CN106250836B - Two benches facial image sorting technique and system under a kind of condition of small sample - Google Patents
Two benches facial image sorting technique and system under a kind of condition of small sample Download PDFInfo
- Publication number
- CN106250836B CN106250836B CN201610598570.4A CN201610598570A CN106250836B CN 106250836 B CN106250836 B CN 106250836B CN 201610598570 A CN201610598570 A CN 201610598570A CN 106250836 B CN106250836 B CN 106250836B
- Authority
- CN
- China
- Prior art keywords
- sample
- unmarked
- subset
- label
- new
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 230000001815 facial effect Effects 0.000 title claims abstract description 26
- 239000003550 marker Substances 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims description 45
- 238000013480 data collection Methods 0.000 claims description 8
- 238000004321 preservation Methods 0.000 claims description 5
- 238000013481 data capture Methods 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims description 2
- 238000003909 pattern recognition Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001151 other effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the two benches facial image sorting technique and system under a kind of condition of small sample, wherein method includes:Sample expands the stage, so that the unmarked sample of facial image is carried out collaboration expression to marker samples by semi-supervised mode, obtains collaboration and indicates coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum cooperate with is indicated that the unmarked sample corresponding to coefficient is added in label subset, marker samples are expanded, using the label subset after expansion as training sample;Simultaneously using remaining unmarked sample as new unmarked sample;Facial image sorting phase expands the new unmarked sample that the stage obtains to sample using the label subset after expansion based on collaboration presentation class device and classifies, and obtains final classification results.The present invention improves the accuracy of supervised classification method, while making full use of the judgement information of unmarked sample, and semi-supervised learning problem is converted to supervised learning problem using sample extended mode.
Description
Technical field
The present invention relates to area of pattern recognition, more specifically to the two benches face figure under a kind of condition of small sample
As sorting technique and system.
Background technology
With the rapid development of pattern-recognition and computer vision technique, recognition of face is obtained since it is widely applied
Concern much from every field researcher becomes an importance in modern mode identification technology research.
However, recognition of face is a small sample problem in practical applications, many traditional face identification methods are bases
In a large amount of training sample premises, therefore in the case where extreme lacks marker samples, the recognition methods largely supervised by
Limitation, recognition capability can be weakened.The face identification method mainly used at present includes KNN (k neighbours), LDA (linear discriminants
Analysis), SRC (rarefaction representation grader), CRC (collaboration presentation class device) etc..In theorem in Euclid space, KNN be it is a kind of based on away from
The strategy classified from similarity.It can effectively keep the structural relation between sample local neighbor, and still, it is to noise
It is very sensitive, and depend on Euclidean distance.Label information is utilized as a kind of sorting technique having supervision in LDA, passes through selection one
A projection vector makes similar point after projection as close to inhomogeneous point disperses as far as possible after projection, to realize
Classification feature.But LDA and KNN have the shortcomings that identical, are exactly that they cannot achieve robustness to noise.
In view of the robustness to noise, there are many achievements in research.The thought of SRC is gone using training data
Sparse reconstructs test sample, and classification results are obtained by comparing reconstructed residual.Since Sparse methods are in pattern-recognition
Success, therefore it is used in every field extensively.But it is to solve for l1Problem needs to take a substantial amount of time, therefore can not be fast
Speed obtains classification results.There is the method that researcher proposes CRC to classify recently, and can be with the final classification knot of quick obtaining
Fruit.But the classifying quality of SRC and CRC depends on the quantity of training sample, therefore cannot be satisfied small sample in actual use
Condition.
Relative to the marker samples for being difficult to obtain, a large amount of unmarked sample is simple and easy to get, is not marked largely in allowing for
Remember that the judgement information in sample, semi-supervised method can be used directly to classify, but if directly uses semi-supervised
Classification policy, with the progress of classification, error in classification can add up, and can be caused in this way to final classification results great
It influences.
Although having above-mentioned a variety of face classification methods, in actual classification human face data, classify for small sample
How fully still more difficult using the judgement information of a large amount of unmarked samples and the semantic information of a small amount of marker samples problem is
With realization.Therefore, classify to a large amount of unmarked samples under condition of small sample, be the emphasis studied at present and difficulty
Point.
Invention content
The purpose of the present invention is exactly to solve to provide one in the classification problem of a large amount of unmarked samples under condition of small sample
Two benches facial image sorting technique and system under kind condition of small sample, it, which has, combines the semi-supervised method with supervision,
The advantages of effective facial image classification is realized under conditions of small sample.
To achieve the goals above, the present invention adopts the following technical scheme that:
A kind of two benches facial image sorting technique under condition of small sample, steps are as follows:
Sample expands the stage, so that the unmarked sample of facial image is carried out collaboration table to marker samples by semi-supervised mode
Show, obtains collaboration and indicate coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum is cooperateed with and indicates coefficient
Corresponding unmarked sample be added to label subset in, marker samples are expanded, using the label subset after expansion as
Training sample;Simultaneously using remaining unmarked sample as new unmarked sample, it is saved in new unmarked subset;
Facial image sorting phase expands the stage using the label subset after expanding based on collaboration presentation class device to sample
New unmarked sample in obtained new unmarked subset is classified, and obtains final classification results.
The sample expands the stage, and maximum cooperate with is indicated that the unmarked sample corresponding to coefficient is added to label subset
In, the corresponding label of unmarked sample being added in label subset is obtained, if the label corresponding to unmarked sample is
It is a, then it will be labeled as in multiple unmarked Sample preservation to new unmarked subset.
Sample expands the stage, includes the following steps:
Step (1):Face image data collection is obtained, wherein label subset is Itrain, unmarked subset is Utest, C is sample
This classification number;
Step (2):The label subset of the i-th class is expanded respectively respectively, the value range of i is 1 to C, to the i-th class
J-th of marker samplesUse unmarked subset UtestIt obtains corresponding collaboration and indicates coefficient, obtain maximum collaboration and indicate system
The corresponding unmarked sample of numberAnd maximum collaboration is indicated to the unmarked sample corresponding to coefficientIt is added to mark
Remember subset ItrainIn;Simultaneously by unmarked subset UtestMiddle maximum collaboration indicates the unmarked sample corresponding to coefficientIf
It is set to full 0;
Step (3):Cycle is iterated to step (2), until marking subset ItrainMarker samples quantity reach setting
Terminate when threshold value.
In the step (3):
The tag along sort that maximum collaboration indicates the unmarked sample corresponding to coefficient is obtained in iteration cycle process, to mark
The quantity of label is judged, to determine that maximum collaboration indicates whether the unmarked sample corresponding to coefficient is put into label subset
ItrainIn.
The quantity to label judges that step is:
If unmarked sample has multiple labels or is not labeled, unmarked sample is marked as Inf, is labeled
It is put into unmarked subset for the sample of Inf, obtains new unmarked subset Unew_test;
If unmarked sample tool is there are one label, unmarked sample is added directly into label subset ItrainIn, finally
The label subset I expandednew_train。
The given threshold of the step (3) is unmarked subset UtestUnmarked sample size half.
Facial image sorting phase, steps are as follows:
By the label subset I of expansionnew_trainAs training sample, new unmarked subset Unew_testFor test sample, make
Use Inew_trainAs training sample and using the grader indicated based on collaboration to new unmarked subset Unew_testDivided
Class obtains final classification results.
The grader indicated based on collaboration is as follows:
Wherein, αiIndicate the expression coefficient of the i-th class,Indicate that the collaboration of the i-th class indicates coefficient,
Secondly, identity (y)=argmin is used | ei| reconstructed residual is calculated, is divided according to the size of reconstructed residual
Class;
Wherein,
The collaboration of the step (2) indicates that the calculating step of coefficient is:
Wherein, UtestFor unmarked subset,For j-th of marker samples of the i-th class, αijFor j-th of label of the i-th class
The corresponding collaboration of sample indicates that coefficient, λ are the calculating that regularization parameter is used for controlling coefficient of concordance, and it is to indicate to prevent over-fitting, α
Coefficient.
A kind of two benches facial image categorizing system under condition of small sample, including:
Sample enlargement module, be configured as by semi-supervised mode make the unmarked sample of facial image to marker samples into
Row collaboration indicates, obtains collaboration and indicates coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum is cooperateed with
It indicates that the unmarked sample corresponding to coefficient is added in label subset, marker samples is expanded, the label after expansion
Subset is as training sample;Simultaneously using remaining unmarked sample as new unmarked sample, it is saved in new unmarked son
It concentrates;
Facial image sort module expands the stage using the label subset after expanding based on collaboration presentation class device to sample
New unmarked sample in obtained new unmarked subset is classified, and obtains final classification results.
Maximum cooperate with is indicated that the unmarked sample corresponding to coefficient is added to label subset by the sample enlargement module
In, the corresponding label of unmarked sample being added in label subset is obtained, if the label corresponding to unmarked sample is
It is a, then it will be labeled as in multiple unmarked Sample preservation to new unmarked subset.
Sample enlargement module, including:
Data capture unit:It is configured as obtaining face image data collection, wherein label subset is Itrain, unmarked subset
For Utest, C is sample class number;
Mark subset expansion unit:It is configured to expand the label subset of the i-th class respectively, the value model of i
It is 1 to C to enclose, to j-th of marker samples of the i-th classUse unmarked subset UtestIt obtains corresponding collaboration and indicates coefficient, obtain
Maximum collaboration is taken to indicate the unmarked sample corresponding to coefficientAnd maximum cooperate with is indicated into not marking corresponding to coefficient
Remember sampleIt is added to label subset ItrainIn;Simultaneously by unmarked subset UtestMiddle maximum collaboration indicates corresponding to coefficient
Unmarked sampleIt is set as full 0;
Iteration unit:It is configured as to marking the work of subset expansion unit to be iterated cycle, until marking subset
ItrainMarker samples quantity terminate when reaching given threshold.
The iteration unit:It is configured as obtaining not marking corresponding to maximum collaboration expression coefficient in iteration cycle process
The tag along sort for remembering sample, judges the quantity of label, to determine that maximum collaboration indicates unmarked corresponding to coefficient
Whether sample is put into label subset ItrainIn.
The quantity to label judges, refers to:
If unmarked sample has multiple labels or is not labeled, unmarked sample is marked as Inf, is labeled
It is put into unmarked subset for the sample of Inf, obtains new unmarked subset Unew_test;
If unmarked sample tool is there are one label, unmarked sample is added directly into label subset ItrainIn, finally
The label subset I expandednew_train。
The given threshold of the iteration unit is unmarked subset UtestUnmarked sample size half.
Facial image sort module, is configured as:
By the label subset I of expansionnew_trainAs training sample, new unmarked subset Unew_testFor test sample, make
Use Inew_trainAs training sample and using the grader indicated based on collaboration to new unmarked subset Unew_testDivided
Class obtains final classification results.
The beneficial effects of the invention are as follows:
1. the method for the present invention considers the relationship between collaboration expression neighbour, and takes full advantage of the subspace knot of data
Structure.Make full use of collaboration sparse information and subspace that information is kept to carry out pattern-recognition.
2. in view of the accumulation of error of pure semi-supervised method, sample expansion is carried out using semi-supervised mode.
3. the method using supervision is classified, it can effectively slow down the accumulation of error in this way, it is contemplated that supervise the knowledge of algorithm
Other effect is far more than unsupervised and semi-supervised algorithm, and the final method using supervision can fully enhance the precision of identification.
Description of the drawings
Fig. 1 is overall flow figure of the present invention;
Fig. 2 is the system framework figure of the present invention.
Specific implementation mode
The invention will be further described with embodiment below in conjunction with the accompanying drawings.
As shown in Figure 1, including:
Sets of image data I is obtained, described image data acquisition system includes training image data subset ItrainAnd test image
Data subset Utest;
Assuming that face image data collection I consists of two parts, I=[X, label], wherein X=[x1, x2..., xn], it is each
A sample xi{ i=1 ... n } is the vector of a p × 1 dimensions, and p is sample dimension, and C is number of samples.
Sample data set I can be divided into I on demand againtrainAnd Utest.10 times of cross validations are used in the present invention, by sample
Data set is divided into 10 parts uniformly at random, takes a copy of it as test sample collection I every timetrain, remaining nine parts are then used as Utest,
Experiment can be repeated 10 times.
The sample set I obtained according to step (1)train, first to ItrainIn some sample point xi{ i=1L n }, is obtained
Take all samples of a certain class label.
Then according to step (2) to each marker samples of every one kindUse UtestIt obtains corresponding collaboration and indicates system
Simultaneously the corresponding unmarked sample of maximum coefficient value is added for numberSpecific method is:
Wherein UtestFor unmarked subset,For j-th of marker samples of the i-th class, αijFor j-th of label of the i-th class
The corresponding coefficient of concordance of sample;Obtain the unmarked sample corresponding to the value of maximum coefficient of concordanceThis sample is added
Itrain, and to UtestInIt is configured as full 0;
According to step (3) until ItrainQuantity reach original UtestQuantity half when terminate to be expanded label sample
This collection;
The tag along sort of extension sample is obtained according to step (3), wherein if there are certain sample multiple labels must be labeled as
Otherwise Inf is directly added into Itrain;
Image data in human face data collection (YaleB) is verified, includes 38 people in YaleB face databases,
Share the sample of 2414 known types, pixel 32x32.Experiment uses 10 times of cross validations, and all data are random
Uniformly it is divided into 10 parts, chooses one group every time and be used as test data, as training data, experiment, which is repeated 10 times, is taken 10 times for remaining
For average value as final recognition accuracy, accuracy rate is as shown in table 1.
The facial image sorting phase:The expansion label subset I obtained according to step (3)new_trainAs training sample
This, new unmarked subset Unew_testFor test sample, I is usednew_trainUtilize the grader pair indicated based on collaboration
Unew_testClassify, and obtains final classification results.Obtain the accuracy of identification (%) under different samplings.
As shown in Fig. 2, the two benches facial image categorizing system under a kind of condition of small sample, including:
Sample enlargement module, be configured as by semi-supervised mode make the unmarked sample of facial image to marker samples into
Row collaboration indicates, obtains collaboration and indicates coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum is cooperateed with
It indicates that the unmarked sample corresponding to coefficient is added in label subset, marker samples is expanded, the label after expansion
Subset is as training sample;Simultaneously using remaining unmarked sample as new unmarked sample, it is saved in new unmarked son
It concentrates;
Facial image sort module expands the stage using the label subset after expanding based on collaboration presentation class device to sample
New unmarked sample in obtained new unmarked subset is classified, and obtains final classification results.
Maximum cooperate with is indicated that the unmarked sample corresponding to coefficient is added to label subset by the sample enlargement module
In, the corresponding label of unmarked sample being added in label subset is obtained, if the label corresponding to unmarked sample is
It is a, then it will be labeled as in multiple unmarked Sample preservation to new unmarked subset.
Sample enlargement module, including:
Data capture unit:It is configured as obtaining face image data collection, wherein label subset is Itrain, unmarked subset
For Utest, C is sample class number;
Mark subset expansion unit:It is configured to expand the label subset of the i-th class respectively, the value model of i
It is 1 to C to enclose, to j-th of marker samples of the i-th classUse unmarked subset UtestIt obtains corresponding collaboration and indicates coefficient,
Obtain the unmarked sample corresponding to maximum collaboration expression coefficientAnd maximum is cooperateed with and is indicated corresponding to coefficient not
Marker samplesIt is added to label subset ItrainIn;Simultaneously by unmarked subset UtestMiddle maximum collaboration indicates that coefficient institute is right
The unmarked sample answeredIt is set as full 0;
Iteration unit:It is configured as to marking the work of subset expansion unit to be iterated cycle, until marking subset
ItrainMarker samples quantity terminate when reaching given threshold.
The accuracy of identification of the 1 upper six kinds of sorting techniques of human face data collection YaleB of table compares
For the performance of inspection-classification method, by multiple data sets using the grader of the present invention to using PCA
Low-dimensional sample data after dimensionality reduction is classified, and compared with being denoted as with primitive class, obtains accuracy of identification.Due to the use of semi-supervised
Sample extended mode, a large amount of marker samples can be obtained, in order to contain accumulated error that semi-supervised mode classification generates,
The mode that the present invention is converted into supervision in final step is classified, therefore can obtain preferable classification results.
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (10)
1. the two benches facial image sorting technique under a kind of condition of small sample, characterized in that steps are as follows:
Sample expands the stage, so that the unmarked sample of facial image is carried out collaboration expression to marker samples by semi-supervised mode,
It obtains collaboration and indicates coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum is cooperateed with and indicates coefficient institute
Corresponding unmarked sample is added in label subset, is expanded marker samples, using the label subset after expansion as instruction
Practice sample;Simultaneously using remaining unmarked sample as new unmarked sample, it is saved in new unmarked subset;
Facial image sorting phase obtains the sample expansion stage using the label subset after expansion based on collaboration presentation class device
New unmarked subset in new unmarked sample classify, and obtain final classification results.
2. the method as described in claim 1, characterized in that the sample expands the stage, and maximum is cooperateed with and indicates that coefficient institute is right
The unmarked sample answered is added in label subset, obtains the corresponding label of unmarked sample being added in label subset, such as
Label corresponding to the unmarked sample of fruit is, then will be labeled as multiple unmarked Sample preservation to new unmarked subset
In.
3. the method as described in claim 1, characterized in that sample expands the stage, includes the following steps:
Step (1):Face image data collection is obtained, wherein label subset is Itrain, unmarked subset is Utest, C is sample class
Other number;
Step (2):The label subset of the i-th class is expanded respectively respectively, the value range of i is 1 to C, to the jth of the i-th class
A marker samplesUse unmarked subset UtestIt obtains corresponding collaboration and indicates coefficient, obtain maximum collaboration and indicate coefficient
Corresponding unmarked sampleAnd maximum collaboration is indicated to the unmarked sample corresponding to coefficientIt is added to label
Subset ItrainIn;Simultaneously by unmarked subset UtestMiddle maximum collaboration indicates the unmarked sample corresponding to coefficientSetting
For full 0;
Step (3):Cycle is iterated to step (2), until marking subset ItrainMarker samples quantity reach given threshold
When terminate.
4. method as claimed in claim 3, characterized in that in the step (3):
The tag along sort that maximum collaboration indicates the unmarked sample corresponding to coefficient is obtained in iteration cycle process, to label
Quantity is judged, to determine that maximum collaboration indicates whether the unmarked sample corresponding to coefficient is put into label subset Itrain
In.
5. method as claimed in claim 4, characterized in that the quantity to label judges that step is:
If unmarked sample has multiple labels or is not labeled, unmarked sample is marked as Inf, is marked as
The sample of Inf is put into unmarked subset, obtains new unmarked subset Unew_test;
If unmarked sample tool is there are one label, unmarked sample is added directly into label subset ItrainIn, it finally obtains
The label subset I of expansionnew_train。
6. the method as described in claim 1, characterized in that facial image sorting phase, steps are as follows:
By the label subset I of expansionnew_trainAs training sample, new unmarked subset Unew_testFor test sample, use
Inew_trainAs training sample and using the grader indicated based on collaboration to new unmarked subset Unew_testClassify,
Obtain final classification results.
7. the two benches facial image categorizing system under a kind of condition of small sample, characterized in that including:
Sample enlargement module is configured as making the unmarked sample of facial image assist marker samples by semi-supervised mode
With indicating, obtains and cooperate with expression coefficient;Obtain the unmarked sample corresponding to maximum collaboration expression coefficient;Maximum collaboration is indicated
Unmarked sample corresponding to coefficient is added in label subset, is expanded marker samples, the label subset after expansion
As training sample;Simultaneously using remaining unmarked sample as new unmarked sample, it is saved in new unmarked subset;
Facial image sort module obtains the sample expansion stage using the label subset after expansion based on collaboration presentation class device
New unmarked subset in new unmarked sample classify, and obtain final classification results.
8. system as claimed in claim 7, characterized in that
Maximum cooperate with is indicated that the unmarked sample corresponding to coefficient is added in label subset, obtained by the sample enlargement module
The corresponding label of unmarked sample being added in label subset is taken, if the label corresponding to unmarked sample is,
It will be labeled as in multiple unmarked Sample preservation to new unmarked subset.
9. system as claimed in claim 7, characterized in that sample enlargement module, including:
Data capture unit:It is configured as obtaining face image data collection, wherein label subset is Itrain, unmarked subset is
Utest, C is sample class number;
Mark subset expansion unit:It is configured to expand the label subset of the i-th class respectively, the value range of i is 1
To C, to j-th of marker samples of the i-th classUse unmarked subset UtestIt obtains corresponding collaboration and indicates coefficient, obtain most
Big collaboration indicates the unmarked sample corresponding to coefficientAnd maximum collaboration is indicated to the unmarked sample corresponding to coefficient
ThisIt is added to label subset ItrainIn;Simultaneously by unmarked subset UtestMiddle maximum collaboration indicates corresponding to coefficient not
Marker samplesIt is set as full 0;
Iteration unit:It is configured as to marking the work of subset expansion unit to be iterated cycle, until marking subset Itrain's
Marker samples quantity terminates when reaching given threshold.
10. system as claimed in claim 9, characterized in that
The iteration unit:It is configured as obtaining the unmarked sample corresponding to maximum collaboration expression coefficient in iteration cycle process
This tag along sort, judges the quantity of label, to determine that maximum collaboration indicates the unmarked sample corresponding to coefficient
Whether label subset I is put intotrainIn;
The quantity to label judges, refers to:
If unmarked sample has multiple labels or is not labeled, unmarked sample is marked as Inf, is marked as
The sample of Inf is put into unmarked subset, obtains new unmarked subset Unew_test;
If unmarked sample tool is there are one label, unmarked sample is added directly into label subset ItrainIn, it finally obtains
The label subset I of expansionnew_train。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610598570.4A CN106250836B (en) | 2016-07-26 | 2016-07-26 | Two benches facial image sorting technique and system under a kind of condition of small sample |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610598570.4A CN106250836B (en) | 2016-07-26 | 2016-07-26 | Two benches facial image sorting technique and system under a kind of condition of small sample |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106250836A CN106250836A (en) | 2016-12-21 |
CN106250836B true CN106250836B (en) | 2018-09-14 |
Family
ID=57604775
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610598570.4A Expired - Fee Related CN106250836B (en) | 2016-07-26 | 2016-07-26 | Two benches facial image sorting technique and system under a kind of condition of small sample |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106250836B (en) |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2864920B1 (en) * | 2012-06-21 | 2023-05-10 | Philip Morris Products S.A. | Systems and methods for generating biomarker signatures with integrated bias correction and class prediction |
CN104598917B (en) * | 2014-12-08 | 2018-04-27 | 上海大学 | A kind of support vector machine classifier IP kernel |
CN105046673B (en) * | 2015-07-13 | 2017-11-03 | 哈尔滨工业大学 | High spectrum image and visual image fusion sorting technique based on self study |
CN105303198B (en) * | 2015-11-17 | 2018-08-17 | 福州大学 | A kind of remote sensing image semisupervised classification method learnt from fixed step size |
CN105740908B (en) * | 2016-01-31 | 2017-05-24 | 中国石油大学(华东) | Classifier design method based on kernel space self-explanatory sparse representation |
-
2016
- 2016-07-26 CN CN201610598570.4A patent/CN106250836B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN106250836A (en) | 2016-12-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Sudholt et al. | Phocnet: A deep convolutional neural network for word spotting in handwritten documents | |
Wu et al. | Traffic sign detection based on convolutional neural networks | |
CN110321967B (en) | Image classification improvement method based on convolutional neural network | |
Bodesheim et al. | Local novelty detection in multi-class recognition problems | |
CN112528928B (en) | Commodity identification method based on self-attention depth network | |
CN107943856A (en) | A kind of file classification method and system based on expansion marker samples | |
CN110543906B (en) | Automatic skin recognition method based on Mask R-CNN model | |
CN106156805A (en) | A kind of classifier training method of sample label missing data | |
Masita et al. | Pedestrian detection using R-CNN object detector | |
Farajzadeh et al. | Study on the performance of moments as invariant descriptors for practical face recognition systems | |
CN105930792A (en) | Human action classification method based on video local feature dictionary | |
Wang et al. | S 3 D: Scalable pedestrian detection via score scale surface discrimination | |
Tian et al. | An accurate eye pupil localization approach based on adaptive gradient boosting decision tree | |
Ren et al. | Image set classification using candidate sets selection and improved reverse training | |
Chen et al. | A multi-scale fusion convolutional neural network for face detection | |
Sunitha et al. | Novel content based medical image retrieval based on BoVW classification method | |
Ghayoumi et al. | Local sensitive hashing (LSH) and convolutional neural networks (CNNs) for object recognition | |
CN111488797B (en) | Pedestrian re-identification method | |
Harizi et al. | Deep-learning based end-to-end system for text reading in the wild | |
CN106250836B (en) | Two benches facial image sorting technique and system under a kind of condition of small sample | |
Rachmawati et al. | Integrating semantic features in fruit recognition based on perceptual color and semantic template | |
CN104573727A (en) | Dimension reduction method of handwritten digital image | |
Blandon et al. | An enhanced and interpretable feature representation approach to support shape classification from binary images | |
Ayyalasomayajula et al. | Feature evaluation for handwritten character recognition with regressive and generative Hidden Markov Models | |
Zhao et al. | Sign text detection in street view images using an integrated feature |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180914 |