CN106096668B - The recognition methods and identifying system of watermarked image - Google Patents
The recognition methods and identifying system of watermarked image Download PDFInfo
- Publication number
- CN106096668B CN106096668B CN201610688726.8A CN201610688726A CN106096668B CN 106096668 B CN106096668 B CN 106096668B CN 201610688726 A CN201610688726 A CN 201610688726A CN 106096668 B CN106096668 B CN 106096668B
- Authority
- CN
- China
- Prior art keywords
- watermark
- image
- images
- candidate region
- convolutional neural
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention discloses the recognition methods of watermarked image and identifying systems, choose all watermark candidate regions of images to be recognized；And watermark identification is carried out to each watermark candidate region by depth convolutional neural networks classifier, judge whether images to be recognized is watermarked image；Realize the identification of watermarked image.The present invention can conveniently and efficiently obtain a large amount of image training data, establish depth convolutional neural networks classifier by convolutional neural networks algorithm using a large amount of image training data, solve the problems, such as that training data is insufficient in the prior art.The problems such as depth convolutional neural networks classifier that the present invention establishes, effectively simulates human eye vision processing system, can recognize that the subtle watermark texture in part, preferably solves in watermarked image, and watermark occupied area is small, of light color, transparency is high.The present invention can reduce the identification process to no-watermark region, shorten recognition time, improve recognition efficiency.
Description
Technical field
The present invention relates to field of image recognition, and in particular to a kind of recognition methods and identifying system of watermarked image.
Background technique
Image contains abundant and intuitive information and requires currently in fields such as the social activities, shopping and tourism of internet
A large amount of image transmits information to user.It is more and more personal since the propagation of internet information is extremely quickly and portable
It is embedded in watermark information to owned image with organizational choice, the watermark of trade mark or network address is such as stamped in image section region,
Thereby protect the ownership of image information.Therefore, the provider of image information needs to examine image before using image
Core identifies in image whether contain watermark information, avoids the occurrence of the behavior of misuse and infringement.With the rapid development of Internet,
All multipaths such as image provider can be uploaded daily using user, crawler downloading obtain great amount of images information, and quantity is much
More than the limit of manual examination and verification.Therefore, image information is audited automatically using computer, identify that wherein the image with watermark becomes
Urgent demand.
The features such as vision significance of watermark information in the picture is very low, has area small, of light color, and transparency is high, band
Often very little, discrimination are lower for difference between watermarking images and non-watermarked image.At this stage, to watermarked image identification
Research not yet deeply expansion, rarely has effective watermarked image identification technology, realizes that accurately identifying for watermarked image is one
Challenging task.
Summary of the invention
It is an object of the invention to the image etc. with watermark information cannot be recognized accurately in the prior art ask to overcome
The appearance of topic；The recognition methods and identifying system of a kind of watermarked image are provided.
In order to achieve the above object, the invention is realized by the following technical scheme:
A kind of recognition methods of watermarked image, the recognition methods includes:
Choose all watermark candidate regions of images to be recognized；
Watermark identification carried out to each watermark candidate region by depth convolutional neural networks classifier, described in judgement
Whether images to be recognized is watermarked image.
Preferably, the method for establishing the depth convolutional neural networks classifier includes:
Generate image training data；
The depth convolutional neural networks are established by convolutional neural networks algorithm using described image training data to classify
Device.
Preferably, the implementation method for generating image training data are as follows:
Collect several anhydrous watermark images；
Intercept several rectangular region images, each rectangular region image at random in every anhydrous watermark image
As the rectangular image for not including watermark information；
Collect a variety of watermark information figures；Each watermark information figure is respectively embedded into each described not comprising watermark
In the rectangular image of information, a kind of image comprising the corresponding watermark information figure is formed；
It regard each rectangular image not comprising watermark information, each image comprising watermark information figure as institute
State image training data.
Preferably, the implementation method for establishing the depth convolutional neural networks classifier are as follows:
A1 initializes each layer parameter of depth convolutional neural networks classifier；
A2 is obtained defeated after successively calculating each described image training data by convolutional neural networks algorithm progress
It is worth out；
A3 calculates the error for obtaining each output valve and corresponding described image training data generic；According to
Minimum error principle carries out each layer parameter in depth convolutional neural networks classifier described in layer-by-layer correction by the error；
A4, repeating said steps A2, A3, until error convergence, realizes building for the depth convolutional neural networks classifier
It is vertical.
Preferably, including in the step of choosing all watermark candidate regions of images to be recognized:
The images to be recognized is divided into multiple candidate regions, the inspection of watermark feature point is carried out to each candidate region
It surveys, counts all watermark feature point quantity of each candidate region；When the watermark feature point of any candidate region is total
When quantity is greater than characteristic point threshold value, the watermark candidate regions of the corresponding candidate region as the images to be recognized
Domain.
Preferably, judging whether the images to be recognized is that the realization step of watermarked image includes:
The image data of each of the images to be recognized watermark candidate region is input to the depth convolution mind
In network classifier, each of the depth convolutional neural networks classifier the last layer output watermark candidate regions are obtained
Domain includes the probability vector of watermark information figure；
Calculating judges that the maximum probability value of all probability vectors indicates whether as comprising watermark information figure；When all
When the maximum value of the probability vector is represented as not including watermark information figure, figure of the images to be recognized without watermark
Picture；When the maximum value of at least one probability vector is expressed as comprising watermark information figure, the images to be recognized is band
Watermarking images.
A kind of identifying system of watermarked image, the identifying system includes:
Acquiring unit, for generating image training data；
Training unit is connect with the acquiring unit；The training unit obtains described image training data, establishes depth
Convolutional neural networks classifier；
Selecting unit is connect with the training unit；All watermarks that the selecting unit chooses images to be recognized are candidate
Region, and the image data of each watermark candidate region is input to the depth convolution mind that the training unit is established
Through in network classifier；
Recognition unit is connect with the training unit；The recognition unit obtains all general of the training unit output
Rate vector carry out the images to be recognized whether be watermarked image judgement.
Preferably, the acquiring unit includes:
Anhydrous watermark image generation module, connect with the training unit；The anhydrous watermark image generation module is collected several
Zhang Wushui watermark image, and several rectangular region images are intercepted at random in every anhydrous watermark image, it generates multiple and does not wrap
Rectangular image with watermarked information；
Generation module containing watermarking images is connect with the anhydrous watermark image generation module, the training unit respectively；It is described
Generation module containing watermarking images obtain it is all described in do not include watermark information rectangular image and collect a variety of watermark information figures
Shape, by each watermark information figure be respectively embedded into it is all described in do not include the rectangular image of watermark information, formed a kind of
Image comprising the corresponding watermark information figure.
Preferably, the selecting unit includes:
The images to be recognized is divided into multiple candidate regions by characteristic point detection module, to each candidate region
The detection of watermark feature point is carried out, all watermark feature point quantity of each candidate region are counted；
Watermark candidate region selecting module is connect with the characteristic point detection module, the training unit respectively；The water
Print candidate region selecting module obtains all watermark feature point quantity of each candidate region, judges each candidate regions
Whether the watermark feature point total quantity in domain is greater than characteristic point threshold value；When being greater than, the corresponding candidate region is as described wait know
The watermark candidate region of one of other image.
On the basis of common knowledge of the art, above-mentioned each optimum condition, can any combination to get each preferable reality of the present invention
Example.
The positive effect of the present invention is that:
The recognition methods and identifying system of a kind of watermarked image disclosed by the invention, firstly, choosing images to be recognized
All watermark candidate regions；Secondly, carrying out watermark knowledge to each watermark candidate region by depth convolutional neural networks classifier
Not, judge whether images to be recognized is watermarked image；Realize the identification of watermarked image.The present invention can be obtained conveniently and efficiently
A large amount of image training data is taken, and depth convolution is established by convolutional neural networks algorithm using a large amount of image training data
Neural network classifier solves the problems, such as that training data is insufficient in the prior art.The depth convolutional Neural net that the present invention establishes
Network classifier effectively simulates human eye vision processing system, can recognize that the subtle watermark texture in part, preferably solves
In watermarked image, the problems such as watermark occupied area is small, of light color, transparency is high.The present invention can be reduced to no-watermark area
The identification process in domain, shortens recognition time, improves recognition efficiency.The present invention also has practicability compared with the prior art
By force, the good advantage of scalability.
Detailed description of the invention
Fig. 1 is the overall structure diagram of the identifying system of watermarked image of the present invention.
Fig. 2 is the overall flow schematic diagram of the recognition methods of watermarked image of the present invention.
Specific embodiment
The present invention is further illustrated below by the mode of embodiment, but does not therefore limit the present invention to the reality
It applies among a range.
As shown in Figure 1, a kind of identifying system of watermarked image, identifying system includes: acquiring unit 1, training unit 2,
Selecting unit 3 and recognition unit 4.Wherein, training unit 2 is connect with acquiring unit 1, selecting unit 3 and recognition unit 4 respectively.
In the present invention, acquiring unit 1 is for generating image training data.Training unit 2 obtains image training data, establishes
Depth convolutional neural networks classifier.Selecting unit 3 chooses all watermark candidate regions of images to be recognized, and by each watermark
The image data of candidate region is input in depth convolutional neural networks classifier.Recognition unit 4 is used as watermark determination module, obtains
Take training unit 2 export all probability vectors carry out images to be recognized whether be watermarked image judgement.
As shown in Figure 1, acquiring unit 1 includes: anhydrous watermark image generation module 11, generation module containing watermarking images 12；Its
In, generation module containing watermarking images 12 is connect with anhydrous watermark image generation module 11.
In the present invention, anhydrous watermark image generation module 11 is used to collect several anhydrous watermark images automatically, and in every nothing
Several rectangular region images for intercepting certain proportion size in watermarking images at random, by all squares not comprising watermark information
Shape image generates no-watermark class image collection.
In the present invention, generation module containing watermarking images 12 collects the watermark information figure in a variety of present invention to be identified automatically
Shape, and no-watermark class image collection is obtained by anhydrous watermark image generation module 11.Generation module containing watermarking images 12 will be each
Watermark information figure to be identified is respectively embedded into all rectangular images not comprising watermark information, is formed a kind of comprising corresponding
The image collection of watermark information figure to be identified.
As shown in Figure 1, selecting unit 3 includes: characteristic point detection module 31, watermark candidate region selecting module 32；Wherein,
Watermark candidate region selecting module 32 is connect with characteristic point detection module 31.
In the present invention, images to be recognized is divided into multiple candidate regions by characteristic point detection module 31, to each candidate regions
Domain carries out the detection of watermark feature point, counts all watermark feature point quantity of each candidate region.
In the present invention, watermark candidate region selecting module 32 obtains all watermark feature point quantity of each candidate region,
Judge whether the watermark feature point total quantity of each candidate region is greater than characteristic point threshold value；When being greater than, corresponding candidate region is made
For a watermark candidate region of images to be recognized.
As shown in Fig. 2, a kind of recognition methods of watermarked image, recognition methods includes:
S1, acquiring unit 1 generate image training data.Concrete methods of realizing is as follows:
The anhydrous watermark image generation module 11 of S1.1, acquiring unit 1 collect several anhydrous watermark images.
S1.2, anhydrous watermark image generation module 11 intercept several rectangular area figures at random in every anhydrous watermark image
Picture, each rectangular region image is as the rectangular image for not including watermark information, and by all squares not comprising watermark information
Shape image forms no-watermark class image collection.
In the present embodiment, it is 1: 2 that anhydrous watermark image generation module 11 intercepts depth-width ratio at random in every anhydrous watermark image
Rectangular region image with 1: 3.6 size is as the rectangular image for not including watermark information.
S1.3, the generation module containing watermarking images 12 of acquiring unit 1 collect a variety of watermark information figures to be identified；It will be every
A watermark information figure to be identified is respectively embedded into each rectangular image not comprising watermark information, is formed a kind of comprising corresponding to
Watermark information figure to be identified image collection.
In the present embodiment, every kind of watermark information figure to be identified is first carried out gray scale two by generation module containing watermarking images 12
Value processing, then by this kind of watermark information figure to be identified with the not equal transparency of 60%-80% be respectively embedded into it is all not
In rectangular image comprising watermark information, as the image comprising corresponding watermark information figure.
For example, anhydrous watermark image generation module 11 collects 3000 anhydrous watermark images, every anhydrous watermark image intercepts 3 at random
A rectangular region image then generates altogether 9000 rectangular images for not including watermark information.Generation module containing watermarking images 12
3 kinds of watermark information figures to be identified are collected, then every kind of watermark information figure to be identified can generate 9000 comprising corresponding to
The image of watermark information figure；Then generation module containing watermarking images 12, which has altogether, generates 27000 figures comprising watermark information figure
Picture.
S1.4, by each rectangular image not comprising watermark information, each the image comprising watermark information to be identified is equal
As image training data.
In the present embodiment, generate 9000 of anhydrous watermark image generation module 11 do not include the image, aqueous of watermark information
Watermark image generation module 12 generates 27000 images comprising watermark information figure and is used as image training data.
S2, training unit 2 establish depth convolutional neural networks by convolutional neural networks algorithm using image training data
Classifier.In the present invention, the concrete methods of realizing for establishing depth convolutional neural networks classifier is as follows:
A1 initializes each layer parameter of depth convolutional neural networks classifier.
Each image training data exported after successively calculating by the convolutional neural networks algorithm by A2
Value.
A3 calculates the error for obtaining each output valve and corresponding image training data generic.According to minimal error
Criterion carries out each layer parameter in depth convolutional neural networks classifier described in layer-by-layer correction by the error.
In the present embodiment, image training data generic refers to that image training data is the rectangle not comprising watermark information
Image type or rectangular image type comprising variety classes watermark information figure.Such as generic includes: 1, does not include water
The rectangular image type of official seal breath, 2, the rectangular image type of the first type watermark information figure, 3, second of type watermark
The rectangular image type of information graphic, 4, the rectangular image type of the third type watermark information figure.
A4, repeating said steps A2, A3, until error convergence, realizes building for the depth convolutional neural networks classifier
It is vertical.
In the present embodiment, the image comprising watermark information figure and/or do not include watermark information that training unit 2 inputs
The size of rectangular image be uniformly adjusted to 227 × 227 pixel sizes.Depth convolutional neural networks are arranged in training unit 2
Classifier uses 8 layers of structure, and first 5 layers are convolutional layer, and the 6th layer and the 7th layer is full articulamentum, and the 8th layer is output layer.Wherein, it rolls up
The realization of lamination includes 3 convolution, activation and pondization steps.
Each layer activation primitive is amendment linear function f (x)=max (0, x).Each layer pond mode is max pooling, pond
Changing unit area is preferably 3 × 3 pixel sizes, and pond step-length is 2 pixels.The convolution kernel size of convolutional layer is by each layer
Depending on input, in the present embodiment, level 1 volume product core size is 11 × 11 × 3, and convolution kernel number is 96, step-length 4.2-5 layers
Convolution kernel size is respectively 5 × 5 × 96,3 × 3 × 256,3 × 3 × 384 and 3 × 3 × 384, convolution kernel number is respectively 256,
384,384 and 256, convolution step-length is 1.6th, 7 layer of output number is 4096, and the output of output layer is according to watermark to be identified
Depending on the type number of information.
S3, selecting unit 3 choose all watermark candidate regions of images to be recognized.It is specific to include following step in the present invention
It is rapid:
Images to be recognized is divided into multiple candidate regions by the characteristic point detection module 31 of S3.1, selecting unit 3, to each
Candidate region carries out the detection of watermark feature point, counts all watermark feature point quantity of each candidate region.
In the present invention, the type of watermark feature point includes but is not limited to SIFT, SURF, ORB etc..
S3.2, the watermark candidate region selecting module 32 of selecting unit 3 obtain all watermark features of each candidate region
Point quantity, judges whether the watermark feature point total quantity of each candidate region is greater than characteristic point threshold value；It is corresponding candidate when being greater than
A watermark candidate region of the region as images to be recognized.
In the present embodiment, the type of watermark feature point is ORB, and the candidate region width that characteristic point detection module 31 divides is
The 1/3 of images to be recognized width, the depth-width ratio of candidate region are 1: 2, and there are the overlapping of 50% area, characteristic point threshold in neighboring candidate region
Value is set as 10.
S4, training unit 2 carry out watermark identification to each watermark candidate region by depth convolutional neural networks classifier,
Recognition unit 4 judges whether images to be recognized is watermarked image.In the present invention, specific implementation step includes:
The image data of each watermark candidate region of images to be recognized is input to training unit 2 by S4.1, selecting unit 3
In the depth convolutional neural networks classifier of foundation, each watermark candidate region of images to be recognized is obtained in depth convolutional Neural
Network classifier the last layer output probability vector.
In the present embodiment, the picture size of each watermark candidate region is uniformly scaled 227 × 227 pictures by selecting unit 3
Training unit 2 is input to after plain size.Depth convolutional neural networks classifier inputs the image of a watermark candidate region every time
Data, after the calculating of depth convolutional neural networks, the watermark candidate region for obtaining this input includes watermark information figure
Probability vector.
S4.2, all watermark candidate regions that recognition unit 4 obtains the images to be recognized that training unit 2 exports include watermark
The probability vector of information graphic.Recognition unit 4 calculating judge the maximum probability value of all probability vectors indicate whether for comprising
Watermark information figure；It is described to be identified when the maximum value of all probability vectors is represented as not including watermark information figure
Image is the image without watermark；When the maximum value of at least one probability vector is expressed as comprising watermark information figure, wait know
Other image is watermarked image.
For example, images to be recognized includes 3 watermark candidate regions, each watermark candidate region is in depth convolutional neural networks
The output of classifier the last layer is a four-dimensional probability vector, respectively corresponds watermark candidate region and belongs to not comprising watermark information
Rectangular image type and the rectangular image type comprising other 3 kinds of type watermark information figures probability, calculate each four-dimension
Maximum probability value in probability vector, to judge whether current watermark candidate region includes watermark information figure.When 3 water
Print candidate region is judged as not comprising watermark information figure, then it represents that images to be recognized is the image without watermark.When 3 water
Any watermark candidate region is judged as comprising watermark information figure in print candidate region, then it represents that images to be recognized is band watermark
Image.
Although specific embodiments of the present invention have been described above, it will be appreciated by those of skill in the art that these
It is merely illustrative of, protection scope of the present invention is defined by the appended claims.Those skilled in the art is not carrying on the back
Under the premise of from the principle and substance of the present invention, many changes and modifications may be made, but these are changed
Protection scope of the present invention is each fallen with modification.
Claims (6)
1. a kind of recognition methods of watermarked image, which is characterized in that the recognition methods includes:
Choose all watermark candidate regions of images to be recognized；
Watermark identification is carried out to each watermark candidate region by depth convolutional neural networks classifier, judgement is described wait know
Whether other image is watermarked image；
The method for establishing the depth convolutional neural networks classifier includes:
Generate image training data；
The depth convolutional neural networks classifier is established by convolutional neural networks algorithm using described image training data；
The implementation method for generating image training data are as follows:
Collect several anhydrous watermark images；
Intercept several rectangular region images, each rectangular region image conduct at random in every anhydrous watermark image
Rectangular image not comprising watermark information；
Collect a variety of watermark information figures；Each watermark information figure is subjected to binarization of gray value processing, and will be by ash
The watermark information figure of degree binary conversion treatment is respectively embedded into each described not comprising watermark letter with the transparency that 60%-80% is not waited
In the rectangular image of breath, a kind of image comprising the corresponding watermark information figure is formed；
It regard each rectangular image not comprising watermark information, each image comprising watermark information figure as the figure
As training data.
2. the recognition methods of watermarked image as described in claim 1, which is characterized in that described to establish the depth convolution mind
Implementation method through network classifier are as follows:
A1 initializes each layer parameter of depth convolutional neural networks classifier；
Each described image training data exported after successively calculating by the convolutional neural networks algorithm by A2
Value；
A3 calculates the error for obtaining each output valve and corresponding described image training data generic；According to minimum
Error criterion carries out each layer parameter in depth convolutional neural networks classifier described in layer-by-layer correction by the error；
A4, repeating said steps A2, A3, until error convergence, realizes the foundation of the depth convolutional neural networks classifier.
3. the recognition methods of watermarked image as described in claim 1, which is characterized in that choosing all of images to be recognized
Include in the step of watermark candidate region:
The images to be recognized is divided into multiple candidate regions, the detection of watermark feature point is carried out to each candidate region,
Count all watermark feature point quantity of each candidate region；When the watermark feature point total quantity of any candidate region
When greater than characteristic point threshold value, the watermark candidate region of the corresponding candidate region as the images to be recognized.
4. the recognition methods of watermarked image as claimed in claim 2, which is characterized in that whether judge the images to be recognized
Include for the realization step of watermarked image:
The image data of each of the images to be recognized watermark candidate region is input to the depth convolutional Neural net
In network classifier, each of described depth convolutional neural networks classifier the last layer output watermark candidate region packet is obtained
The probability vector of figure with watermarked information；
Calculating judges that the maximum probability value of all probability vectors indicates whether as comprising watermark information figure；Described in all
When the maximum value of probability vector is represented as not including watermark information figure, the images to be recognized is the image without watermark；
When the maximum value of at least one probability vector is expressed as comprising watermarking images, the images to be recognized is band watermark figure
Picture.
5. a kind of identifying system of watermarked image, which is characterized in that the identifying system includes:
Acquiring unit, for generating image training data；
Training unit is connect with the acquiring unit；The training unit obtains described image training data, establishes depth convolution
Neural network classifier；
Selecting unit is connect with the training unit；The selecting unit chooses all watermark candidate regions of images to be recognized,
And the image data of each watermark candidate region is input to the depth convolutional Neural net that the training unit is established
In network classifier；
Recognition unit is connect with the training unit；The recognition unit obtain all probability of training unit output to
Amount carry out the images to be recognized whether be watermarked image judgement；
The acquiring unit includes:
Anhydrous watermark image generation module, connect with the training unit；The anhydrous watermark image generation module collects several nothings
Watermarking images, and several rectangular region images are intercepted at random in every anhydrous watermark image, multiple are generated not comprising water
The rectangular image of official seal breath；
Generation module containing watermarking images is connect with the anhydrous watermark image generation module, the training unit respectively；It is described aqueous
Watermark image generation module obtains all rectangular images not comprising watermark information and collects a variety of watermark information figures, will
Each watermark information figure carries out binarization of gray value processing, and by the watermark information figure by binarization of gray value processing with
Equal transparency is not respectively embedded into all rectangular images not comprising watermark information to 60%-80%, forms one kind and includes
The image of the corresponding watermark information figure.
6. the identifying system of watermarked image as claimed in claim 5, which is characterized in that the selecting unit includes:
The images to be recognized is divided into multiple candidate regions by characteristic point detection module, is carried out to each candidate region
The detection of watermark feature point, counts all watermark feature point quantity of each candidate region；
Watermark candidate region selecting module is connect with the characteristic point detection module, the training unit respectively；The watermark is waited
Favored area selecting module obtains all watermark feature point quantity of each candidate region, judges each candidate region
Whether watermark feature point total quantity is greater than characteristic point threshold value；When being greater than, the corresponding candidate region is as the figure to be identified
The watermark candidate region of one of picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610688726.8A CN106096668B (en) | 2016-08-18 | 2016-08-18 | The recognition methods and identifying system of watermarked image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610688726.8A CN106096668B (en) | 2016-08-18 | 2016-08-18 | The recognition methods and identifying system of watermarked image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106096668A CN106096668A (en) | 2016-11-09 |
| CN106096668B true CN106096668B (en) | 2019-06-18 |
Family
ID=58069617
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610688726.8A Active CN106096668B (en) | 2016-08-18 | 2016-08-18 | The recognition methods and identifying system of watermarked image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106096668B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108986295B (en) * | 2017-06-02 | 2020-10-20 | 深圳怡化电脑股份有限公司 | Black watermark recognition method and device and terminal equipment |
| CN107730453A (en) * | 2017-11-13 | 2018-02-23 | 携程计算机技术（上海）有限公司 | Picture quality method for improving |
| CN107808358B (en) * | 2017-11-13 | 2021-11-05 | 携程计算机技术（上海）有限公司 | Automatic detection method for image watermark |
| CN107911753B (en) * | 2017-11-28 | 2021-01-22 | 百度在线网络技术（北京）有限公司 | Method and device for adding digital watermark in video |
| US10699358B2 (en) | 2018-02-22 | 2020-06-30 | Mcafee, Llc | Image hidden information detector |
| CN110211015B (en) * | 2018-02-28 | 2022-12-20 | 佛山科学技术学院 | Watermark method based on characteristic object protection |
| CN108596916B (en) * | 2018-04-16 | 2021-02-26 | 深圳市联软科技股份有限公司 | Watermark identification method, system, terminal and medium with similar colors |
| CN108960256A (en) * | 2018-06-28 | 2018-12-07 | 东软集团股份有限公司 | A kind of determination method, device and equipment of components damage degree |
| CN109102451B (en) * | 2018-07-24 | 2023-03-21 | 齐鲁工业大学 | Anti-counterfeiting halftone intelligent digital watermark manufacturing method for paper media output |
| CN109325169A (en) * | 2018-07-25 | 2019-02-12 | 北京奔流网络信息技术有限公司 | A kind of copyright image filtering method and device |
| CN109636710A (en) * | 2018-10-16 | 2019-04-16 | 平安好房（上海）电子商务有限公司 | Automatically remove method, equipment, storage medium and the device of watermark |
| CN109598231B (en) * | 2018-12-03 | 2021-03-02 | 广州市百果园信息技术有限公司 | Video watermark identification method, device, equipment and storage medium |
| CN109859372A (en) * | 2018-12-07 | 2019-06-07 | 保定钞票纸业有限公司 | Watermark recognition methods, device, cloud server and the system of anti-forge paper |
| CN109784181B (en) * | 2018-12-14 | 2024-03-22 | 平安科技（深圳）有限公司 | Picture watermark identification method, device, equipment and computer readable storage medium |
| CN110222752B (en) * | 2019-05-28 | 2021-11-16 | 北京金山数字娱乐科技有限公司 | Image processing method, system, computer device, storage medium and chip |
| CN110852242A (en) * | 2019-11-06 | 2020-02-28 | 北京字节跳动网络技术有限公司 | Watermark identification method, device, equipment and storage medium based on multi-scale network |
| CN110991488B (en) * | 2019-11-08 | 2023-10-20 | 广州坚和网络科技有限公司 | Picture watermark identification method using deep learning model |
| CN111445376B (en) * | 2020-03-24 | 2023-08-18 | 五八有限公司 | Video watermark detection method, device, electronic equipment and storage medium |
| CN113837914A (en) * | 2020-06-08 | 2021-12-24 | 北京金山办公软件股份有限公司 | Watermark identification method and system based on artificial intelligence |
| CN112132460A (en) * | 2020-09-22 | 2020-12-25 | 京东城市(北京)数字科技有限公司 | Method, device and system for identifying potential danger area and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1945597A (en) * | 2006-11-03 | 2007-04-11 | 北京启明星辰信息技术有限公司 | Blind detecting system and method for digital watermarking flooding |
| CN102592256A (en) * | 2011-12-28 | 2012-07-18 | 辽宁师范大学 | Digital image watermark detection method based on support vector machine correction |
| CN102903071A (en) * | 2011-07-27 | 2013-01-30 | 阿里巴巴集团控股有限公司 | Watermark adding method and system as well as watermark identifying method and system |
| CN104778702A (en) * | 2015-04-15 | 2015-07-15 | 中国科学院自动化研究所 | Image stego-detection method on basis of deep learning |
-
2016
- 2016-08-18 CN CN201610688726.8A patent/CN106096668B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1945597A (en) * | 2006-11-03 | 2007-04-11 | 北京启明星辰信息技术有限公司 | Blind detecting system and method for digital watermarking flooding |
| CN102903071A (en) * | 2011-07-27 | 2013-01-30 | 阿里巴巴集团控股有限公司 | Watermark adding method and system as well as watermark identifying method and system |
| CN102592256A (en) * | 2011-12-28 | 2012-07-18 | 辽宁师范大学 | Digital image watermark detection method based on support vector machine correction |
| CN104778702A (en) * | 2015-04-15 | 2015-07-15 | 中国科学院自动化研究所 | Image stego-detection method on basis of deep learning |
Non-Patent Citations (2)
| Title |
|---|
| "Deep learning for steganalysis via convolutional neural networks";Yinlong Qian etc，;《Proceedings of SPIE-TheInternational Society for Optical Engineering》;20151231;摘要、第2.3节、图5、第4节 |
| "基于特征点的抗几何攻击的图像盲水印技术研究";余艳玮;《万方数据知识服务平台》;20110328;第4.2.1-4.2.2节、第5.1节、第5.2.3节、图5.1 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106096668A (en) | 2016-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106096668B (en) | The recognition methods and identifying system of watermarked image | |
| Yang et al. | An embedding cost learning framework using GAN | |
| Yang et al. | Source camera identification based on content-adaptive fusion residual networks | |
| CN106530200B (en) | Steganographic image detection method and system based on deep learning model | |
| CN105117729B (en) | A kind of method and apparatus of identification reproduction image | |
| CN111415316A (en) | Defect data synthesis algorithm based on generation of countermeasure network | |
| CN108596818B (en) | Image steganalysis method based on multitask learning convolutional neural network | |
| CN109558806A (en) | The detection method and system of high score Remote Sensing Imagery Change | |
| Wu et al. | Steganalysis via deep residual network | |
| CN113762138B (en) | Identification method, device, computer equipment and storage medium for fake face pictures | |
| CN104636764B (en) | A kind of image latent writing analysis method and its device | |
| CN112801846B (en) | Watermark embedding and extracting method and device, computer equipment and storage medium | |
| CN102722858B (en) | Blind steganalysis method based on symmetric neighborhood information | |
| Chen et al. | SNIS: A signal noise separation-based network for post-processed image forgery detection | |
| Hou et al. | Detection of hue modification using photo response nonuniformity | |
| Zhang et al. | Distinguishing photographic images and photorealistic computer graphics using visual vocabulary on local image edges | |
| Wang et al. | Wavelet based region duplication forgery detection | |
| CN108230269B (en) | Grid removing method, device and equipment based on depth residual error network and storage medium | |
| Chen et al. | Image splicing localization using residual image and residual-based fully convolutional network | |
| CN110533575A (en) | A kind of depth residual error steganalysis method based on isomery core | |
| CN113052923A (en) | Tone mapping method, tone mapping apparatus, electronic device, and storage medium | |
| Li et al. | Distinguishing computer graphics from photographic images using a multiresolution approach based on local binary patterns | |
| Mohamed et al. | Detecting Secret Messages in Images Using Neural Networks | |
| Le et al. | Representing visual complexity of images using a 3d feature space based on structure, noise, and diversity | |
| Kumar et al. | Image steganography analysis based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |