CN104680473A - Machine learning-based color image watermark embedding and detecting method - Google Patents
Machine learning-based color image watermark embedding and detecting method Download PDFInfo
- Publication number
- CN104680473A CN104680473A CN201410793279.3A CN201410793279A CN104680473A CN 104680473 A CN104680473 A CN 104680473A CN 201410793279 A CN201410793279 A CN 201410793279A CN 104680473 A CN104680473 A CN 104680473A
- Authority
- CN
- China
- Prior art keywords
- low frequency
- image
- component
- frequency coefficient
- normalized
- 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
Landscapes
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The invention discloses a machine learning-based color image watermark embedding and detecting method. The method comprises the following steps of firstly, using a matrix-based image normalizing technology to map a carrier image to a geometric invariant space, and combining a invariant centroid theory to determine an important area in a normalized color image; secondly, performing non-subsample Contourlet conversion on the important area of the G-component normalized color image; finally, according to a G-component visual mask model, adaptively determining embedding intensity, so that an image watermark is embedded in a low-frequency area, and the embedded watermark is guaranteed to have very good transparency and very high robustness; during watermark detection, by using the height relativity among color image components, selecting a stable feature vector training SVR (Support Vector Regression) model, and using the SVR training model to extract digital watermark information.
Description
Technical field
The invention belongs to Information hiding and digital watermark technology field in multi-media information security, especially one not only has preferably not sentience, and all has the embedding of the Color digital watermarking based on machine learning and the detection method of good robustness to normal signal process (medium filtering, edge sharpening, superimposed noise and JPEG compression etc.) and desynchronization attack (rotation, translation, convergent-divergent, shearing, upset etc.).
Background technology
Digital watermarking (Digital Watermarking) is as effective means of supplementing out economy of conventional encryption methods; it is a kind of new technology can protecting copyright and certification source and integrality under open network environment; cause people to pay much attention in recent years, and become a focus of international academic community research.So-called digital figure watermark; the mark (watermark) of certain sense will be had exactly; the method utilizing data to embed is hidden in digital picture product; in order to prove the entitlement of creator to its works; and as qualification, the illegal foundation of encroaching right of prosecution; simultaneously by ensureing the complete reliability of numerical information to the determination and analysis of watermark, thus become intellectual property protection and the false proof effective means of digital multimedia.
So-called desynchronization attack, not refers to that this kind of attack can remove watermark information from containing watermarking images, and refers to that it can destroy synchronous (namely the changing watermark embedment position) of digital watermarking component, thus cause detecting device to can not find effective watermark.Desynchronization attack comprises global affine transformation (i.e. rotation, convergent-divergent, translation) and general desynchronization attack (shearing, change of scale, ranks removal etc.).In recent years, people mainly adopt the New Scheme of Image Watermarking of three kinds of anti-desynchronization attack of Measure Design, are respectively constructive geometry invariant, hide masterplate, utilize original image key character.
In recent years, anti-desynchronization attack digital figure watermark embeds and detection method research has made great progress, but existing most image watermark embedded mobile GIS is all for gray level image, and the digital watermarking algorithm being directly used in color host image is less.Even if initial carrier is coloured image, most of method is also just by extracting its monochrome information or using the digital watermarking of individual color channel information insertion.Also just says, existing algorithm fails to embody very well and the specific contact of reservation different color components in whole color space, thus must affect robustness and the not sentience of digital watermarking.
Summary of the invention
The present invention is directed to the problems referred to above that conventional images water mark method exists, there is provided one not only to have preferably not sentience, and to normal signal process (medium filtering, edge sharpening, superimposed noise and JPEG compression etc.) and desynchronization attack (rotation, translation, convergent-divergent, shearing, upset etc.), all there is the embedding of the Color digital watermarking based on machine learning and the detection method of good robustness.
Technical solution of the present invention is: a kind of Color digital watermarking embedding grammar based on machine learning, it is characterized in that carrying out in accordance with the following steps successively:
Step 1: utilize the image normalization technology based on square to be mapped in the geometry invariant space by original color image F, respectively each component of coloured image is normalized, to obtain normalized color image;
Step 2: determine important area in conjunction with Invariant centroid theory in normalized color image, extract important area
;
Step 3: according to visual masking model, self-adaptation determination embedment strength, by digital watermark embedding in the low frequency region of non-downsampling Contourlet conversion, obtains moisture printed color picture
.
Described step 1 is as follows:
Step 11: RGB tri-components extracting original color image respectively;
Step 12: gray level image normalized is carried out to R component;
Step 13: the barycenter calculating R component normalized image, according to Invariant centroid definition, it can be used as the barycenter of G component and B component normalized image, carries out gray level image normalized respectively to G component and B component;
Step 14: in conjunction with the normalized image of RGB tri-components, obtains normalized color image.
Described step 2 is as follows:
Step 21: utilize Gaussian filter to the smoothing process of normalized color image, disturb with stress release treatment;
Step 22: according to Invariant centroid definition, calculate the barycenter of R component normalized image characteristic area
, and it can be used as the Invariant centroid initial value of normalized image;
Step 23: according to Invariant centroid definition, calculate with
for the center of circle,
for the Invariant centroid of the border circular areas of radius
;
Step 24: if
, then 25 are gone to step; Otherwise, order
, and go to step 23;
Step 25:
be the Invariant centroid of R component normalized image;
Step 26: point centered by the Invariant centroid of R component normalized image, choosing size is
rectangular area as the important area of R component normalized image;
In like manner, extract the important area of G component and B component normalized image respectively, obtain the important area of normalized color image.
Described step 3 is as follows:
Step 31: 3 grades of non-downsampling Contourlet conversions are implemented to the important area of G component normalized image, chooses low frequency region
aas digital watermark embedding district, obtain the low frequency coefficient of non-downsampling Contourlet conversion accordingly;
Step 32: be in size
moving window in by low frequency region
abe divided into low frequency coefficient block
, utilize key
, and according to G component visual masking value, at low frequency region
ain choose and not overlap
individual position as the embedded location of watermark signal, wherein, front portion
individual low frequency coefficient block
for watermark embedment, rear portion
individual low frequency coefficient block
sVR for watermark extracting trains;
Step 33: in the important area of G component normalized image, selected by amendment
individual low frequency coefficient block
b k interior low frequency coefficient, embed to complete watermark information, the embedding method adopted is:
Wherein,
with
be respectively low frequency coefficient block before and after amendment
b k interior low frequency coefficient,
v k for low frequency coefficient block
b k interior low frequency coefficient average,
for the embedment strength of watermark;
Step 34: according to G component visual masking model, the embedment strength of Automatic adjusument watermark
, the regulation strategy adopted is:
Wherein,
m f for final HVS model,
for weights, value, between 0.4 to 0.5, is used for correcting distortion,
m l for Intensity model,
m t for texture model,
for the important area of G component normalized image,
for edge detecting operation,
the edge treated of expansion,
m e for edge model,
for wave filter;
Step 35: the acquisition of moisture printed color picture, replaces former low frequency coefficient with the low frequency coefficient of amended non-downsampling Contourlet conversion and carries out inverse non-downsampling Contourlet conversion, to obtain the normalization important area containing watermark; In conjunction with the insignificant part in former normalized image region, to obtain containing watermark normalization characteristic region; Carry out pre-distortion compensated to obtain containing watermark G component image; Utilize containing watermark G component image and former R component, B component image, moisture printed color picture can be obtained
.
The detection method corresponding with above-mentioned embedding grammar is as follows:
Step 41: utilize the image normalization technology based on square, treating sense colors image is
be normalized, to obtain corresponding normalized color image;
Step 42: calmodulin binding domain CaM Invariant centroid is theoretical, extracts important area from normalized color image;
Step 43: respectively 3 grades of non-downsampling Contourlet conversions are implemented to the normalization important area of R component and G component, and choose corresponding low frequency region
with
;
Step 44: be in size
moving window in by low frequency region
with
be divided into low frequency coefficient block
with
, utilize the key identical with telescopiny
, respectively at low frequency region
with
in choose and not overlap
individual low frequency coefficient block, wherein,
individual low frequency coefficient block
with
for SVR training,
individual low frequency coefficient block
with
for the extraction of digital watermarking;
Step 45: establish low frequency coefficient block
the average of interior low frequency coefficient is
if, low frequency coefficient block
the average of interior low frequency coefficient is
, will
as training characteristics value, and incite somebody to action
as training objective value, so following vector set can be defined:
Carry out training study (suitable kernel function will be selected during training), SVR training pattern can be obtained;
Step 46: establish low frequency coefficient block
the average of interior low frequency coefficient is
, will
as training characteristics, utilize acquired SVR training pattern to carry out data prediction, corresponding output vector value can be obtained
, by comparing
with output vector value
size extract digital watermarking, extracting method is
Described
for low frequency coefficient block
the average of interior low frequency coefficient;
Step 47: to binary sequence
carry out a liter dimension process, and carry out inverted disorderly deciphering, namely obtain extracted binary image watermarking
.
The present invention is in conjunction with the high correlation between color image color component, propose a kind of Color digital watermarking based on machine learning to embed and detection method, not only there is good not sentience, and to normal signal process (medium filtering, edge sharpening, superimposed noise and JPEG compression etc.) and desynchronization attack (rotation, translation, convergent-divergent, shearing, upset etc.), all there is good robustness.In addition the present invention also have calculate simple, easily realize, when extracting watermark without the need to features such as initial carrier images, greatly strengthen its practicality for digital picture Works copyright protection.
Embodiment
Embedding grammar of the present invention comprises the following steps:
Step 1: utilize the image normalization technology based on square to be mapped in the geometry invariant space by original color carrier image F, be normalized respectively to each component of coloured image, to obtain corresponding normalized color image, this step is as follows:
Step 11: RGB tri-components extracting original color image respectively;
Step 12: gray level image normalized is carried out to R component;
Step 13: the barycenter calculating R component normalized image, according to Invariant centroid definition, it can be used as the barycenter of G component and B component normalized image, carries out gray level image normalized respectively to G component and B component;
Step 14: in conjunction with the normalized image of RGB tri-components, obtains corresponding normalized color image
.
Step 2: determine important area in conjunction with Invariant centroid theory in normalized color image, extract important area
, this step:
Step 21: utilize Gaussian filter to the smoothing process of normalized color image, disturb with stress release treatment;
Step 22: according to Invariant centroid definition, calculate the barycenter of R component normalized image characteristic area
, and it can be used as the Invariant centroid initial value of normalized image;
Step 23: according to Invariant centroid definition, calculate with
for the center of circle,
for the Invariant centroid of the border circular areas of radius
;
Step 24: if
, then 25 are gone to step; Otherwise, order
, and go to step 23;
Step 25:
for the Invariant centroid of R component normalized image;
Step 26: point centered by the Invariant centroid of R component normalized image, choosing size is
rectangular area as the important area of R component normalized image;
In like manner, extract the important area of G component and B component normalized image respectively, obtain the important area of corresponding normalized color image.
Step 3: according to visual masking model, self-adaptation determination embedment strength, by digital watermark embedding in the low frequency region of non-downsampling Contourlet conversion, this step is as follows:
Step 31: choosing of watermark embedment position.
In size be
moving window in by low frequency region
abe divided into low frequency coefficient block
.Utilize key
, and according to G component visual masking value, at low frequency region
ain choose and not overlap
individual position is as the embedded location of watermark signal.Wherein, front portion
individual low frequency coefficient block
for watermark embedment, rear portion
individual low frequency coefficient block
sVR for watermark extracting trains;
Step 32: embedding algorithm.
In the important area of G component normalized image, selected by amendment
individual low frequency coefficient block
b k interior low frequency coefficient, embed to complete watermark information, the embedding method adopted is:
Wherein,
with
be respectively low frequency coefficient block before and after amendment
b k interior low frequency coefficient,
v k for low frequency coefficient block
b k interior low frequency coefficient average,
for the embedment strength of watermark.The present invention according to G component visual masking model, the embedment strength of Automatic adjusument watermark
.The regulation strategy adopted is:
Wherein,
m f for final HVS model,
for weights, value, between 0.4 to 0.5, is used for correcting distortion.
m l for Intensity model,
m t for texture model.
for the important area of G component normalized image,
for edge detecting operation,
the edge treated of expansion,
m e for edge model,
for wave filter.
Step 33: the acquisition of moisture printed color picture, comprises the following steps:
Step 331: replace former low frequency coefficient with the low frequency coefficient of amended non-downsampling Contourlet conversion and carry out inverse non-downsampling Contourlet conversion, to obtain the normalization important area containing watermark;
Step 332: in conjunction with the insignificant part in former normalized image region, to obtain containing watermark normalization characteristic region;
Step 333: carry out pre-distortion compensated to obtain containing watermark G component image, its object is to reduce normalization and operates the distortion caused carrier image, namely improve the transparency of digital watermarking.Concrete operations are: the difference (image) calculating original normalized image and contain between watermark normalized image
; To difference (image)
do inverse normalization operation, obtain unfavourable balance value (image)
; By unfavourable balance value (image)
directly be superimposed on original image, can obtain containing watermark G component image.
Step 334: utilize containing watermark G component image and former R component, B component image, moisture printed color picture can be obtained
.
As follows to corresponding Color digital watermarking detecting step with above-mentioned embedding grammar:
Step 41: utilize the image normalization technology based on square, treat sense colors image
be normalized, to obtain corresponding normalized color image;
Step 42: calmodulin binding domain CaM Invariant centroid is theoretical, extracts important area from normalized color image;
Step 43: respectively 3 grades of non-downsampling Contourlet conversions are implemented to the normalized image important area of R component and G component, and choose corresponding low frequency region
with
;
Step 44: be in size
moving window in by low frequency region
with
be divided into low frequency coefficient block
with
.Utilize the key identical with telescopiny
, respectively at low frequency region
with
in choose and not overlap
individual low frequency coefficient block.Wherein,
individual low frequency coefficient block
with
for SVR training,
individual low frequency coefficient block
with
for the extraction of digital watermarking;
Step 45:SVR trains.
If low frequency coefficient block
the average of interior low frequency coefficient is
.If low frequency coefficient block
the average of interior low frequency coefficient is
.
Will
as training characteristics value, and incite somebody to action
as training objective value, so following vector set can be defined:
Carry out training study (suitable kernel function will be selected during training), SVR training pattern can be obtained;
Step 46: data prediction and watermark extracting.
Step 461: establish low frequency coefficient block
the average of interior low frequency coefficient is
.The present invention will
as training characteristics;
Step 462: utilize acquired SVR training pattern to carry out data prediction, corresponding output vector value can be obtained
;
Step 463: by comparing
(
for low frequency coefficient block
the average of interior low frequency coefficient) and output vector value
size extract digital watermarking, extracting method is
Step 47: to binary sequence
carry out a liter dimension process, and carry out inverted disorderly deciphering, namely obtain extracted binary image watermarking
.
Claims (5)
1., based on a Color digital watermarking embedding grammar for machine learning, it is characterized in that successively according to such as
Lower step is carried out:
Step 1: utilize the image normalization technology based on square to be mapped in the geometry invariant space by original color image F, respectively each component of coloured image is normalized, to obtain normalized color image;
Step 2: determine important area in conjunction with Invariant centroid theory in normalized color image, extract important area
;
Step 3: according to visual masking model, self-adaptation determination embedment strength, by digital watermark embedding in the low frequency region of non-downsampling Contourlet conversion, obtains moisture printed color picture
.
2. the Color digital watermarking embedding grammar based on machine learning according to claim 1, its feature
Be that described step 1 is as follows:
Step 11: RGB tri-components extracting original color image respectively;
Step 12: gray level image normalized is carried out to R component;
Step 13: the barycenter calculating R component normalized image, according to Invariant centroid definition, it can be used as the barycenter of G component and B component normalized image, carries out gray level image normalized respectively to G component and B component;
Step 14: in conjunction with the normalized image of RGB tri-components, obtains normalized color image.
3. the Color digital watermarking embedding grammar based on machine learning according to claim 2, is characterized in that described step 2 is as follows:
Step 21: utilize Gaussian filter to the smoothing process of normalized color image, disturb with stress release treatment;
Step 22: according to Invariant centroid definition, calculate the barycenter of R component normalized image characteristic area
, and it can be used as the Invariant centroid initial value of normalized image;
Step 23: according to Invariant centroid definition, calculate with
for the center of circle,
rfor the Invariant centroid of the border circular areas of radius
;
Step 24: if
, then 25 are gone to step; Otherwise, order
, and go to step 23;
Step 25:
be the Invariant centroid of R component normalized image;
Step 26: point centered by the Invariant centroid of R component normalized image, choosing size is
s1 × S2rectangular area as the important area of R component normalized image;
In like manner, extract the important area of G component and B component normalized image respectively, obtain the important area of normalized color image.
4. the Color digital watermarking embedding grammar based on machine learning according to claim 3, is characterized in that described step 3 is as follows:
Step 31: 3 grades of non-downsampling Contourlet conversions are implemented to the important area of G component normalized image, chooses low frequency region
aas digital watermark embedding district, obtain the low frequency coefficient of non-downsampling Contourlet conversion accordingly;
Step 32: by low frequency region in the moving window that size is 2 × 2
abe divided into low frequency coefficient block
, utilize key
k 1 , and according to G component visual masking value, at low frequency region
ain choose and not overlap
p × Q+Hindividual position as the embedded location of watermark signal, wherein, front portion P × Q low frequency coefficient block
for watermark embedment, rear portion H low frequency coefficient block
sVR for watermark extracting trains;
Step 33: in the important area of G component normalized image, P × Q low frequency coefficient block selected by amendment
b k interior low frequency coefficient, embed to complete watermark information, the embedding method adopted is:
Wherein,
with
be respectively low frequency coefficient block before and after amendment
b k interior low frequency coefficient,
v k for low frequency coefficient block
b k interior low frequency coefficient average, △ is the embedment strength of watermark;
Step 34: according to G component visual masking model, the embedment strength △ of Automatic adjusument watermark, the regulation strategy adopted is:
Wherein,
m f for final HVS model,
pfor weights, value, between 0.4 to 0.5, is used for correcting distortion,
m l for Intensity model,
m t for texture model,
for the important area for G component normalized image,
for edge detecting operation,
the edge treated of expansion,
m e for edge model,
for wave filter;
Step 35: the acquisition of moisture printed color picture, replaces former low frequency coefficient with the low frequency coefficient of amended non-downsampling Contourlet conversion and carries out inverse non-downsampling Contourlet conversion, to obtain the normalization important area containing watermark; In conjunction with the insignificant part in former normalized image region, to obtain containing watermark normalization characteristic region; Carry out pre-distortion compensated to obtain containing watermark G component image; Utilize containing watermark G component image and former R component, B component image, moisture printed color picture can be obtained
.
5. one kind embeds with the Color digital watermarking based on machine learning described in claim 1,2,3 or 4
The detection method that method is corresponding, is characterized in that carrying out as follows:
Step 41: utilize the image normalization technology based on square, treating sense colors image is
be normalized, to obtain corresponding normalized color image;
Step 42: calmodulin binding domain CaM Invariant centroid is theoretical, extracts important area from normalized color image;
Step 43: respectively 3 grades of non-downsampling Contourlet conversions are implemented to the normalization important area of R component and G component, and choose corresponding low frequency region
with
;
Step 44: by low frequency region in the moving window that size is 2 × 2
with
be divided into low frequency coefficient block
with
, utilize the key identical with telescopiny
k 1 , respectively at low frequency region
with
in choose and not overlap
p × Q+Hindividual low frequency coefficient block, wherein, H low frequency coefficient block
with
for SVR training, P × Q low frequency coefficient block
with
for the extraction of digital watermarking;
Step 45: establish low frequency coefficient block
the average of interior low frequency coefficient is
if, low frequency coefficient block
the average of interior low frequency coefficient is
, will
as training characteristics value, and incite somebody to action
as training objective value, so following vector set can be defined:
Carry out training study, SVR training pattern can be obtained;
Step 46: establish low frequency coefficient block
the average of interior low frequency coefficient is
, will
as training characteristics, utilize acquired SVR training pattern to carry out data prediction, corresponding output vector value can be obtained
, by comparing
with output vector value
size extract digital watermarking, extracting method is
Described
for low frequency coefficient block
the average of interior low frequency coefficient;
Step 47: to binary sequence
carry out a liter dimension process, and carry out inverted disorderly deciphering, namely obtain extracted binary image watermarking
.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410793279.3A CN104680473A (en) | 2014-12-20 | 2014-12-20 | Machine learning-based color image watermark embedding and detecting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410793279.3A CN104680473A (en) | 2014-12-20 | 2014-12-20 | Machine learning-based color image watermark embedding and detecting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104680473A true CN104680473A (en) | 2015-06-03 |
Family
ID=53315473
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410793279.3A Pending CN104680473A (en) | 2014-12-20 | 2014-12-20 | Machine learning-based color image watermark embedding and detecting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104680473A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9521291B2 (en) | 2013-07-19 | 2016-12-13 | Digimarc Corporation | Feature-based watermark localization in digital capture systems |
| CN106408497A (en) * | 2016-08-31 | 2017-02-15 | 南京师范大学 | Strong-robustness watermark embedding and extraction method for original remote sensing images |
| CN107194863A (en) * | 2017-04-07 | 2017-09-22 | 广东精点数据科技股份有限公司 | Image digital watermark embedded system and method |
| CN107240059A (en) * | 2017-04-07 | 2017-10-10 | 广东精点数据科技股份有限公司 | The modeling method of image digital watermark embedment strength regressive prediction model |
| US9892301B1 (en) | 2015-03-05 | 2018-02-13 | Digimarc Corporation | Localization of machine-readable indicia in digital capture systems |
| CN107862648A (en) * | 2017-11-24 | 2018-03-30 | 齐鲁工业大学 | Color watermark insertion and extracting method based on human-eye visual characteristic |
| CN108052981A (en) * | 2017-12-22 | 2018-05-18 | 辽宁师范大学 | Image classification method based on non-downsampling Contourlet conversion and convolutional neural networks |
| US10198648B1 (en) | 2015-04-10 | 2019-02-05 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| CN112241929A (en) * | 2019-07-16 | 2021-01-19 | 阿里巴巴集团控股有限公司 | Watermark embedding method, watermark detecting method, watermark embedding device, watermark detecting device and electronic equipment |
| CN112637440A (en) * | 2020-12-17 | 2021-04-09 | 合肥图溯信息技术有限公司 | Picture invisible watermark embedding method and system compatible with scanner equipment |
| CN114255151A (en) * | 2020-09-25 | 2022-03-29 | 浙江工商大学 | High-resolution image robust digital watermarking method based on key point detection and deep learning |
| US11409845B2 (en) | 2019-01-17 | 2022-08-09 | Nxp B.V. | Method for determining if a machine learning model has been copied |
| US11409843B2 (en) | 2019-10-10 | 2022-08-09 | Nxp B.V. | Method for protecting a software program from copying |
| US11500970B2 (en) | 2019-08-02 | 2022-11-15 | Nxp B.V. | Machine learning model and method for determining if the machine learning model has been copied |
| US11586989B2 (en) | 2019-07-15 | 2023-02-21 | Nxp B.V. | Method for detecting if a machine learning model has been copied using intermediate outputs of the machine learning model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930499A (en) * | 2012-11-16 | 2013-02-13 | 崔得龙 | Method for adding and removing digital watermark to and from digital image signal |
| CN104091303A (en) * | 2014-07-11 | 2014-10-08 | 湖南大学 | Robust image hashing method and device based on Radon transformation and invariant features |
-
2014
- 2014-12-20 CN CN201410793279.3A patent/CN104680473A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930499A (en) * | 2012-11-16 | 2013-02-13 | 崔得龙 | Method for adding and removing digital watermark to and from digital image signal |
| CN104091303A (en) * | 2014-07-11 | 2014-10-08 | 湖南大学 | Robust image hashing method and device based on Radon transformation and invariant features |
Non-Patent Citations (2)
| Title |
|---|
| 牛盼盼: "面向彩色图像版权保护的数字水印技术研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 王向阳 等: "基于图像归一化的Contourlet域数字水印方案", 《小型微型计算机系统》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10460413B2 (en) | 2013-07-19 | 2019-10-29 | Digimarc Corporation | Feature-based signal localization in digital capture systems |
| US9521291B2 (en) | 2013-07-19 | 2016-12-13 | Digimarc Corporation | Feature-based watermark localization in digital capture systems |
| US9984429B2 (en) | 2013-07-19 | 2018-05-29 | Digimarc Corporation | Feature-based signal localization in digital capture systems |
| US9892301B1 (en) | 2015-03-05 | 2018-02-13 | Digimarc Corporation | Localization of machine-readable indicia in digital capture systems |
| US11194984B2 (en) | 2015-03-05 | 2021-12-07 | Digimarc Corporation | Localization of machine-readable indicia in digital capture systems |
| US10650209B2 (en) | 2015-03-05 | 2020-05-12 | Digimarc Corporation | Localization of machine-readable indicia in digital capture systems |
| US10198648B1 (en) | 2015-04-10 | 2019-02-05 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| US11244183B2 (en) | 2015-04-10 | 2022-02-08 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| CN106408497A (en) * | 2016-08-31 | 2017-02-15 | 南京师范大学 | Strong-robustness watermark embedding and extraction method for original remote sensing images |
| CN107194863A (en) * | 2017-04-07 | 2017-09-22 | 广东精点数据科技股份有限公司 | Image digital watermark embedded system and method |
| CN107240059A (en) * | 2017-04-07 | 2017-10-10 | 广东精点数据科技股份有限公司 | The modeling method of image digital watermark embedment strength regressive prediction model |
| CN107862648A (en) * | 2017-11-24 | 2018-03-30 | 齐鲁工业大学 | Color watermark insertion and extracting method based on human-eye visual characteristic |
| CN107862648B (en) * | 2017-11-24 | 2021-01-19 | 齐鲁工业大学 | Color watermark embedding and extracting method based on human visual characteristics |
| CN108052981A (en) * | 2017-12-22 | 2018-05-18 | 辽宁师范大学 | Image classification method based on non-downsampling Contourlet conversion and convolutional neural networks |
| CN108052981B (en) * | 2017-12-22 | 2021-08-17 | 辽宁师范大学 | Image classification method based on nonsubsampled Contourlet transformation and convolutional neural network |
| US11409845B2 (en) | 2019-01-17 | 2022-08-09 | Nxp B.V. | Method for determining if a machine learning model has been copied |
| US11586989B2 (en) | 2019-07-15 | 2023-02-21 | Nxp B.V. | Method for detecting if a machine learning model has been copied using intermediate outputs of the machine learning model |
| CN112241929A (en) * | 2019-07-16 | 2021-01-19 | 阿里巴巴集团控股有限公司 | Watermark embedding method, watermark detecting method, watermark embedding device, watermark detecting device and electronic equipment |
| CN112241929B (en) * | 2019-07-16 | 2022-11-11 | 阿里巴巴集团控股有限公司 | Watermark embedding method, watermark detecting method, watermark embedding device, watermark detecting device and electronic equipment |
| US11500970B2 (en) | 2019-08-02 | 2022-11-15 | Nxp B.V. | Machine learning model and method for determining if the machine learning model has been copied |
| US11409843B2 (en) | 2019-10-10 | 2022-08-09 | Nxp B.V. | Method for protecting a software program from copying |
| CN114255151A (en) * | 2020-09-25 | 2022-03-29 | 浙江工商大学 | High-resolution image robust digital watermarking method based on key point detection and deep learning |
| CN114255151B (en) * | 2020-09-25 | 2024-05-14 | 浙江工商大学 | High-resolution image robust digital watermarking method based on key point detection and deep learning |
| CN112637440A (en) * | 2020-12-17 | 2021-04-09 | 合肥图溯信息技术有限公司 | Picture invisible watermark embedding method and system compatible with scanner equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104680473A (en) | Machine learning-based color image watermark embedding and detecting method | |
| Kumar et al. | Image transformation technique using steganography methods using LWT technique | |
| CN104091302A (en) | Robust watermark embedding and extracting method based on multi-scale space | |
| CN102682418B (en) | Method for embedding and extracting multiple zero watermarks of digital image | |
| CN105701758A (en) | Digital watermarking image generation method based on graphic code and apparatus thereof | |
| CN101393635B (en) | Watermark information embedding, extracting method and processing system | |
| CN107578362B (en) | Method for embedding secret information in image based on texture synthesis | |
| CN102024249A (en) | Digital image watermarking method based on visual perception characteristics | |
| CN105512999A (en) | Dual-transform color image holographic watermarking method | |
| CN102930499A (en) | Method for adding and removing digital watermark to and from digital image signal | |
| CN104766263A (en) | Color image watermark embedding and detecting method based on quaternion Legendre moment correction | |
| CN104504643A (en) | Robustness digital water mark embedding and detection method based on local content features | |
| Su | Color image watermarking: algorithms and technologies | |
| CN104484850A (en) | Robust image watermark detecting method based on fuzzy classification | |
| CN116579908B (en) | Method and device for implanting encrypted hidden information into image | |
| CN109544438A (en) | A kind of digital watermark method based on neural network and dct transform | |
| CN111640052B (en) | Robust high-capacity digital watermarking method based on mark code | |
| CN114255151A (en) | High-resolution image robust digital watermarking method based on key point detection and deep learning | |
| Dutta et al. | An efficient and secure digital image watermarking using features from iris image | |
| CN104504644A (en) | Color image watermark detection method based on invariant-moment geometric correction | |
| CN202929680U (en) | Digital watermarking device based on image feature and Huffman coding theory | |
| Yang et al. | Robust digital watermarking based on local invariant radial harmonic fourier moments | |
| Fang et al. | Image Watermarking Resisting to Geometrical Attacks Based on Histogram | |
| KR101725095B1 (en) | Digital Watermark Detecting Method for Image file With Unknown Rotation Angle | |
| Wu | A RST invariant watermarking scheme utilizing support vector machine and image moments for synchronization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150603 |