for feeding back the judgment result to the watermark removing network model and optimizing the watermark removing network model based on the judgment result.
Optionally, the discriminator includes a global discriminator and/or a local discriminator, and the apparatus further includes:
and the second calculation module is used for calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator, and the convolution kernel parameters to be learned are adjusted according to the loss function.
Optionally, the determining module is further configured to respectively extract the image without the watermark and the corresponding overall image feature of the image without the watermark, and determine whether the overall image of the image without the watermark is a real image according to the extracted overall image feature; and/or
Respectively extracting local characteristics of a corresponding watermark removal area image part in the image without the watermark and a corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
and the optimization module is further used for feeding back the judgment result to the global discriminator and/or the local discriminator, optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
Optionally, the apparatus further comprises:
the collecting module is used for collecting a large number of images without watermarks and watermark images;
and the watermarking module is used for taking the image without the watermark as an original image, adding the watermark image to the original image with random transparency to synthesize a watermarked image, and taking the watermarked image as a training sample to generate a training sample image set.
Optionally, the generating module is further configured to perform scale normalization preprocessing on the input training sample image.
Optionally, the generator is a full convolution network with an encode-decode structure, and includes an encoder and a decoder;
the encoder comprises a first convolution module, a second convolution module and a third convolution module, and the decoder comprises a first deconvolution module, a second deconvolution module and a third deconvolution module;
each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module.
According to another aspect of the present invention, there is also provided a computer storage medium storing computer program code which, when run on a computing device, causes the computing device to perform any one of the above-mentioned method for generating network-based watermark removal.
According to another aspect of the present invention, there is also provided a computing device comprising:
a processor;
a memory storing computer program code;
the computer program code, when executed by the processor, causes the computing device to perform any of the above-described generative network-based watermark removal methods.
The embodiment of the invention provides a watermark removing scheme based on a generating type network, wherein the generating type confrontation network comprises a generator, in the embodiment, a training sample image is obtained, the obtained training sample image is input to the generator to generate a corresponding watermark-free image, then a generator loss function is calculated, a convolution kernel parameter to be learned is trained according to the calculated generator loss function, the steps are repeated, and when the generator loss function is converged, the trained convolution kernel parameter and the generator are stored as a trained watermark removing network model. Based on the scheme provided by the embodiment of the invention, the real distribution of the image can be better simulated, the original pixels of the image cannot be damaged while the watermark is effectively removed, the loss of image information in the watermark removing process is avoided, the watermark removing effect is improved, and the generalization capability is better.
The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.
The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments thereof, taken in conjunction with the accompanying drawings.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
fig. 1 is a flowchart of a watermark removal method based on a generative countermeasure network according to an embodiment of the present invention;
fig. 2 is a flowchart of a watermark removal method based on a generative countermeasure network according to another embodiment of the present invention;
FIG. 3 is a schematic diagram of a training process according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a generator according to an embodiment of the invention;
fig. 5 is a block diagram of a watermark removal apparatus based on a generative countermeasure network according to an embodiment of the present invention;
fig. 6 is a block diagram of a watermark removal apparatus based on a generative countermeasure network according to another embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
It should be noted that the features of the embodiments and preferred embodiments of the present invention can be combined with each other without conflict.
Fig. 1 is a flowchart of a watermark removal method based on a generative countermeasure network according to an embodiment of the present invention, wherein the generative countermeasure network includes a generator. As shown in fig. 1, the watermark removing method based on the generative countermeasure network according to the embodiment of the present invention at least includes the following steps S102 to S110:
step S102, acquiring a training sample image;
step S104, inputting the obtained training sample image into a generator to generate a corresponding watermark-free image;
step S106, calculating a generator loss function, and training a convolution kernel parameter to be learned according to the generator loss function;
step S108, repeating the steps for many times, and when the loss function of the generator is converged, storing the convolution kernel parameters obtained by training and the generator as a trained watermark removal network model;
step S110, inputting the image containing the watermark into a watermark removal network model to obtain a corresponding image after the watermark is removed.
The embodiment of the invention provides a watermark removing scheme based on a generating type network, wherein the generating type confrontation network comprises a generator, in the embodiment, a training sample image is obtained, the obtained training sample image is input to the generator to generate a corresponding watermark-free image, then a generator loss function is calculated, a convolution kernel parameter to be learned is trained according to the calculated generator loss function, the steps are repeated, and when the generator loss function is converged, the trained convolution kernel parameter and the generator are stored as a trained watermark removing network model. Based on the scheme provided by the embodiment of the invention, the real distribution of the image can be better simulated, the original pixels of the image cannot be damaged while the watermark is effectively removed, the loss of image information in the watermark removing process is avoided, the watermark removing effect is improved, and the generalization capability is better.
The loss function (loss function) is also called a cost function (cost function) and is used for evaluating the degree of inconsistency of a predicted value and a true value of the model, the loss function in deep learning is a 'baton' of the whole network model, and the learning of network parameters is guided through back propagation of errors generated by marking predicted samples and the true samples. The loss function is also an optimized objective function in the neural network actually, the process of training or optimizing the neural network is a process of minimizing the loss function, the smaller the loss function is, the closer the predicted value of the model is to the true value, and the better the robustness of the model is. Common loss functions include Hinge loss functions, cross entropy loss functions, logarithmic loss functions, squared loss functions, 0-1 loss functions, and the like.
The Loss function mentioned in step S106 may be a cross-entropy Loss function, and in the embodiment of the present invention, the Loss function of a single sample may be expressed as Loss- [ log y ^ + (1-y) log (1-y ^) ], and the total Loss function of N samples is a superposition of N losses, where y is a real sample label [0,1], which represents a negative class and a positive class, respectively. When y is 1, the prediction output is closer to the true sample label 1, the Loss function Loss is smaller, the prediction output is closer to 0, the Loss is larger, when y is 0, the prediction output is closer to the true sample label 0, the Loss function Loss is smaller, the prediction output is closer to 1, and the Loss is larger. Loss characterizes the difference between the prediction output and y, and the more the prediction output is different from y, the larger the value of Loss, i.e., the greater the "penalty" for the current model, and the more the similar exponential growth level. This is because the log function itself is characterized, which may bias the model towards making the prediction output closer to the true sample label. In practical applications, an appropriate loss function may be selected according to the situation, and the present invention is not particularly limited thereto.
At the beginning of training, because the convolution kernel parameters are initialized randomly, in this case, the watermark-free image generated by the generator has a large difference from the output expected by the user, and the Loss is also large. For step S106, it is preferable to calculate the parameter value to be updated by back propagation of the gradient, so that the Loss becomes small.
And step S108 is executed, after a plurality of times of iterative training, the Loss is gradually reduced, and the image output by the network is closer to the image which is expected to be subjected to watermark removal. When Loss converges, i.e. the Loss function approaches a stable value, no longer becomes significantly smaller, and the network learns better parameters. The image output by the network model is very close to the image which is expected by the user after the watermark is removed, namely the output expected by the user is satisfied.
In an embodiment, before step S102 is executed, parameter initialization may also be performed on the network model. Optionally, parameter initialization is performed on the convolution kernel to be learned, and an initial value of the parameter of the convolution kernel to be learned is given. The initialization of parameters is important for the training of the network, poor initialization parameters can cause the problem of gradient propagation and reduce the training speed, and good initialization parameters can accelerate convergence and find a better solution. The good initialization method can not only increase the learning speed, but also improve the accuracy. In a preferred embodiment of the present invention, the parameters of the convolution kernel to be learned are initialized using a gaussian distribution with a mean of 0 and a variance of 0.01. Based on the scheme of the preferred embodiment, the training process can be shortened, and the convergence can be accelerated on the premise of ensuring the accuracy.
Aiming at the step S104, the training sample image is input into the generator to generate a corresponding watermark-free image, and the invention provides a preferable scheme. In the scheme, firstly, a watermark-containing sample image is input into a generator, and then is sequentially subjected to convolution and deconvolution operations with convolution kernel parameters, so that a corresponding watermark-free image is output. The convolution process (taking two-dimensional convolution as an example) is that a moving window with the same size as the template is opened from the upper left corner of the image matrix, the window image and the template pixel are multiplied and added, the pixel brightness value at the center of the window is replaced by the calculation result, then, the moving window moves to the right for a preset sliding step length, the same operation is performed, and so on, a new image matrix can be obtained from left to right and from top to bottom, and the deconvolution operation is the inverse operation of the convolution operation and is not repeated herein.
Based on the watermark removal network model obtained by the technical scheme, when the image containing the watermark is input, the image after the watermark is removed can be obtained, but under the normal condition, the network model can be optimized in order to obtain a better watermark removal effect and better meet the requirements of users.
In an embodiment of the present invention, the generative confrontation network further comprises a discriminator, and the discriminator can be used to evaluate the effect of the network model. As shown in fig. 2, the watermark removing method based on the generative countermeasure network may further include:
step S202, the image after removing the watermark and the original image without the watermark, which are output by the watermark removing network model, are input into a discriminator together;
step S204, judging whether the image after the watermark removal output by the watermark removal network model meets the requirements or not;
and step S206, feeding back the judgment result to the watermark removal network model, and optimizing the watermark removal network model based on the judgment result.
In this embodiment, the watermark removal network model is optimized according to the comparison determination result by comparing and determining the image after removing the watermark and the original image without the watermark. Based on the scheme of the embodiment, the effect of image watermark removal can be evaluated, and meanwhile, the network model can be optimized, so that the watermark removal effect is further improved.
In order to acquire more image information as much as possible, both global and local information need to be considered. In a preferred embodiment of the present invention, the discriminator includes a global discriminator and/or a local discriminator, and the watermark removing method based on the generative countermeasure network may further include: and calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator, and adjusting the convolution kernel parameters to be learned according to the loss function. In this embodiment, the loss function comprises a loss function portion of the global discriminator and/or a loss function portion of the local discriminator, and further the convolution kernel parameters of the network model may be adjusted according to the loss function. Better model parameters can be learned based on the scheme provided by the preferred embodiment, and the watermark removing effect is greatly improved.
The global discriminator uses the whole image as input, and further judges whether the input image is the image after removing the watermark or the original image without the watermark, in other words, the global information is used for evaluating the effect of removing the watermark. The local discriminator observes a partial area with the watermark as the center, and takes a local image as input, thereby further improving the local consistency of the watermark removal area by utilizing local information. It should be noted that, in practical applications, one or more global discriminators and one or more local discriminators may be provided, and the number of discriminators is not specifically limited in the embodiment of the present invention.
Preferably, the features can be extracted by a convolutional neural network, and the features are represented by vectors.
In view of the above step S204, the present invention provides an alternative embodiment. In this embodiment, the image without the watermark and the corresponding overall image feature without the watermark may be extracted, and whether the overall image of the image without the watermark is a real image or not may be determined according to the extracted overall image feature; and/or respectively extracting the local characteristics of the watermark removal area image part corresponding to the image without the watermark and the watermark area image part corresponding to the image without the watermark, and judging whether the local image of the image without the watermark is a real image or not according to the extracted local characteristics. In this embodiment, the step S206 may include: and feeding back the judgment result to the global discriminator and/or the local discriminator, and optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
According to the scheme provided by the embodiment of the invention, the overall and/or local characteristics of the image without the watermark and the original image without the watermark are extracted for comparison and judgment, then the judgment result is fed back to the corresponding overall discriminator and/or local discriminator, when the overall image and/or the local image is a real image, the watermark removal network model is optimized, and the optimized network model is stored for direct use by a user. According to the scheme provided by the invention, the effect of removing the watermark can be evaluated by utilizing the global information of the image, and the local consistency of the watermark removing area is further improved by considering the local information.
In addition, in the solution of the embodiment of the present invention, the training sample images are some watermark-containing images, and these watermark-containing images are also input to the network model. In an embodiment of the present invention, the watermark removing method based on the generative countermeasure network may further include: collecting a large number of images without watermarks and watermark images; taking the image without the watermark as an original image, and adding the watermark image to the original image with random transparency to synthesize the image with the watermark; and generating a training sample image set by using the watermark-containing image as a training sample. Based on the scheme provided by the embodiment of the invention, various training samples, namely watermark-containing images, can be obtained, and abundant sample data enables a network model to learn more effective and accurate parameters, thereby improving the accuracy of network model training.
The method comprises the steps of adding a watermark to an original image to generate a composite watermark image (including a watermark image), selecting a plurality of characters, symbols, numbers and a plurality of different fonts to obtain the watermark, preferably selecting the transparency value between 0 and 9, and obtaining abundant training data by the scheme so as to improve the accuracy of model training.
In a preferred embodiment of the present invention, in the training process of the network model (see fig. 3), a watermark image is first added to an original image to generate a watermarked image, the watermarked image is input to a generator, and the watermarked image is sequentially passed through an encoder and a decoder in the generator to generate an image after removing the watermark. Then, performing optimization training on the network model by using a global discriminator and a local discriminator, specifically, taking the original image and the image without the watermark as the input of the global discriminator, and judging whether the image without the watermark is an original real image; and the corresponding watermark region image part in the original image and the corresponding watermark removal region image in the image after the watermark removal are used as the input of the local discriminator, so that the local consistency of the watermark removal region is improved. And performing optimization training on the network model through the global discriminator and the local discriminator to finally achieve the effect that the two discriminators cannot distinguish whether the final image is the original image without the watermark or the image after the watermark is removed. The scheme of the preferred embodiment can give consideration to both integral and local information, and the training effect is better.
In the training process, the parameters of the network model include some artificially set hyper-parameters, such as an optimization method, the size of the input image of the network model to be trained, Batch _ size, and the like, in addition to the parameters to be learned in the training process. For example, the learning rate of the optimization method may be set to 0.001, the image size may be set to 256 × 256, and the Batch _ size may be set to 32.
In one embodiment, when the size of the input image of the network model does not meet the preset image size, the input training sample image is subjected to scale normalization preprocessing. In this embodiment, the preprocessing further includes performing noise reduction on the image to remove interference colors in the image, and clipping a blank area around the image and then performing scale normalization.
Fig. 4 is a schematic structural diagram of a generator according to an embodiment of the present invention, and as shown in fig. 4, the generator is a full convolution network with an encode-decode structure, and includes an encoder and a decoder, the encoder includes a first convolution module 10, a second convolution module 11, and a third convolution module 12, and the decoder includes a first deconvolution module 20, a second deconvolution module 21, and a third deconvolution module 22; each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module. The hole convolution layer is added, so that the receptive field of the characteristic is increased, and the watermark can be removed by using more information.
The image containing the watermark is sequentially subjected to 3 groups of convolution modules (3 times of down sampling) of the encoder and 3 groups of deconvolution modules (3 times of up sampling) of the decoder, and the image after the watermark is removed is obtained. The following describes the watermark removal process by the watermark removal network model in detail with the change of the image (feature map) size. An original image is composed of 3 color channels of red, green and blue, the length is H, the width is W, the size of the image is W H3, firstly, the watermark is added on the original image to synthesize a watermark-containing image, at this time, the size of the watermark-containing image is W H3, then the watermark-containing image is input into a watermark removing network model, after passing through a first convolution module in an encoder, the size of the image (characteristic diagram) is changed into (W/2) C1, after passing through a second convolution module in the encoder, the size of the image (characteristic diagram) is changed into (W/4) C2, after passing through a third convolution module in the encoder, the size of the image (characteristic diagram) is changed into (W/8) (H/8) C3, after passing through a first reverse rolling machine module of a decoder, the size of the image (characteristic diagram) is changed into (W/4) C2, after passing through the decoder second rewinder module, the image (feature map) size becomes (W/2) × (H/2) × C1, and after passing through the decoder third rewinder module, the image (feature map) size becomes W × H3. The output of the third deconvolution module in the encoder is here the image after removal of the watermark, i.e. the image from which the watermark mask model was removed.
C1, C2, and C3 above indicate the number of channels corresponding to the feature map, and 2, 4, and 8 refer to the reduction in the length and width of the feature map relative to the input image by a factor. The feature map size becomes 1/2 for the input image after the first convolution module, 1/4 for the input image after the second convolution module, and 1/8 for the input image after the third convolution module.
In an embodiment of the present invention, the convolution kernel size of the first convolution layer in the first convolution module is 3 × 3, the number of output channels is 32, the sliding step length is 1, the filling manner is SAME, and the second convolution layer is consistent with the first convolution layer of the SAME convolution module except for the sliding step length being 2.
The convolution kernel size of the first convolution layer in the second convolution module is 3 multiplied by 3, the output channel number is 64, the sliding step length is 1, the filling mode is SAME, and the second convolution layer is consistent with the first convolution layer of the SAME convolution module except the sliding step length is 2.
The convolution kernel size of the first convolution layer in the third convolution module is 3 x 3, the number of output channels is 128, the sliding step length is 1, the filling mode is SAME, and the second convolution layer is consistent with the first convolution layer of the SAME convolution module except for the sliding step length of 2.
The deconvolution module is an inverse process of the convolution module, wherein the sizes of convolution kernels in the convolution layer are all selected to be 3 × 3, and the number of output channels is reduced in turn.
Corresponding to the above watermark removing method based on the generative countermeasure network, an embodiment of the present invention further provides a watermark removing apparatus based on the generative countermeasure network, where the generative countermeasure network includes a generator, as shown in fig. 5, the watermark removing apparatus based on the generative countermeasure network at least includes:
an obtaining module 502, configured to obtain a training sample image;
a generating module 504, configured to input the training sample image into the generator to generate a corresponding watermark-free image;
the first calculating module 506 is configured to calculate a generator loss function, and train a convolution kernel parameter to be learned according to the generator loss function;
a storage module 508, configured to repeat the above steps for multiple times, and when the generator loss function converges, store the trained convolution kernel parameters and the generator as a trained watermark removal network model;
the watermark removing module 510 is configured to input the watermark-containing image into a watermark removing network model, and obtain a corresponding image from which the watermark is removed.
In a preferred embodiment, as shown in fig. 6, the watermark removing apparatus based on the generative countermeasure network may further include:
the initialization module 512 is configured to perform parameter initialization on a convolution kernel to be learned before obtaining a training sample, and assign an initial value to the convolution kernel to be learned.
In a preferred embodiment, the initialization module 512 is further configured to initialize the convolution kernel parameters to be learned by using a gaussian distribution with a mean value of 0 and a variance of 0.01.
In a preferred embodiment, the generating module 504 is further configured to input the watermark-containing sample image into the generator, sequentially perform convolution and deconvolution operations on the watermark-containing sample image and the initial values of the convolution kernel parameters, and output a corresponding watermark-free image.
In a preferred embodiment, the generative countermeasure network further includes an identifier, and referring to fig. 6, the watermark removing apparatus based on the generative countermeasure network may further include:
an input module 514, configured to input the image without the watermark output by the watermark removal network model and the original image without the watermark to the discriminator together;
a judging module 516, configured to judge whether the image with the watermark removed, output by the watermark removal network model, meets the requirement;
and an optimizing module 518, configured to feed back the determination result to the watermark removal network model, and optimize the watermark removal network model based on the determination result.
In a preferred embodiment, the discriminator includes a global discriminator and/or a local discriminator, and the watermark removing apparatus based on the generative countermeasure network may further include:
and a second calculating module 520, configured to calculate a loss function of the watermark removal network model, where the loss function includes a loss function of the global discriminator and/or a loss function of the local discriminator, and adjust the convolution kernel parameter to be learned according to the loss function.
In a preferred embodiment, the determining module 516 is further configured to extract the features of the image without the watermark and the corresponding overall image without the watermark, and determine whether the overall image of the image without the watermark is a real image according to the extracted features of the overall image; and/or respectively extracting the local characteristics of the corresponding watermark removal area image part in the image without the watermark and the corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
the optimization module 518 is further configured to feed back the determination result to the global discriminator and/or the local discriminator, and optimize and store the watermark removal network model when the whole image and/or the local image is a real image.
In a preferred embodiment, referring to fig. 6, the watermark removing apparatus based on the generative countermeasure network may further include:
a collection module 522 for collecting a plurality of watermark-free images and watermark images;
and a watermarking module 524, configured to take the image without the watermark as an original image, add the watermark image to the original image with random transparency to synthesize a watermarked image, and use the watermarked image as a training sample to generate a training sample image set.
In a preferred embodiment, the generating module 504 is further configured to perform a pre-processing of scale normalization on the input training sample image.
In a preferred embodiment, referring to fig. 4, the generator is a full convolutional network of an encode-decode structure, including an encoder and a decoder;
the encoder comprises a first convolution module 10, a second convolution module 12 and a third convolution module 14, and the decoder 2 comprises a first deconvolution module 20, a second deconvolution module 22 and a third deconvolution module 24;
each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module.
Based on the same inventive concept, the embodiment of the present invention further provides a computer storage medium, which stores computer program codes, and when the computer program codes are run on a computing device, the computer storage medium causes the computing device to execute any one of the above watermark removing methods based on the generative network
Based on the same inventive concept, an embodiment of the present invention further provides a computing device, including:
a processor;
a memory storing computer program code;
the computer program code, when executed by the processor, causes the computing device to perform any of the above-described generative network-based watermark removal methods.
The embodiment of the invention provides a watermark removing scheme based on a generating type network, wherein the generating type confrontation network comprises a generator, in the embodiment, a training sample image is obtained, the obtained training sample image is input to the generator to generate a corresponding watermark-free image, then a generator loss function is calculated, a convolution kernel parameter to be learned is trained according to the calculated generator loss function, the steps are repeated, and when the generator loss function is converged, the trained convolution kernel parameter and the generator are stored as a trained watermark removing network model. Based on the scheme provided by the embodiment of the invention, the real distribution of the image can be better simulated, the original pixels of the image cannot be damaged while the watermark is effectively removed, the loss of image information in the watermark removing process is avoided, the watermark removing effect is improved, and the generalization capability is better.
In addition, the functional units in the embodiments of the present invention may be physically independent of each other, two or more functional units may be integrated together, or all the functional units may be integrated in one processing unit. The integrated functional units may be implemented in the form of hardware, or in the form of software or firmware.
Those of ordinary skill in the art will understand that: the integrated functional units, if implemented in software and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computing device (e.g., a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention when the instructions are executed. And the aforementioned storage medium includes: u disk, removable hard disk, Read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disk, and other various media capable of storing program code.
Alternatively, all or part of the steps of implementing the foregoing method embodiments may be implemented by hardware (such as a computing device, e.g., a personal computer, a server, or a network device) associated with program instructions, which may be stored in a computer-readable storage medium, and when the program instructions are executed by a processor of the computing device, the computing device executes all or part of the steps of the method according to the embodiments of the present invention.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments can be modified or some or all of the technical features can be equivalently replaced within the spirit and principle of the present invention; such modifications or substitutions do not depart from the scope of the present invention.
According to an aspect of the present invention, a1, a watermark removing method based on a generative countermeasure network, the generative countermeasure network including a generator, the method including:
acquiring a training sample image;
inputting the training sample image into the generator to generate a corresponding waterless image;
calculating a generator loss function, and training a convolution kernel parameter to be learned according to the generator loss function;
repeating the steps for multiple times, and when the loss function of the generator is converged, storing the convolution kernel parameters obtained by training and the generator as a trained watermark removal network model;
and inputting the image containing the watermark into the watermark removing network model to obtain a corresponding image with the watermark removed.
A2, the watermark removing method based on the generative countermeasure network according to A1, wherein before the obtaining of the training sample image, the method comprises:
and carrying out parameter initialization on the convolution kernel to be learned, and giving an initial value to the convolution kernel to be learned.
A3, the watermark removing method based on the generative countermeasure network according to A2, wherein the parameter initialization of the convolution kernel to be learned includes:
and initializing the convolution kernel parameters to be learned by using a Gaussian distribution with a mean value of 0 and a variance of 0.01.
A4, the watermark removing method based on the generative countermeasure network according to A1, wherein the inputting the training sample image into the generator to generate a corresponding watermark-free image comprises:
and inputting the water-containing print sample image into the generator, and sequentially performing convolution and deconvolution operations on the water-containing print sample image and the initial value of the convolution kernel parameter to output a corresponding water-free print image.
A5, the watermark removal method based on generative countermeasure network according to a1, wherein the generative countermeasure network further comprises a discriminator, the method further comprising:
the image after the watermark is removed and outputted by the watermark removing network model and the original image without the watermark are inputted into a discriminator together;
judging whether the image after the watermark removal output by the watermark removal network model meets the requirements or not;
and feeding back the judgment result to a watermark removal network model, and optimizing the watermark removal network model based on the judgment result.
A6, the watermark removal method based on generative countermeasure network according to a5, wherein the discriminator comprises a global discriminator and/or a local discriminator, the method further comprising:
calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator;
and adjusting the convolution kernel parameters to be learned according to the loss function.
A7, the method for removing watermark based on generation-based countermeasure network according to a5, wherein the determining whether the image after removing watermark output by the watermark removing network model meets the requirement, feeding back the determination result to the generator, and optimizing the watermark removing network model based on the determination result includes:
respectively extracting the image without the watermark and the corresponding overall image characteristics of the image without the watermark, and judging whether the overall image of the image without the watermark is a real image or not according to the extracted overall image characteristics; and/or
Respectively extracting local characteristics of a corresponding watermark removal area image part in the image without the watermark and a corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
and feeding back the judgment result to the global discriminator and/or the local discriminator, and optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
A8, the watermark removing method based on the generative countermeasure network according to A1, wherein before the obtaining of the training sample image, the method further comprises:
collecting a large number of images without watermarks and watermark images;
taking the image without the watermark as an original image, and adding the watermark image to the original image with random transparency to synthesize a watermark-containing image;
and generating a training sample image set by using the watermark-containing image as a training sample.
A9, the watermark removal method based on the generative countermeasure network according to a1, wherein before the training sample image is input into the generator to generate the corresponding watermark-free image, the method further comprises:
and carrying out scale normalization preprocessing on the input training sample image.
A10, the watermark removing method based on the generation-based countermeasure network according to A1, wherein the generator is a full convolution network with an encode-decode structure and comprises an encoder and a decoder;
the encoder comprises a first convolution module, a second convolution module and a third convolution module, and the decoder comprises a first deconvolution module, a second deconvolution module and a third deconvolution module;
each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module.
According to an aspect of the invention, B11, a watermark removing apparatus based on a generative countermeasure network is also disclosed, the generative countermeasure network includes a generator, the apparatus includes:
the acquisition module is used for acquiring a training sample image;
the generating module is used for inputting the training sample image into the generator to generate a corresponding watermark-free image;
the first calculation module is used for calculating the generator loss function and training the convolution kernel parameters to be learned according to the generator loss function;
the storage module is used for repeating the steps for multiple times, and when the loss function of the generator is converged, the convolution kernel parameters obtained by training and the generator are stored as a trained watermark removal network model;
and the watermark removing module is used for inputting the watermark-containing image into the watermark removing network model to obtain a corresponding image with the watermark removed.
B12, the watermark removing device based on the generative countermeasure network according to B11, further comprising:
and the initialization module is used for carrying out parameter initialization on the convolution kernel to be learned before the training sample image is obtained and endowing an initial value to the convolution kernel to be learned.
B13, the watermark removal device based on the generative countermeasure network according to B11, wherein,
the initialization module is further configured to initialize the convolution kernel parameters to be learned by using a gaussian distribution with a mean value of 0 and a variance of 0.01.
B14, the watermark removal device based on the generative countermeasure network according to B11, wherein,
and the generating module is also used for inputting the water-containing print sample image into the generator, sequentially carrying out convolution and deconvolution operations on the water-containing print sample image and the initial value of the convolution kernel parameter, and outputting a corresponding water-free print image.
B15, the watermark removal device based on the generative countermeasure network according to B11, wherein the generative countermeasure network further comprises a discriminator, the device further comprising:
the input module is used for inputting the image after the watermark is removed and output by the watermark removal network model and the original image without the watermark into the discriminator together;
the judging module is used for judging whether the image after the watermark is removed and output by the watermark removing network model meets the requirements or not;
and the optimizing module is used for feeding back the judgment result to the watermark removing network model and optimizing the watermark removing network model based on the judgment result.
B16, the watermark removal apparatus based on generative countermeasure network according to B15, wherein the discriminator comprises a global discriminator and/or a local discriminator, the apparatus further comprising:
and the second calculation module is used for calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator, and the convolution kernel parameters to be learned are adjusted according to the loss function.
B17, the watermark removal device based on the generative countermeasure network according to B15, wherein,
the judging module is further configured to respectively extract the image with the watermark removed and the corresponding overall image features of the image without the watermark, and judge whether the overall image of the image with the watermark removed is a real image according to the extracted overall image features; and/or
Respectively extracting local characteristics of a corresponding watermark removal area image part in the image without the watermark and a corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
and the optimization module is further used for feeding back the judgment result to the global discriminator and/or the local discriminator, optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
B18, the watermark removing device based on the generative countermeasure network according to B11, wherein the device further comprises:
the collecting module is used for collecting a large number of images without watermarks and watermark images;
and the watermarking module is used for taking the image without the watermark as an original image, adding the watermark image to the original image with random transparency to synthesize a watermarked image, and taking the watermarked image as a training sample to generate a training sample image set.
B19, the watermark removal device based on the generative countermeasure network according to B11, wherein,
the generation module is also used for carrying out scale normalization preprocessing on the input training sample image.
B20, the watermark removing device based on the generation type countermeasure network according to B11, wherein the generator is a full convolution network with an encode-decode structure and comprises an encoder and a decoder;
the encoder comprises a first convolution module, a second convolution module and a third convolution module, and the decoder comprises a first deconvolution module, a second deconvolution module and a third deconvolution module;
each convolution module and each deconvolution module respectively comprise two convolution layers connected in series, and 4 void convolution layers connected in series are arranged between the third convolution module and the first deconvolution module.
According to an aspect of the present invention, C21, a computer storage medium storing computer program code which, when run on a computing device, causes the computing device to execute the method for watermark removal based on generative networks as defined in any of a1-a10, is also disclosed.
Based on an aspect of the invention, there is also disclosed D22, a computing device, comprising:
a processor;
a memory storing computer program code;
the computer program code, when executed by the processor, causes the computing device to perform any of the generative network-based watermark removal methods of A1-A10.
Claims (10)
1. A watermark removal method based on a generative confrontation network, the generative confrontation network comprising a generator, the method comprising:
acquiring a training sample image;
inputting the training sample image into the generator to generate a corresponding waterless image;
calculating a generator loss function, and training a convolution kernel parameter to be learned according to the generator loss function;
repeating the steps for multiple times, and when the loss function of the generator is converged, storing the convolution kernel parameters obtained by training and the generator as a trained watermark removal network model;
and inputting the image containing the watermark into the watermark removing network model to obtain a corresponding image with the watermark removed.
2. The watermark removal method based on the generative countermeasure network according to claim 1, wherein prior to the obtaining of the training sample image, comprising:
and carrying out parameter initialization on the convolution kernel to be learned, and giving an initial value to the convolution kernel to be learned.
3. The watermark removing method based on the generative countermeasure network according to claim 2, wherein the parameter initialization of the convolution kernel to be learned comprises:
and initializing the convolution kernel parameters to be learned by using a Gaussian distribution with a mean value of 0 and a variance of 0.01.
4. The watermark removal method based on the generative countermeasure network as claimed in claim 1, wherein the inputting the training sample image into the generator to generate a corresponding watermark-free image comprises:
and inputting the water-containing print sample image into the generator, and sequentially performing convolution and deconvolution operations on the water-containing print sample image and the initial value of the convolution kernel parameter to output a corresponding water-free print image.
5. The watermark removal method based on the generative countermeasure network of claim 1, wherein the generative countermeasure network further comprises a discriminator, the method further comprising:
the image after the watermark is removed and outputted by the watermark removing network model and the original image without the watermark are inputted into a discriminator together;
judging whether the image after the watermark removal output by the watermark removal network model meets the requirements or not;
and feeding back the judgment result to a watermark removal network model, and optimizing the watermark removal network model based on the judgment result.
6. The watermark removal method based on the generative countermeasure network of claim 5, wherein the discriminator comprises a global discriminator and/or a local discriminator, the method further comprising:
calculating a loss function of the watermark removal network model, wherein the loss function comprises a loss function of the global discriminator and/or a loss function of the local discriminator;
and adjusting the convolution kernel parameters to be learned according to the loss function.
7. The watermark removing method based on the generative countermeasure network as claimed in claim 5, wherein the determining whether the image after removing the watermark output by the watermark removing network model meets the requirement, feeding back the determination result to the generator, and optimizing the watermark removing network model based on the determination result comprises:
respectively extracting the image without the watermark and the corresponding overall image characteristics of the image without the watermark, and judging whether the overall image of the image without the watermark is a real image or not according to the extracted overall image characteristics; and/or
Respectively extracting local characteristics of a corresponding watermark removal area image part in the image without the watermark and a corresponding watermark area image part in the image without the watermark, and judging whether the local image without the watermark is a real image or not according to the extracted local characteristics;
and feeding back the judgment result to the global discriminator and/or the local discriminator, and optimizing and storing the watermark removal network model when the whole image and/or the local image is a real image.
8. A watermark removal apparatus based on a generative confrontation network, the generative confrontation network comprising a generator, the apparatus comprising:
the acquisition module is used for acquiring a training sample image;
the generating module is used for inputting the training sample image into the generator to generate a corresponding watermark-free image;
the first calculation module is used for calculating the generator loss function and training the convolution kernel parameters to be learned according to the generator loss function;
the storage module is used for repeating the steps for multiple times, and when the loss function of the generator is converged, the convolution kernel parameters obtained by training and the generator are stored as a trained watermark removal network model;
and the watermark removing module is used for inputting the watermark-containing image into the watermark removing network model to obtain a corresponding image with the watermark removed.
9. A computer storage medium having computer program code stored thereon, which, when run on a computing device, causes the computing device to perform the method of generative network-based watermark removal of any of claims 1 to 7.
10. A computing device, comprising:
a processor;
a memory storing computer program code;
the computer program code, when executed by the processor, causes the computing device to perform the method of generative network-based watermark removal of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910190722.0A CN111696046A (en) | 2019-03-13 | 2019-03-13 | Watermark removing method and device based on generating type countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910190722.0A CN111696046A (en) | 2019-03-13 | 2019-03-13 | Watermark removing method and device based on generating type countermeasure network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111696046A true CN111696046A (en) | 2020-09-22 |
Family
ID=72475816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910190722.0A Pending CN111696046A (en) | 2019-03-13 | 2019-03-13 | Watermark removing method and device based on generating type countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111696046A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102141A (en) * | 2020-09-24 | 2020-12-18 | 腾讯科技（深圳）有限公司 | Watermark detection method, watermark detection device, storage medium and electronic equipment |
| CN112330522A (en) * | 2020-11-09 | 2021-02-05 | 深圳市威富视界有限公司 | Watermark removal model training method and device, computer equipment and storage medium |
| CN112419135A (en) * | 2020-11-19 | 2021-02-26 | 广州华多网络科技有限公司 | Watermark recognition online training, sampling and removing method, device, equipment and medium |
| CN112487369A (en) * | 2020-12-16 | 2021-03-12 | 湖南大学 | Frame loss resistant GIF dynamic image copyright authentication method |
| CN112686794A (en) * | 2020-12-21 | 2021-04-20 | 福建新大陆软件工程有限公司 | Watermark removing method based on generating type countermeasure network |
| CN113160156A (en) * | 2021-04-12 | 2021-07-23 | 佛山市顺德区美的洗涤电器制造有限公司 | Method for processing image, processor, household appliance and storage medium |
| CN113691885A (en) * | 2021-09-09 | 2021-11-23 | 深圳万兴软件有限公司 | Video watermark removing method and device, computer equipment and storage medium |
| CN114119335A (en) * | 2022-01-26 | 2022-03-01 | 南京信息工程大学 | Neural network watermark generation method, system and storage medium based on GAN |
-
2019
- 2019-03-13 CN CN201910190722.0A patent/CN111696046A/en active Pending
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112102141A (en) * | 2020-09-24 | 2020-12-18 | 腾讯科技（深圳）有限公司 | Watermark detection method, watermark detection device, storage medium and electronic equipment |
| CN112330522A (en) * | 2020-11-09 | 2021-02-05 | 深圳市威富视界有限公司 | Watermark removal model training method and device, computer equipment and storage medium |
| CN112330522B (en) * | 2020-11-09 | 2024-06-04 | 深圳市威富视界有限公司 | Watermark removal model training method, device, computer equipment and storage medium |
| CN112419135A (en) * | 2020-11-19 | 2021-02-26 | 广州华多网络科技有限公司 | Watermark recognition online training, sampling and removing method, device, equipment and medium |
| CN112487369A (en) * | 2020-12-16 | 2021-03-12 | 湖南大学 | Frame loss resistant GIF dynamic image copyright authentication method |
| CN112686794A (en) * | 2020-12-21 | 2021-04-20 | 福建新大陆软件工程有限公司 | Watermark removing method based on generating type countermeasure network |
| CN112686794B (en) * | 2020-12-21 | 2023-06-02 | 福建新大陆软件工程有限公司 | Watermark removing method based on generation type countermeasure network |
| CN113160156A (en) * | 2021-04-12 | 2021-07-23 | 佛山市顺德区美的洗涤电器制造有限公司 | Method for processing image, processor, household appliance and storage medium |
| CN113691885A (en) * | 2021-09-09 | 2021-11-23 | 深圳万兴软件有限公司 | Video watermark removing method and device, computer equipment and storage medium |
| CN113691885B (en) * | 2021-09-09 | 2024-01-30 | 深圳万兴软件有限公司 | Video watermark removal method and device, computer equipment and storage medium |
| CN114119335A (en) * | 2022-01-26 | 2022-03-01 | 南京信息工程大学 | Neural network watermark generation method, system and storage medium based on GAN |
| CN114119335B (en) * | 2022-01-26 | 2022-04-26 | 南京信息工程大学 | Neural network watermark generation method, system and storage medium based on GAN |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111696046A (en) | Watermark removing method and device based on generating type countermeasure network | |
| Jancsary et al. | Regression Tree Fields—An efficient, non-parametric approach to image labeling problems | |
| CN113658051A (en) | Image defogging method and system based on cyclic generation countermeasure network | |
| CN115661144B (en) | Adaptive medical image segmentation method based on deformable U-Net | |
| CN112307714B (en) | Text style migration method based on dual-stage depth network | |
| CN112784929B (en) | Small sample image classification method and device based on double-element group expansion | |
| CN110766708B (en) | Image comparison method based on contour similarity | |
| CN114998602B (en) | Domain adaptive learning method and system based on low confidence sample contrast loss | |
| CN114332466B (en) | Continuous learning method, system, equipment and storage medium for image semantic segmentation network | |
| CN112836602B (en) | Behavior recognition method, device, equipment and medium based on space-time feature fusion | |
| CN115063318A (en) | Adaptive frequency-resolved low-illumination image enhancement method and related equipment | |
| CN115953784A (en) | Laser coding character segmentation method based on residual error and feature blocking attention | |
| CN114820303A (en) | Method, system and storage medium for reconstructing super-resolution face image from low-definition image | |
| Camacho et al. | Convolutional neural network initialization approaches for image manipulation detection | |
| CN112819731B (en) | Gray scale image enhancement method, device, computer equipment and storage medium | |
| CN112560034B (en) | Malicious code sample synthesis method and device based on feedback type deep countermeasure network | |
| CN114022706A (en) | Method, device and equipment for optimizing image classification model and storage medium | |
| Lin | Comparative Analysis of Pix2Pix and CycleGAN for image-to-image translation | |
| CN115375966A (en) | Image countermeasure sample generation method and system based on joint loss function | |
| CN115294424A (en) | Sample data enhancement method based on generation countermeasure network | |
| CN112907503B (en) | Penaeus vannamei Boone quality detection method based on self-adaptive convolutional neural network | |
| Altakrouri et al. | Image to image translation networks using perceptual adversarial loss function | |
| Qaderi et al. | A Gated Deep Model for Single Image Super-Resolution Reconstruction | |
| CN110807777B (en) | Marine mammal image segmentation method based on convolutional neural network | |
| Zhangjie et al. | The secure steganography for hiding images via GAN |
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 |