CN108460405A - A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study - Google Patents
A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study Download PDFInfo
- Publication number
- CN108460405A CN108460405A CN201810104070.XA CN201810104070A CN108460405A CN 108460405 A CN108460405 A CN 108460405A CN 201810104070 A CN201810104070 A CN 201810104070A CN 108460405 A CN108460405 A CN 108460405A
- Authority
- CN
- China
- Prior art keywords
- grader
- classifier
- integrated
- precision
- optimization method
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
-
- 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
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of image latent writings based on deeply study to analyze Ensemble classifier optimization method.Concrete operation step is as follows:(1)Choose several base graders under certain feature;(2)Using Bagging integrated approaches, several strong integrated classifiers are generated;(3)The fixed sub-classifier number generated, and it is identical in subspace number, repeatedly generate different data sets;(4)Deeply learning model DQN is established, to step(3)In data set be trained and filter out the grader set in the constant or better less number of precision;(5)Steganographic data to be discriminated is input in model, the integrated classifier precision after calculation optimization and grader number.The present invention can effectively optimize the precision and sub-classifier number of integrated classifier, be suitable for the case where a large amount of combining classifiers are adjudicated, reach better classification performance.
Description
Technical field
The present invention relates to a kind of image latent writings based on deeply study to analyze Ensemble classifier optimization method.
Background technology
Steganography is that secret information insertion Digital Media is carried out covert communications, and steganalysis is to detect digital matchmaker
The presence of secret hiding data in body.In most of Steganalysis, some extracted from original and hiding media are used
Sensitive features train grader, and " normal " or " containing close " decision is made to suspicious media.
In steganalysis, the application of grader collects ingredient from the binary classifier of early stage to recent multi classifier
Class device.And the performance of integrated classifier is far superior to single grader.In existing sorting technique, Ensemble classifier achieves very
Good classifying quality, therefore, integrated classifier is in steganalysis using more and more extensive.But integrated classifier selects a large amount of bases
Grader carries out comprehensive judgement, and grader set has certain redundancy, meanwhile, with base grader quantity in integrated classifier
Increase, predetermined speed of model can decline, while its memory space needed can also sharply increase.
Invention content
Purpose of the present invention is to the deficiencies for existing Stego-detection method, propose a kind of figure based on deeply study
As steganalysis Ensemble classifier optimization method.
In order to achieve the above objectives, the present invention adopts the following technical scheme that:
A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study, concrete operation step are as follows:
(1)Choose several base graders under certain feature;
(2)Using Bagging integrated approaches, several strong integrated classifiers are generated;
(3)The fixed sub-classifier number generated, and it is identical in subspace number, repeatedly generate different data sets;
(4)Deeply learning model DQN is established, to step(3)In data set be trained and filter out constant in precision
Or preferably in the case of less number grader set;
(5)Steganographic data to be discriminated is input in model, the integrated classifier precision after calculation optimization and grader number.
The step(1)In, if the butt grader that the distinct methods for choosing identical embedded rate generate is as primary data.
The step(2)In, wherein Bagging integrated approaches are also known as self-service aggregation, are a kind of to be distributed according to non-uniform probability
The duplicate sampling from data(It puts back to)Method, on the self-service sample set that each sampling generates, one base grader of training;It is right
Trained listening group is voted, and test sample is assigned in the highest class of gained vote.
The step(3)In, integrated classifier includes many a base graders, while carrying out integrated study training,
Obtain step(4)In required data set.
The step(4)In, DQN models are established, according to depth Q learning algorithms, build DQN models, are filtered out ideal
Sub-classifier set, concrete operation step are:
1)Prepare data set D, i.e. grader set, serial number is added to each row;
2)Grader set C is initialized, the precision a and sub-classifier number n of current data set are calculated, as trained reference number
Value;
3)Aimed at precision A and object classifiers number N are determined by certain experiment, the target as the study of DQN decisions;
4)By DQN networks, the grader for reaching desired value is selected, is added in set C;
5)By adjusting desired value in a certain range, step 4 is repeated), constantly set D is screened, is selected best
As a result;
6)Calculate the precision a ' and grader number n ' of the data set that final choice goes out.
Compared with prior art, the present invention has the advantage that:
Deeply study applied in the screening of steganalysis grader, is passed through the sieve to base grader by the method for the present invention
Choosing, can select effective base grader, and the number of base grader is reduced while improving classification accuracy, can be apparent
The case where improving the performance of steganalysis model, being suitable for the judgement of a large amount of combining classifiers, reaches better classification performance, more
Suitable for practical application scene.
Description of the drawings
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is the method for the present invention structure chart.
Fig. 3 is that deeply learns DQN selection course figures.
Fig. 4 is precision curve graph.
Fig. 5 is sub-classifier number statistical chart.
Specific implementation mode
In order to facilitate the understanding of those skilled in the art, being carried out to the present invention below in conjunction with attached drawing and embodiment further
Description.
As depicted in figs. 1 and 2, the image latent writing analysis Ensemble classifier based on deeply study that the present embodiment proposes is excellent
Change method, mainly includes the following steps that:
(1)Choose several base graders under certain feature;
(2)Using Bagging integrated approaches, several strong integrated classifiers are generated;
(3)The fixed sub-classifier number generated, and it is identical in subspace number, repeatedly generate different data sets;
(4)Deeply learning model DQN is established, to step(3)In data set be trained and filter out constant in precision
Or preferably in the case of less number grader set;
(5)Steganographic data to be discriminated is input in model, the integrated classifier precision after calculation optimization and grader number,
Obtain final result.
Bagging methods are described in detail below in this example:
Concentrated from the initial data that size is n, independently randomly choose a samples of n ' and form self-service data set, and by this
A process independently carries out many times, until generating many independent self-service data sets.Then, each self-service data set is only
It is on the spot used to train one " component classifier ", the judgement of final classification device will be according to these " component classifier " respective judgements
As a result it chooses in a vote.
In this example, with reference to figure 3, the step(4)In, DQN models are established, according to depth Q learning algorithms, build DQN
Model, filters out ideal sub-classifier set, and concrete operation step is:
1)Prepare data set D, i.e. grader set, serial number is added to each row;
2)Grader set C is initialized, the precision a and sub-classifier number n of current data set are calculated, as trained reference number
Value;
3)Aimed at precision A and object classifiers number N are determined by certain experiment, the target as the study of DQN decisions;
4)By DQN networks, the grader for reaching desired value is selected, is added in set C;
5)By adjusting desired value in a certain range, step 4 is repeated), constantly set D is screened, is selected best
As a result;
6)Calculate the precision a ' and grader number n ' of the data set that final choice goes out.
With reference to figure 4, common integrated study is significantly better than by the grader set performance that the above method filters out, is not only existed
It increases in precision, has even more selected the sub-classifier set after optimization, with reference to figure 5, number is obviously more integrated than common
Lack.
In conclusion deeply study is used in the optimization of integrated study by this example, learn using deeply
Outstanding strategy improves integrated precision and saves the space of on-line study to select the integrated classifier after optimization.
Several embodiments of the invention above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
Cannot the limitation to the scope of the claims of the present invention therefore be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the guarantor of the present invention
Protect range.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (5)
1. a kind of image latent writing based on deeply study analyzes Ensemble classifier optimization method, which is characterized in that concrete operations
Steps are as follows:
(1)Choose several base graders under certain feature;
(2)Using Bagging integrated approaches, several strong integrated classifiers are generated;
(3)The fixed sub-classifier number generated, and it is identical in subspace number, repeatedly generate different data sets;
(4)Deeply learning model DQN is established, to step(3)In data set be trained and filter out constant in precision
Or preferably in the case of less number grader set;
(5)Steganographic data to be discriminated is input in model, the integrated classifier precision after calculation optimization and grader number.
2. the image latent writing according to claim 1 based on deeply study analyzes Ensemble classifier optimization method, special
Sign is, the step(1)In, if the butt grader generated under the distinct methods of the identical embedded rate of selection is as initial number
According to.
3. the image latent writing according to claim 1 based on deeply study analyzes Ensemble classifier optimization method, special
Sign is, the step(2)In, wherein Bagging integrated approaches are also known as self-service aggregation, be it is a kind of according to non-uniform probability distribution from
The method of duplicate sampling in data, each to sample on the self-service sample set generated, one base grader of training;To point trained
Class device is voted, and test sample is assigned in the highest class of gained vote.
4. the image latent writing according to claim 1 based on deeply study analyzes Ensemble classifier optimization method, special
Sign is, the step(3)In, integrated classifier is obtained comprising many a base graders while carrying out integrated study training
To step(4)In required data set.
5. the image latent writing according to claim 1 based on deeply study analyzes Ensemble classifier optimization method, special
Sign is, the step(4)In, DQN models are established, according to depth Q learning algorithms, DQN models is built, filters out ideal son
Grader set, concrete operation step are:
1)Prepare data set D, i.e. grader set, serial number is added to each row;
2)Grader set C is initialized, the precision a and sub-classifier number n of current data set are calculated, as trained reference number
Value;
3)Aimed at precision A and object classifiers number N are determined by certain experiment, the target as the study of DQN decisions;
4)By DQN networks, the grader for reaching desired value is selected, is added in set C;
5)By adjusting desired value in a certain range, step 4 is repeated), constantly set D is screened, is selected best
As a result;
6)Calculate the precision a ' and grader number n ' of the data set that final choice goes out.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810104070.XA CN108460405A (en) | 2018-02-02 | 2018-02-02 | A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810104070.XA CN108460405A (en) | 2018-02-02 | 2018-02-02 | A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108460405A true CN108460405A (en) | 2018-08-28 |
Family
ID=63239329
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810104070.XA Pending CN108460405A (en) | 2018-02-02 | 2018-02-02 | A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108460405A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113158043A (en) * | 2021-04-20 | 2021-07-23 | 湖南海龙国际智能科技股份有限公司 | Intelligent tourism resource recommendation system adopting reinforcement learning and integrated learning |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105872555A (en) * | 2016-03-25 | 2016-08-17 | 中国人民武装警察部队工程大学 | Steganalysis algorithm specific to H.264 video motion vector information embedment |
US20170032245A1 (en) * | 2015-07-01 | 2017-02-02 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and Methods for Providing Reinforcement Learning in a Deep Learning System |
-
2018
- 2018-02-02 CN CN201810104070.XA patent/CN108460405A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170032245A1 (en) * | 2015-07-01 | 2017-02-02 | The Board Of Trustees Of The Leland Stanford Junior University | Systems and Methods for Providing Reinforcement Learning in a Deep Learning System |
CN105872555A (en) * | 2016-03-25 | 2016-08-17 | 中国人民武装警察部队工程大学 | Steganalysis algorithm specific to H.264 video motion vector information embedment |
Non-Patent Citations (3)
Title |
---|
CHRISTOS DIMITRAKAKIS等: "Online adaptive policies for ensemble classifiers", 《NEUROCOMPUTING》 * |
IOANNIS PARTALAS等: "Pruning an ensemble of classifiers via reinforcement learning", 《NEUROCOMPUTING》 * |
YANG GAO等: "Learning classifier system ensemble and compact rule set", 《CONNECTION SCIENCE》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113158043A (en) * | 2021-04-20 | 2021-07-23 | 湖南海龙国际智能科技股份有限公司 | Intelligent tourism resource recommendation system adopting reinforcement learning and integrated learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111126386B (en) | Sequence domain adaptation method based on countermeasure learning in scene text recognition | |
CN108229550B (en) | Cloud picture classification method based on multi-granularity cascade forest network | |
CN111507884A (en) | Self-adaptive image steganalysis method and system based on deep convolutional neural network | |
CN109754002A (en) | A kind of steganalysis hybrid integrated method based on deep learning | |
CN108304876A (en) | Disaggregated model training method, device and sorting technique and device | |
CN107393542A (en) | A kind of birds species identification method based on binary channels neutral net | |
CN104537647A (en) | Target detection method and device | |
CN108399431A (en) | Disaggregated model training method and sorting technique | |
CN108399378A (en) | A kind of natural scene image recognition methods based on VGG depth convolutional networks | |
CN107944460A (en) | One kind is applied to class imbalance sorting technique in bioinformatics | |
CN111145145B (en) | Image surface defect detection method based on MobileNet | |
CN109259764B (en) | Method for determining dynamic brain function network threshold | |
CN110288048A (en) | A kind of submarine pipeline methods of risk assessment of SVM directed acyclic graph | |
CN112819063B (en) | Image identification method based on improved Focal loss function | |
CN107273916A (en) | The unknown Information Hiding & Detecting method of steganographic algorithm | |
CN110348448A (en) | A kind of license plate character recognition method based on convolutional neural networks | |
CN105354600A (en) | Automatic classification method for sandstone microsections | |
CN107492084A (en) | Typical packed cell core image combining method based on randomness | |
CN114500396B (en) | MFD chromatographic feature extraction method and system for distinguishing anonymous Torr application flow | |
CN109523514A (en) | To the batch imaging quality assessment method of Inverse Synthetic Aperture Radar ISAR | |
CN106250913A (en) | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis | |
CN108460405A (en) | A kind of image latent writing analysis Ensemble classifier optimization method based on deeply study | |
CN105069480B (en) | Polarization SAR terrain classification method based on gaussian filtering and PSO | |
CN111461135B (en) | Digital image local filtering evidence obtaining method integrated by convolutional neural network | |
CN104463207A (en) | Knowledge self-encoding network and polarization SAR image terrain classification method thereof |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180828 |