CN104778702A - Image stego-detection method on basis of deep learning - Google Patents
Image stego-detection method on basis of deep learning Download PDFInfo
- Publication number
- CN104778702A CN104778702A CN201510178034.4A CN201510178034A CN104778702A CN 104778702 A CN104778702 A CN 104778702A CN 201510178034 A CN201510178034 A CN 201510178034A CN 104778702 A CN104778702 A CN 104778702A
- Authority
- CN
- China
- Prior art keywords
- degree
- image
- depth
- detection method
- training set
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 51
- 238000013135 deep learning Methods 0.000 title abstract 2
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000001914 filtration Methods 0.000 claims abstract description 20
- 238000013527 convolutional neural network Methods 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 description 18
- 239000011159 matrix material Substances 0.000 description 17
- 238000000034 method Methods 0.000 description 16
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000013461 design Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides an image stego-detection method on the basis of deep learning, which comprises the following steps: filtering images which are correspondingly provided with steganography marks or reality marks in a training set by a high-pass filter so as to obtain a training set comprising steganography residue images and reality residue images; carrying out learning on the training set on a deep network model so as to obtain the trained deep network detection model; filtering an image to be detected by the high-pass filter so as to obtain a residue image to be detected; detecting the residue image to be detected on the deep network detection model so as to determine whether the residue image to be detected is a steganography image. The image stego-detection method provided by the invention can realize creation of a blind-detection model by automatic learning and can accurately identify the steganography image.
Description
Technical field
The invention belongs to image processing field, particularly relate to the image latent writing detection method based on degree of depth study.
Background technology
Along with the digitizing of media resource in recent years and developing rapidly and applying of internet, the acquisition of digital picture and the exchange on network and transmission become very easily with general.Thus be also condition of providing convenience based on the Information hiding of digital picture.Steganography is used for secret information being embedded into the apperceive characteristic not changing carrier in normal carrier, thus realizes secret transmissions information., there is a large amount of steganography method in flourish along with Information Hiding Techniques.People can obtain and use multiple steganography instrument to communicate news on internet easily.The abuse of Steganography brings the information security issue become increasingly conspicuous, and brings potential serious harm to country and society.This makes society very urgent to the demand of the Steganalysis of digital picture.Whether the object of digital picture steganalysis (Digital ImageSteganalysis) is by judging containing extra secret information in image to view data analysis, even can estimated information embedded quantity, estimation key, acquisition secret information etc.The image hidden Info can be found by image induct, thus effectively supervision Steganography use, prevent the illegal application of Steganography, the network information security is of great importance.
At present, Steganalysis mainly contains two large classes: for the special method of specific steganography instrument or a certain class embedded technology with not for the universal method of specific embedding grammar.The general verification and measurement ratio of special method is all very high, but practicality is not strong, because can not exhaustive all hidden algorithms in practical application.Meanwhile, new steganographic algorithm is in continuous appearance, and the importance of general steganalysis becomes increasingly conspicuous, and obviously strengthens in the last few years to the research of these class methods.General Steganalysis is also called blind Detecting technology, and the method for classifying modes of its normally feature based, comprises feature extraction and two steps trained by sorter.The Detection accuracy of current general steganalysis method depends primarily on manual features design.In current image induct field, method about characteristic Design is more, typical in document [J.Fridrich and J.Kodovsky, " Rich Models for Steganalysis of Digital Images, " IEEE Trans.on Info.Forensics and Security, vol.7 (3), pp.868-882, 2012] and [V.Holub and J.Fridrich, " Random projections of residuals for digital imagesteganalysis, " IEEE Transactions on Information Forensics and Security, vol.8, no.12, pp.1996 – 2006, 2013.] the design screening of these features, arranging of parameter depends on specific data set very much, and the energy that requires a great deal of time, and the experimental knowledge of people is had high requirements.In practical application, the complicacy of real world images data and diversity, bring more challenges to characteristic Design.
In recent years, along with the progress of degree of depth study, utilize degree of depth learning method automatic learning feature from view data to obtain in the problems such as recognition and classification and pay close attention to widely and apply.Degree of depth study is the class machine learning method being carried out realization character study by the network model of training containing multilayered nonlinear structure, feature extraction and classifying can be fused in a network model by it, by training whole network to carry out Optimized model parameter, realization character study simultaneously and classification.Typical degree of depth learning method is as document [Hinton G E, Salakhutdinov R R. " Reducing the dimensionality ofdata with neural networks; " Science, 2006,313 (5786): 504-507.] and [Krizhevsky A, Sutskever I, Hinton G E. " Imagenet classification with deepconvolutional neural networks, " Advances in neural information processingsystems.2012:1097-1105].
Summary of the invention
The invention provides a kind of image latent writing detection method based on degree of depth study, to realize creating blind check model by automatic learning and the image latent writing detection method of hidden image being distinguished more accurately.
The invention provides a kind of image latent writing detection method based on degree of depth study, comprising:
To in training set, filtering is carried out, to obtain the training set comprising steganography class residual image and true class residual image to the image Hi-pass filter that steganography class should be had to mark or true class marks;
Described training set is learnt on degree of depth network model, to obtain the degree of depth network detection model after training;
Filtering is carried out to the described Hi-pass filter of image to be checked, to obtain residual image to be checked;
Described degree of depth network detection model detects described residual image to be checked, to determine that whether residual image to be checked is for hidden image.
Beneficial effect of the present invention is:
The image latent writing detection method image latent writing of the present invention detection method that the present invention is based on degree of depth study is by carrying out filtering to the image marked in advance and forming training set and then train in deep neural network training set, to obtain for the high image latent writing detection model of the versatility of nonspecific classification characteristics of image, achieve and create blind check model by automatic learning and can distinguish hidden image more accurately.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the image latent writing detection method embodiment one that the present invention is based on degree of depth study;
Fig. 2 is the FB(flow block) of the image latent writing detection method embodiment one that the present invention is based on degree of depth study;
Fig. 3 is the structural drawing of degree of depth convolutional neural networks in the image latent writing detection method embodiment one that the present invention is based on degree of depth study;
Fig. 4 for the present invention is based on degree of depth study image latent writing detection method embodiment one in image involved by each step;
In figure, the image set of corresponding true class mark in Fig. 4 (a)-training set, Fig. 4 (b)-steganography class image one example to be checked, Fig. 4 (c)-image filtering gained residual image one example to be checked.
Embodiment
Fig. 1 is the process flow diagram of the image latent writing detection method embodiment one that the present invention is based on degree of depth study, Fig. 2 is the FB(flow block) of image latent writing detection method embodiment one that the present invention is based on degree of depth study, and Fig. 3 is the structural drawing of degree of depth convolutional neural networks in the image latent writing detection method embodiment one that the present invention is based on degree of depth study; As shown in Figure 1, Figure 2 and Figure 3, the image latent writing detection method that the present invention is based on degree of depth study comprises:
S101, in training set to steganography class being had to mark or the image Hi-pass filter of true class mark carries out filtering, to obtain the training set comprising steganography class residual image and true class residual image; Preferably, described image is gray level image, and the size of described image is 256 × 256;
Preferably, described Hi-pass filter carries out filtering and comprises and carry out convolution operation to filtering core template K image I formula (1) Suo Shi:
Wherein, filtering core conventional in the steganalysis that K comprises for described Hi-pass filter, described Hi-pass filter comprises linear filtering core or nonlinear filtering core; This example only provides wherein a kind of feasible pattern, and K adopts the matrix of one 5 × 5 sizes:
S102, described training set to be learnt on degree of depth network model, to obtain the degree of depth network detection model after training;
Preferably, described degree of depth network model is degree of depth convolutional neural networks CNN; Described degree of depth convolutional neural networks CNN comprises successively with at least two convolutional layers of the mutual cascade of input and output, at least one full articulamentum and an output layer; Wherein, convolutional neural networks is degree of depth learning method representatively, is a kind of neural network with special networks structure, and it comprises weights and share, the structural design thoughts such as partial operation region, can directly from the more effective feature representation of data learning;
Preferably, described CNN network is of five storeys convolutional layer and 2 layers of full articulamentum altogether; In practical application, the selection of convolutional layer and the full articulamentum number of plies can be determined according to particular problems such as training image size, data scales; The input and output of described each convolutional layer are multiple two-dimensional matrix, are called characteristic pattern, and the output of each convolutional layer is as the input of lower one deck; Wherein, the quantity of the characteristic pattern of every layer of convolutional layer output is 16, also can select other numerical value in practical application.
Preferably, described every layer of convolutional layer comprises following three kinds of operations:
First, convolution algorithm is carried out according to formula (3):
Wherein,
represent the i-th input matrix of l layer convolutional layer,
represent the convolution kernel of the i-th input matrix of connection l layer convolutional layer and m × m size of jth convolution output matrix,
represent the biased of the jth convolution output matrix of l layer convolutional layer;
Preferably, ground floor and layer 5 convolutional layer convolution kernel size are 5 × 5, second to the 4th layer of convolutional layer convolution kernel size is 3 × 3; Every layer of convolution step-length is 1, and does not carry out zero padding operation; In practical application, also can select the convolution kernel of other sizes;
Secondly, to the output matrix that described convolution algorithm obtains
in each element activate, here, the activation function of the element in described convolutional layer is Gaussian function f (x):
Finally, to activate after output matrix be averaged pondization operate, obtain the final output of this convolutional layer; Namely, preferred, describedly on degree of depth network model, learn to comprise the pond window corresponding to convolutional layer described in degree of depth convolutional neural networks CNN to training set and perform the operation of average pondization, wherein average pond function is averaged to the element in k × k size area nonoverlapping in matrix; Wherein, in every layer of convolutional layer, pond window size is 3 × 3, and step-length is 2, and pond window size and step-length also can be chosen other values as the case may be and replace.
Described full articulamentum every layer comprises multiple unit; In the present embodiment, every layer of full articulamentum includes 128 unit; In practical application, the every layer unit quantity of full articulamentum can select other values as the case may be.
Preferably, the operation of described full articulamentum comprises and connects adjacent two full articulamentums according to formula (5):
Wherein,
represent the i-th input block of the full articulamentum of l layer,
represent the i-th input block of the full articulamentum of connection l layer and the weight of jth output unit,
represent the biased of the jth output unit of the full articulamentum of l layer; Here, each unit is all connected with all unit of front one deck, and wherein ground floor is connected with last one deck of convolutional layer, and its last one deck is connected with output layer, and the output of every layer is as the input of lower one deck; F (x) is activation function, and preferably, in described full articulamentum, the activation function of element is ReLU function (6):
f(x)＝max(0，x) (6)
Preferably, the quantity of described output layer is one, and the operation of described output layer comprises:
First calculate according to formula (7):
y
i＝Σ
iw
ijx
i+b
j(7)
Wherein, x
irepresent the i-th input block of output layer, w
ijrepresent and connect the i-th input block of output layer and the weight of jth output unit, b
jrepresent the biased of the jth output unit of output layer; Wherein, the output of last one deck of full articulamentum inputs as it;
Then activate described output layer, here, in described output layer, the activation function of element is softmax function (8):
Wherein i ∈ { 1,2}.
Preferably, described described training set to be learnt on degree of depth network model, comprises to obtain the degree of depth network detection model after training:
Described training set is learnt on degree of depth convolutional neural networks CNN, to obtain the degree of depth network detection model after training by minimizing function shown in formula (9) according to back-propagation algorithm:
-log z
i(9)
Wherein, i ∈ { 1,2}.
S103, filtering is carried out to the described Hi-pass filter of image to be checked, to obtain residual image to be checked; Described the process of filtering is carried out to image to be checked and S101 similar, repeat no more.
S104, on described degree of depth network detection model, described residual image to be checked to be detected, to determine that whether residual image to be checked is for hidden image, Fig. 4 for the present invention is based on degree of depth study image latent writing detection method embodiment one in image involved by each step; In figure, the image set of corresponding true class mark in Fig. 4 (a)-training set, Fig. 4 (b)-steganography class image one example to be checked, Fig. 4 (c)-image filtering gained residual image one example to be checked.
The image latent writing detection method embodiment one that the present invention is based on degree of depth study is by carrying out filtering to the image marked in advance and forming training set and then train in deep neural network training set, to obtain for the high image latent writing detection model of the versatility of nonspecific classification characteristics of image, achieve and create blind check model by automatic learning and can distinguish hidden image more accurately.
Fig. 1 is the process flow diagram of the image latent writing detection method embodiment one that the present invention is based on degree of depth study, and as shown in Figure 1, the step included by image latent writing detection method embodiment two that the present invention is based on degree of depth study is all identical with embodiment one, and difference is:
S102, described training set to be learnt on degree of depth network model, to obtain the degree of depth network detection model after training;
Preferably, described degree of depth network model is degree of depth convolutional neural networks CNN; Described degree of depth convolutional neural networks CNN comprises successively with at least two convolutional layers of the mutual cascade of input and output, at least one full articulamentum and an output layer;
Preferably, described CNN network is of five storeys convolutional layer and 2 layers of full articulamentum altogether; In practical application, the selection of convolutional layer and the full articulamentum number of plies can be determined according to particular problems such as training image size, data scales; The input and output of described each convolutional layer are multiple two-dimensional matrix, are called characteristic pattern, and the output of each convolutional layer is as the input of lower one deck; Wherein, the quantity of the characteristic pattern of every layer of convolutional layer output is 16, also can select other numerical value in practical application.
Preferably, described every layer of convolutional layer comprises following three kinds of operations:
First, convolution algorithm is carried out according to formula (3):
Wherein,
represent the i-th input matrix of l layer convolutional layer,
represent the convolution kernel of the i-th input matrix of connection l layer convolutional layer and m × m size of jth convolution output matrix,
represent the biased of the jth convolution output matrix of l layer convolutional layer;
Preferably, ground floor and layer 5 convolutional layer convolution kernel size are 5 × 5, second to the 4th layer of convolutional layer convolution kernel size is 3 × 3; Every layer of convolution step-length is 1, and does not carry out zero padding operation; In practical application, also can select the convolution kernel of other sizes;
Secondly, to the output matrix that described convolution algorithm obtains
in each element activate, here, the activation function of the element in described convolutional layer is Gaussian function f (x):
Finally, in the operation that described convolutional layer comprises, the pond window corresponding to convolutional layer described in degree of depth convolutional neural networks CNN performs the operation of maximal value pondization, also namely adopts the operation of maximal value pondization to the output matrix after activation, obtains the final output of this convolutional layer; Wherein maximal value pond function is to the element maximizing in k × k size area nonoverlapping in matrix; Wherein, in every layer of convolutional layer, pond window size is 3 × 3, and step-length is 2, and pond window size and step-length also can be chosen other values as the case may be and replace.
Described full articulamentum every layer comprises multiple unit; In the present embodiment, every layer of full articulamentum includes 128 unit; In practical application, the every layer unit quantity of full articulamentum can select other values as the case may be.
Preferably, the operation of described full articulamentum comprises and connects adjacent two full articulamentums according to formula (5):
Wherein,
represent the i-th input block of the full articulamentum of l layer,
represent the i-th input block of the full articulamentum of connection l layer and the weight of jth output unit,
represent the biased of the jth output unit of the full articulamentum of l layer; Here, each unit is all connected with all unit of front one deck, and wherein ground floor is connected with last one deck of convolutional layer, and its last one deck is connected with output layer, and the output of every layer is as the input of lower one deck; F (x) is activation function, and preferably, in described full articulamentum, the activation function of element is ReLU function (6):
f(x)＝max(0，x) (6)
Preferably, the quantity of described output layer is one, and the operation of described output layer comprises:
First calculate according to formula (7):
y
i＝Σ
iw
ijx
i+b
j(7)
Wherein, x
irepresent the i-th input block of output layer, w
ijrepresent and connect the i-th input block of output layer and the weight of jth output unit, b
jrepresent the biased of the jth output unit of output layer; Wherein, the output of last one deck of full articulamentum inputs as it;
Then activate described output layer, here, in described output layer, the activation function of element is softmax function (8):
Wherein i ∈ { 1,2}.
Preferably, described described training set to be learnt on degree of depth network model, comprises to obtain the degree of depth network detection model after training:
Described training set is learnt on degree of depth convolutional neural networks CNN, to obtain the degree of depth network detection model after training by minimizing function shown in formula (9) according to back-propagation algorithm:
-log z
i(9)
Wherein, i ∈ { 1,2}.
Last it is noted that above each embodiment is only in order to illustrate technical scheme of the present invention, be not intended to limit; Although with reference to foregoing embodiments to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein some or all of technical characteristic; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the scope of various embodiments of the present invention technical scheme.
Claims (10)
1., based on an image latent writing detection method for degree of depth study, it is characterized in that, comprising:
To in training set, filtering is carried out, to obtain the training set comprising steganography class residual image and true class residual image to the image Hi-pass filter that steganography class should be had to mark or true class marks;
Described training set is learnt on degree of depth network model, to obtain the degree of depth network detection model after training;
Filtering is carried out to the described Hi-pass filter of image to be checked, to obtain residual image to be checked;
Described degree of depth network detection model detects described residual image to be checked, to determine that whether residual image to be checked is for hidden image.
2. the image latent writing detection method based on degree of depth study according to claim 1, it is characterized in that, described degree of depth network model is degree of depth convolutional neural networks CNN.
3. the image latent writing detection method based on degree of depth study according to claim 2, it is characterized in that, described degree of depth convolutional neural networks CNN comprises successively with at least two convolutional layers of the mutual cascade of input and output, at least one full articulamentum and an output layer.
4. the image latent writing detection method based on degree of depth study according to claim 3, it is characterized in that, the activation function of the element in described convolutional layer is Gaussian function.
5. the image latent writing detection method based on degree of depth study according to claim 3, it is characterized in that, in described full articulamentum, the activation function of element is ReLU function (6):
f(x)＝max(0，x) (6)。
6. the image latent writing detection method based on degree of depth study according to claim 3, it is characterized in that, in described output layer, the activation function of element is softmax function (8):
7. the image latent writing detection method based on degree of depth study according to claim 2, is characterized in that, describedly on degree of depth network model, learns to comprise the pond window corresponding to described convolutional layer to training set and performs average pondization and operate.
8. the image latent writing detection method based on degree of depth study according to claim 2, is characterized in that, describedly to learn on degree of depth network model described training set, comprise to obtain the degree of depth network detection model after training:
Described training set is learnt on degree of depth convolutional neural networks, to obtain the degree of depth network detection model after training by minimizing function shown in formula (9) according to back-propagation algorithm:
-log z
i(9)
Wherein, i ∈ { 1,2}.
9. the image latent writing detection method based on degree of depth study according to claim 1, it is characterized in that, described Hi-pass filter comprises linear filtering core and nonlinear filtering core.
10. the image latent writing detection method based on degree of depth study according to claim 1, it is characterized in that, described image is gray level image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510178034.4A CN104778702A (en) | 2015-04-15 | 2015-04-15 | Image stego-detection method on basis of deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510178034.4A CN104778702A (en) | 2015-04-15 | 2015-04-15 | Image stego-detection method on basis of deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104778702A true CN104778702A (en) | 2015-07-15 |
Family
ID=53620148
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510178034.4A Pending CN104778702A (en) | 2015-04-15 | 2015-04-15 | Image stego-detection method on basis of deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104778702A (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105956610A (en) * | 2016-04-22 | 2016-09-21 | 中国人民解放军军事医学科学院卫生装备研究所 | Remote sensing image landform classification method based on multi-layer coding structure |
| WO2016165082A1 (en) * | 2015-04-15 | 2016-10-20 | 中国科学院自动化研究所 | Image stego-detection method based on deep learning |
| CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术（上海）有限公司 | The recognition methods of watermarked image and the system of identification |
| US9521291B2 (en) | 2013-07-19 | 2016-12-13 | Digimarc Corporation | Feature-based watermark localization in digital capture systems |
| CN106339753A (en) * | 2016-08-17 | 2017-01-18 | 中国科学技术大学 | Method for effectively enhancing robustness of convolutional neural network |
| CN106530200A (en) * | 2016-10-23 | 2017-03-22 | 深圳大学 | Deep-learning-model-based steganography image detection method and system |
| CN106846463A (en) * | 2017-01-13 | 2017-06-13 | 清华大学 | Micro-image three-dimensional rebuilding method and system based on deep learning neutral net |
| CN106920206A (en) * | 2017-03-16 | 2017-07-04 | 广州大学 | A kind of steganalysis method based on confrontation neutral net |
| US9892301B1 (en) | 2015-03-05 | 2018-02-13 | Digimarc Corporation | Localization of machine-readable indicia in digital capture systems |
| CN107832832A (en) * | 2017-10-19 | 2018-03-23 | 珠海格力电器股份有限公司 | The pond operation method and device of convolutional neural networks |
| CN108596818A (en) * | 2017-12-22 | 2018-09-28 | 华南理工大学 | A kind of image latent writing analysis method based on multi-task learning convolutional neural networks |
| CN108664909A (en) * | 2018-04-28 | 2018-10-16 | 上海爱优威软件开发有限公司 | A kind of auth method and terminal |
| US10198648B1 (en) | 2015-04-10 | 2019-02-05 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| CN109348211A (en) * | 2018-08-06 | 2019-02-15 | 中国科学院声学研究所 | The general information of interframe encode hides detection method in a kind of video frame |
| CN109729070A (en) * | 2018-11-28 | 2019-05-07 | 甘肃农业大学 | A kind of detection method of the concurrent stego-channel of network isomery based on CNN and RNN Fusion Model |
| CN109993048A (en) * | 2018-11-26 | 2019-07-09 | 普达迪泰（天津）智能装备科技有限公司 | A kind of bituminous pavement striped filter method and detection system |
| CN110503157A (en) * | 2019-08-27 | 2019-11-26 | 安徽大学 | The image latent writing analysis method of multitask convolutional neural networks based on fine granularity image |
| CN111131658A (en) * | 2020-01-19 | 2020-05-08 | 中国科学技术大学 | Image steganography method, device, electronic equipment and medium |
| CN113011898A (en) * | 2019-12-20 | 2021-06-22 | 宝洁公司 | Machine learning based imaging method to determine authenticity of a consumer product |
| CN113642535A (en) * | 2021-10-13 | 2021-11-12 | 聊城高新生物技术有限公司 | Biological branch detection method and device and electronic equipment |
-
2015
- 2015-04-15 CN CN201510178034.4A patent/CN104778702A/en active Pending
Non-Patent Citations (1)
| Title |
|---|
| Y QIAN等: "《Deep learning for steganalysis via convolutional neural networks》", 《PROCEEDINGS OF THE SPIE》 * |
Cited By (33)
| 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 |
| US10460413B2 (en) | 2013-07-19 | 2019-10-29 | Digimarc Corporation | Feature-based signal 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 |
| US11244183B2 (en) | 2015-04-10 | 2022-02-08 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| US10198648B1 (en) | 2015-04-10 | 2019-02-05 | Digimarc Corporation | Decoding 1D-barcodes in digital capture systems |
| US10223780B2 (en) | 2015-04-15 | 2019-03-05 | Institute Of Automation Chinese Academy Of Sciences | Image steganalysis based on deep learning |
| WO2016165082A1 (en) * | 2015-04-15 | 2016-10-20 | 中国科学院自动化研究所 | Image stego-detection method based on deep learning |
| CN105956610B (en) * | 2016-04-22 | 2019-02-22 | 中国人民解放军军事医学科学院卫生装备研究所 | A kind of remote sensing images classification of landform method based on multi-layer coding structure |
| CN105956610A (en) * | 2016-04-22 | 2016-09-21 | 中国人民解放军军事医学科学院卫生装备研究所 | Remote sensing image landform classification method based on multi-layer coding structure |
| CN106339753A (en) * | 2016-08-17 | 2017-01-18 | 中国科学技术大学 | Method for effectively enhancing robustness of convolutional neural network |
| CN106096668A (en) * | 2016-08-18 | 2016-11-09 | 携程计算机技术（上海）有限公司 | The recognition methods of watermarked image and the system of identification |
| CN106096668B (en) * | 2016-08-18 | 2019-06-18 | 携程计算机技术（上海）有限公司 | The recognition methods and identifying system of watermarked image |
| CN106530200B (en) * | 2016-10-23 | 2020-01-07 | 深圳大学 | Steganographic image detection method and system based on deep learning model |
| CN106530200A (en) * | 2016-10-23 | 2017-03-22 | 深圳大学 | Deep-learning-model-based steganography image detection method and system |
| CN106846463A (en) * | 2017-01-13 | 2017-06-13 | 清华大学 | Micro-image three-dimensional rebuilding method and system based on deep learning neutral net |
| CN106846463B (en) * | 2017-01-13 | 2020-02-18 | 清华大学 | Microscopic image three-dimensional reconstruction method and system based on deep learning neural network |
| CN106920206A (en) * | 2017-03-16 | 2017-07-04 | 广州大学 | A kind of steganalysis method based on confrontation neutral net |
| CN106920206B (en) * | 2017-03-16 | 2020-04-14 | 广州大学 | Steganalysis method based on antagonistic neural network |
| CN107832832A (en) * | 2017-10-19 | 2018-03-23 | 珠海格力电器股份有限公司 | The pond operation method and device of convolutional neural networks |
| CN108596818B (en) * | 2017-12-22 | 2021-10-26 | 华南理工大学 | Image steganalysis method based on multitask learning convolutional neural network |
| CN108596818A (en) * | 2017-12-22 | 2018-09-28 | 华南理工大学 | A kind of image latent writing analysis method based on multi-task learning convolutional neural networks |
| CN108664909A (en) * | 2018-04-28 | 2018-10-16 | 上海爱优威软件开发有限公司 | A kind of auth method and terminal |
| CN109348211A (en) * | 2018-08-06 | 2019-02-15 | 中国科学院声学研究所 | The general information of interframe encode hides detection method in a kind of video frame |
| CN109993048A (en) * | 2018-11-26 | 2019-07-09 | 普达迪泰（天津）智能装备科技有限公司 | A kind of bituminous pavement striped filter method and detection system |
| CN109729070A (en) * | 2018-11-28 | 2019-05-07 | 甘肃农业大学 | A kind of detection method of the concurrent stego-channel of network isomery based on CNN and RNN Fusion Model |
| CN110503157A (en) * | 2019-08-27 | 2019-11-26 | 安徽大学 | The image latent writing analysis method of multitask convolutional neural networks based on fine granularity image |
| CN110503157B (en) * | 2019-08-27 | 2021-11-19 | 安徽大学 | Image steganalysis method of multitask convolution neural network based on fine-grained image |
| CN113011898A (en) * | 2019-12-20 | 2021-06-22 | 宝洁公司 | Machine learning based imaging method to determine authenticity of a consumer product |
| CN111131658A (en) * | 2020-01-19 | 2020-05-08 | 中国科学技术大学 | Image steganography method, device, electronic equipment and medium |
| CN113642535A (en) * | 2021-10-13 | 2021-11-12 | 聊城高新生物技术有限公司 | Biological branch detection method and device and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104778702A (en) | Image stego-detection method on basis of deep learning | |
| US10223780B2 (en) | Image steganalysis based on deep learning | |
| CN104573669B (en) | Image object detection method | |
| CN105046277B (en) | Robust mechanism study method of the feature significance in image quality evaluation | |
| CN105657402B (en) | A kind of depth map restoration methods | |
| CN108399406A (en) | The method and system of Weakly supervised conspicuousness object detection based on deep learning | |
| CN103824309B (en) | Automatic extracting method of urban built-up area border | |
| CN106920243A (en) | The ceramic material part method for sequence image segmentation of improved full convolutional neural networks | |
| CN108961137A (en) | A kind of image latent writing analysis method and system based on convolutional neural networks | |
| CN110796026A (en) | Pedestrian re-identification method based on global feature stitching | |
| CN110443367A (en) | A kind of method of strength neural network model robust performance | |
| CN108734210A (en) | A kind of method for checking object based on cross-module state multi-scale feature fusion | |
| CN108052980A (en) | Air quality grade detection method based on image | |
| CN107480723B (en) | Texture Recognition based on partial binary threshold learning network | |
| CN105205449A (en) | Sign language recognition method based on deep learning | |
| CN107423747A (en) | A kind of conspicuousness object detection method based on depth convolutional network | |
| CN113205520B (en) | Method and system for semantic segmentation of image | |
| CN106920206A (en) | A kind of steganalysis method based on confrontation neutral net | |
| CN106339753A (en) | Method for effectively enhancing robustness of convolutional neural network | |
| CN111257341A (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN109859091A (en) | Image steganography detection method based on Gabor filtering and convolutional neural network | |
| CN105469063A (en) | Robust human face image principal component feature extraction method and identification apparatus | |
| CN114841319A (en) | Multispectral image change detection method based on multi-scale self-adaptive convolution kernel | |
| CN106203373A (en) | A kind of human face in-vivo detection method based on deep vision word bag model | |
| CN105046286A (en) | Supervision multi-view feature selection method based on automatic generation of view and unit with l1 and l2 norm minimization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150715 |
|
| WD01 | Invention patent application deemed withdrawn after publication |