CN104463137B - The abnormal face image detecting method and system of feature based space decomposition - Google Patents
The abnormal face image detecting method and system of feature based space decomposition Download PDFInfo
- Publication number
- CN104463137B CN104463137B CN201410802645.7A CN201410802645A CN104463137B CN 104463137 B CN104463137 B CN 104463137B CN 201410802645 A CN201410802645 A CN 201410802645A CN 104463137 B CN104463137 B CN 104463137B
- Authority
- CN
- China
- Prior art keywords
- feature space
- data
- feature
- unit
- candidate samples
- 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
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
- G06V40/161—Detection; Localisation; Normalisation
Abstract
A kind of abnormal face of feature based space decomposition is as detection method and system, by the way that the data in a face database are carried out with the feature space model that principal component analysis obtains including principal direction, then leaving-one method is used, it regard each image of data set as candidate samples successively, it is deleted from data set, feature space model is updated using feature space division；Multiple principal directions to new and old feature space carry out the abnormality detection based on angle, and the diagonal entry of spin matrix calculates Outlier factor when dividing eventually through feature space, and detects abnormal face image.
Description
Technical field
The present invention relates to a kind of technology of image processing field, specifically one kind can improve prior art in higher-dimension
The detection side of detection performance and computational efficiency, feature based space decomposition abnormal face image on number and large scale data
Method and system.
Background technology
Lifted with the progress in terms of information technology and imaging sensor, and in computing power, increasing people
Face image or face video appear in people life in, such as gate control system based on recognition of face, based on face video with
Monitoring system of track etc..This kind of system is when implementing, in order to obtain satisfied performance, it usually needs the very big face of analysis one
Image library, trains and extracts faceform；One accurate faceform just can guarantee that the recognition accuracy or tracking of system
Effect.However, in actual life, in the building process of training facial image database, change, biography due to image imaging circumstances
Sensor noise and human factor etc., facial image database are often mixed into some containing excessive noise or not conforming to of seriously being blocked
The image of lattice, even inhuman face image.Many faceforms are very sensitive to abnormal data, that is to say, that a small amount of is different
Ordinary person's face image may have a strong impact on the degree of accuracy of final faceform.Generally, these images can quilt when making database
It is artificial to delete.However, with the development of big data and social networks, the scale of facial image database is increasing, such operation
Mode needs to expend substantial amounts of human cost.Thus, in the urgent need to it is a kind of can machine realize Outlier Detection Algorithm, it can send out
Now it is mixed in those abnormal images in a facial image database.This technical problem that exactly this patent is solved.
Abnormal data is defined as that " observation deviates undue remote of other observations, so that allowing people to suspect it by Hawkins
Generated by other mechanism ".The characteristics of codes or data typically exhibits out high dimension and large scale (large sample amount) during big data.Pass
The Outlier Detection Algorithm of system usually requires to assume that data obey some specific distribution, or needs to generate each data sample
This k- arest neighbors, thus so-called " dimension crisis " problem can be run into.In high-dimensional environment, the observation dimension d of data is usual
The number n of sample observed by being far longer than.Thus, when dimension is larger, less data sample is in data space
Seem very sparse, so that range information becomes unreliable, the concept of neighborhood also becomes nonsensical.Lacking prior information
In the case of, it is very difficult that its distributed model is found from data.In addition, in higher-dimension or large scale problem, generation
The calculating of the k- arest neighbors of each data is quite time-consuming.Abnormality detection in facial image database be a typical higher-dimension and
The dimension of large scale problem, i.e. data is very high, sample it is in a large number.Thus, design a kind of accurate, quick, exception of robust
Facial image detection algorithm is still the task and urgent need to solve the problem of a challenge.
By the retrieval discovery to prior art, Barnett [Barnett V.The ordering of
multivariate data(with discussion)[J].Journal of the royalstatistical
society,1976,Series.A(139):318-354] a kind of T based on Hotelling is proposed2The statistics of statistic is different
Normal detection method, it has main steps that：1) think that data integrally obey a multivariate normal distributions, then using training data
Calculate its average and variance matrix；2) the abnormal score of candidate samples is calculated using mahalanobis distance, it obeys Hotelling T2Point
Cloth, if score is excessive, the candidate samples are an exceptional samples.The major defect of this method is, if the dimension of data is higher,
Its applicability is limited, because it is difficult to it is determined that rational data distribution.Breunig etc. proposes a kind of exception inspection based on density
Survey method, i.e. local outlier factor LOF [Breunig M, Kriegel H P, Ng R T, et al.LOF:Identifying
density‐based local outliers.Proceeding of the 2000ACM SIGMOD International
Conference on Management ofData ACM,2000:93-104], it according to the local density of each sample come really
Its fixed abnormality degree, thus can excavate and be hidden in local anomaly in data global structure, having main steps that for it is first calculated
The k- neighborhoods of each sample, then local density in neighborhood is calculated, if the local density of a point is too big, then that is exactly one
Individual abnormity point.LOF major defect is that computational efficiency is relatively low on higher-dimension or large scale data, since it is desired that calculating each
The k- neighborhoods of sample.Kriegel et al. proposes a kind of method for detecting abnormality based on angle, referred to as angle Outlier factor
fastABOF[Kriegel H P,hubert M S,Zimek A.Angle‐based outlier detection in
high‐dimensionaldata.Proceeding of the 14th ACM SIGKDD international
conference on Knowledgediscovery and data mining ACM,2008:444-452], its main step
Suddenly it is the k- neighborhoods for first calculating each sample, then calculates the angle that any two points and the point are constituted in neighborhood, then calculates institute
The variance of angled cosine value, if the corresponding variance very little of a sample, then it is an exceptional sample.fastABOD
Main determination be that its computational efficiency in large scale data is relatively low because it needs to calculate the k- neighborhoods of each sample.
Chinese patent literature CN103365969A discloses (bulletin) day 2013.10.23, discloses a kind of abnormal data inspection
The method and system of processing is surveyed, it has main steps that：The specific data in the scheduled time are gathered, then are extracted in specific data
Characteristic information, determines the data space of data, if the data of a last sample have exceeded corresponding data space, then it
For abnormal data.The major defect of this method is, when the dimension of data is very high, according only to different to judge per one-dimensional value region
The Detection results of perseverance are limited, and the selection per one-dimensional characteristic area in addition relies on experience.
The content of the invention
It is not enough on the higher-dimension and large scale problem that the present invention is detected for existing method in abnormal face, propose that one kind is based on
The abnormal face image detecting method of feature space division, can detect to contain from a given facial image database and excessively make an uproar
Sound and it is verified and blocks abnormal face image, or inhuman face image.In higher-dimension large scale data set as facial image database
On, it has higher sensitiveness to an exceptional sample, so as to improve detection performance and the calculating of abnormal face image
Efficiency.The basic thought of the present invention is that, to the sensitiveness of abnormal data, under leaving-one method mode, will delete using principal component analysis
The change size of principal direction caused by one candidate samples as abnormality detection criterion.In order to accurately calculate a candidate
The knots modification of principal direction caused by face image, employs the principal component point that feature space division updates the data collection with quick decrement
Analysis, and its result is optimal in theory.It has also been found that the spin matrix of the principal direction in feature space splitting algorithm
Diagonal entry be equal to principal direction angle knots modification cosine similarity.Thus, the present invention is eventually through observation feature space
The diagonal entry value of spin matrix detects to be hidden in the abnormal face image in face database during division.
The abnormal face image detecting method of feature based space decomposition proposed by the present invention comprises the following steps：
Step one：By the two-dimensional data matrix of every piece image in facial image database and the row of title one, a vector is obtained；
Then facial image database can be designated as a data set X={ x containing n observation sample1,x2,…,xn, wherein xi∈Rd；
Step 2：Concentrate total data to calculate principal component analysis data, obtain its feature space model；
Step 3：Using leaving-one method, each sample is selected successively as candidate samples, then by it from data set from deleting
Remove；
Step 4：The feature space model of collection is updated the data using feature space splitting algorithm, spin moment therein is recorded
Battle array；
Step 5：The diagonal entry of spin matrix is extracted, the detection method based on angle is then used, judges the candidate
Whether sample is abnormal face image.
Feature space model in described step two refers to：For a data set X=containing n observation sample
{x1,x2,…,xn, wherein xi∈Rd, X feature space model is the feature decomposition of its covariance matrix, be denoted as Ω (X)=
(N (X), μ (X), Λ (X), U (X)), wherein：N (X) is the number of observation sample in X, and μ (X) is the average of sample, and Λ (X) is X
First k maximum characteristic value { λ of Eigenvalues Decomposition1, λ2..., λkThe diagonal matrix that is constituted, U (X) is these characteristic values correspondence
Characteristic vector, i.e. principal direction.
Feature space division in described step four refers to：Exceptional sample is detected using leaving-one method, every time by one
Candidate samples xtDeleted from complete data set, then carry out abnormal inspection by observing the change of principal direction in feature space
Survey, remaining data after candidate samples are deleted integrates as Z, its corresponding feature space model is Ω (Z)=(N (Z), μ (Z), Λ
(Z), U (Z)), in the case of higher-dimension large scale, Ω (Z) amount of calculation is calculated using the feature decomposition of Z covariance matrix is
Than larger, accordingly present invention employs a kind of update method of decrement, Ω (Z) is realized by carrying out small modifications to Ω (X)
Quick and precisely calculating：Note candidate samples integrate as Y={ xt, it is clear that it is not required to calculate and just obtains its feature space model for Ω
(Y)=(1, xt,0,0)；So, Ω (Z) can be regarded as dividing the later results of Ω (Y) from Ω (X), thus can be with table
It is shown as
Described division processing, specifically includes following steps：
N (Z)=n-1
U (Z)=U (X) R
Wherein：G=U (X)T(Xt- μ (X)), EVD represents Eigenvalues Decomposition.
The abnormal face image detection based on angle in described step five refers to：By set up feature space division and
Natural link between abnormality detection based on angle, only by checking that the diagonal entry of spin matrix just can be with candidate samples
Whether it is abnormal data, amount of calculation is very small：When the feature space model Ω (X) of original data set principal direction is U (X)={ u1,
u2,…,uk, delete candidate samples xtThe feature space model Ω (Z) of remaining data principal direction is U (Z)={ u ' afterwards1,u
′2,…,u′k}.Thus, the cosine similarity of corresponding principal direction isAccording to spy
Levy space decomposition algorithm, U (Z) be U (X) as obtained by rotation transformation R, u ' m=U (X) rm, wherein rmIt is R m row.So,
um Tu′m=Rmm, wherein RmmIt is spin matrix R m-th of diagonal entry.That is, spin matrix R diagonal entry
It is exactly the cosine similarity that feature space model principal direction changes.Finally, candidate samples xtOutlier factor be：If StIt is larger, illustrate candidate samples xtCause the change of larger principal direction, then xtJust
It is an exceptional sample.
Technique effect
Compared with prior art, technique effect of the invention is mainly：First, divided by feature space, the present invention is not
The k- neighborhoods of each sample of calculating are needed, thus computational efficiency is very high, especially on higher-dimension and large scale data set；The
Two, based on the angle method for detecting abnormality of many principal direction strategies, consider abnormal using angle information in the feature space of low-dimensional
Property so that detection performance of the present invention on high dimensional data is substantially due to prior art.
Brief description of the drawings
Fig. 1 is schematic diagram of the present invention.
Fig. 2 is many principal component strategy schematic diagrames.
Fig. 3 is embodiment particular flow sheet.
Embodiment
Embodiments of the invention are elaborated below, the present embodiment is carried out lower premised on technical solution of the present invention
Implement, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to following implementations
Example.
Embodiment 1
As shown in figure 1, for the implementation method of the present embodiment, the present embodiment comprises the following steps：
Step one：By the two-dimensional data matrix of every piece image in facial image database and the row of title one, a vector is obtained；
Then facial image database can be designated as a data set X={ x containing n observation sample1,x2,…,xn, wherein xi∈Rd；
Step 2：Concentrate total data to calculate principal component analysis data, obtain its feature space model；
Step 3：Using leaving-one method, each sample is selected successively as candidate samples, then by it from data set from deleting
Remove；
Step 4：The feature space model of collection is updated the data using feature space splitting algorithm, spin moment therein is recorded
Battle array；
Step 5：The diagonal entry of spin matrix is extracted, the detection method based on angle is then used, judges the candidate
Whether sample is abnormal face image.
As shown in figure 3, in order to detect data-oriented collection X={ x1,x2,…,xnIn exceptional data point, the present embodiment
Specific implementation step is as follows：
1) by the Eigenvalues Decomposition for the covariance matrix for calculating X obtain its feature space Ω (X)=(N (X), μ (X),
Λ(X),U(X))。
2) to each sample x in Xt, t=1,2 ..., n.Proper subspace is updated using feature space division decrement
3) average updates μ (Z)=(n μ (X)-xt)/(n-1)；
4) the projection g=U (X) of average residual error is calculatedT(xt-μ(X))；
5) covariance matrix on feature space is calculated
6) calculateEigenvalues Decomposition R Λ (X) RT。
7) many principal direction strategies are used, the angle changed by principal direction calculates Outlier factor：
Described multiple principal directions, i.e., many principal direction strategies refer to：When sample number is very big in data set, in order to improve calculation
Detection performance of the method on big-sample data collection, the knots modification of first principal direction may be not enough to observe the presence of abnormity point.
Multiple principal direction strategies consider the change that the preceding l principal direction caused by a candidate samples is deleted under leaving-one method pattern
Amount.The tactful basic thought is that secondary leading principal direction is more sensitive to abnormity point than first principal direction.
As shown in Fig. 2 being used for the Visual Explanation of abnormality detection for many principal direction strategies.In Fig. 2 (a) left side subgraph,
Three solid lines represent first three principal direction of the feature space model of raw data set respectively.It is right in Fig. 2 (a) the right subgraph
Side round dot represents a normal sample, now deletes it from data set, the small change of the direction generation of principal component, from original
Carry out principal direction and new solid line position has been rotated to from dotted line position.In Fig. 2 (b) right side subgraph, round dot represents an abnormal sample
This, it deviate from the region of normal sample.Now it is deleted from data set, then the principal direction of feature space model can occur
Larger change, wherein：The anglec of rotation of first principal direction is less than the anglec of rotation of time principal direction.
8) cosine similarity of new and old feature space correspondence principal direction is equal to spin matrix R diagonal entry Rmm；
9) sample x is calculatedtOutlier factorIf StMore than threshold value Td, remember xtFor abnormal sample
This, is otherwise designated as normal sample.
The present embodiment is related to a kind of system for realizing the above method, including：Principal component analysis module, feature space split-mode
Block and angle anomaly data detection module, wherein：Principal component analysis module calculates the feature space model of full dataset and defeated
Go out to feature space division module and carry out model modification, angle anomaly data detection module is according to from principal component analysis module
Feature space and from feature space divide module renewal after feature space model carry out angle detection and output abnormality data
Testing result.
Described feature space division module includes：Average updating block, average Residual projection unit, feature space association side
Poor matrix unit and Eigenvalues Decomposition unit, wherein：Average updating block is connected with average Residual projection unit and transmits deletion
Residual error caused by candidate samples, average Residual projection unit is connected with feature space covariance matrix unit and transmits covariance
Information, feature space covariance matrix unit is connected with Eigenvalues Decomposition unit and transmits covariance matrix, Eigenvalues Decomposition list
Member is connected with angle anomaly data detection module and transmits spin matrix information.
Described angle anomaly data detection module includes：Diagonal entry extraction unit, Outlier factor computing unit and
Threshold value comparing unit, wherein：Diagonal entry extraction unit is connected with Outlier factor computing unit and transmits new and old principal direction
Cosine similarity information, Outlier factor computing unit is connected and transmission abnormality factor information with threshold value comparison model, and threshold value compares
Unit compares Outlier factor and threshold value, and output testing result, the i.e. candidate samples are normal data or abnormal data.
Anomaly data detection algorithm (the ES- for the feature based space decomposition that table 1, table 2 and table 3 are proposed by the present embodiment
OD) the testing result in two groups of conventional real human face image libraries.Table 1 is the specifying information of used test data.For
The detection performance of this algorithm is embodied, experiment is carried out to ES-OD and traditional two kinds of Outlier Detection Algorithms LOF and FastABOD
Compare.For the comprehensive performance for weighing several Outlier Detection Algorithms, evaluation index, knot are used as using AUC scores and calculating time
Fruit is respectively in table 2 and table 3.As can be seen that either detection performance is crossed or computational efficiency, the ES-OD of the present embodiment will
Higher than two kinds conventional methods.
The test data of table 1 is configured
| Dimension | Sample number | |
| FERET | 644 | 720 |
| AR | 644 | 952 |
The ES-OD of table 2 LOF, FastABOD and the present embodiment abnormality detection AUC scores
| LOF | FastABOD | ES‐OD | |
| FERET | 0.9398 | 0.8861 | 0.9248 |
| AR | 0.9715 | 0.9679 | 0.9869 |
The ES-OD of table 3 LOF, FastABOD and the present embodiment abnormality detection calculates the time
| LOF | FastABOD | ES‐OD | |
| FERET | 7.4672 | 15.812 | 1.7460 |
| AR | 15.432 | 25.959 | 2.2726 |
Claims (4)
1. a kind of abnormal face image detecting method of feature based space decomposition, it is characterised in that
Step 1：By the way that the data in a face database are carried out with the feature space that principal component analysis obtains including principal direction
Model；
Step 2：Using leaving-one method, each image of data set is deleted it as candidate samples from data set successively,
Feature space model is updated using feature space division；
Step 3：Multiple principal directions to new and old feature space carry out the abnormality detection based on angle, whether judge the candidate samples
For abnormal face image；
Described feature space division refers to：Every time by a candidate samples xtDeleted from complete data set, note Y={ xt}
For candidate samples collection, its feature space model is Ω (Y)=(1, xt, 0,0), remaining data collection after candidate samples are deleted is
Z, its corresponding feature space model is Ω (Z)=(N (Z), μ (Z), Λ (Z), U (Z)), is further represented asSpecifically include：N (Z)=n-1；
U (Z)=U (X) R；Wherein：G=U (X)T(xt- μ (X)), EVD represents Eigenvalues Decomposition；
The described abnormality detection based on angle refers to：By setting up between feature space division and abnormality detection based on angle
Natural link, only by checking whether the diagonal entry of spin matrix just can be abnormal data with candidate samples, specific step
Suddenly include：
1) when the feature space model Ω (X) of original data set principal direction is U (X)={ u1,u2,…,uk, delete candidate samples xt
The feature space model Ω (Z) of remaining data principal direction is U (Z)={ u ' afterwards1,u′2,…,u′k}；Corresponding principal direction it is remaining
String similarity is
2) according to feature space splitting algorithm, U (Z) be U (X) as obtained by spin matrix R, u 'm=U (X) rm, wherein rmIt is R
M is arranged, so um Tu′m=Rmm, wherein RmmIt is spin matrix R m-th of diagonal entry, i.e. spin matrix R diagonal line element
Element is exactly the cosine similarity that feature space model principal direction changes；
3) candidate samples xtOutlier factor be：Work as StMore than threshold value Td, then xtFor exceptional sample；
Feature space model described in step 2 refers to：For a data set X={ x containing n observation sample1,x2,…,
xn, wherein xi∈Rd, X feature space model is the feature decomposition of its covariance matrix, is denoted as Ω (X)=(N (X), μ
(X), Λ (X), U (X)), wherein：N (X) is the number of observation sample in X, and μ (X) is the average of sample, and Λ (X) is X characteristic values
The first k maximum characteristic value { λ decomposed1, λ2..., λkThe diagonal matrix that is constituted, U (X) is the corresponding feature of these characteristic values
Vector, i.e. principal direction.
2. a kind of system for realizing claim 1 methods described, it is characterised in that including：Principal component analysis module, feature space
Divide module and angle anomaly data detection module, wherein：Principal component analysis module calculates the feature space mould of full dataset
Type is simultaneously exported to feature space division module progress model modification, and angle anomaly data detection module is according to from principal component analysis
The feature space of module and divide feature space model after the renewal of module from feature space and carry out angle detection and exporting different
Regular data testing result.
3. system according to claim 2, it is characterized in that, described feature space division module includes：Average updates single
Member, average Residual projection unit, feature space covariance matrix unit and Eigenvalues Decomposition unit, wherein：Average updating block
It is connected with average Residual projection unit and transmits the residual error caused by deletion candidate samples, average Residual projection unit is empty with feature
Between covariance matrix unit be connected and transmit covariance information, feature space covariance matrix unit and Eigenvalues Decomposition unit phase
Connect and transmit covariance matrix, Eigenvalues Decomposition unit, which is connected with angle anomaly data detection module and transmits spin matrix, to be believed
Breath.
4. the system according to Claims 2 or 3, it is characterized in that, described angle anomaly data detection module includes：Diagonally
Line element extraction unit, Outlier factor computing unit and threshold value comparing unit, wherein：Diagonal entry extraction unit with it is abnormal because
Sub- computing unit is connected and transmits the cosine similarity information of new and old principal direction, Outlier factor computing unit and threshold value comparison model
It is connected and transmission abnormality factor information, threshold value comparing unit compares Outlier factor and threshold value, exports testing result, the i.e. candidate
Sample is normal data or abnormal data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410802645.7A CN104463137B (en) | 2014-12-18 | 2014-12-18 | The abnormal face image detecting method and system of feature based space decomposition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410802645.7A CN104463137B (en) | 2014-12-18 | 2014-12-18 | The abnormal face image detecting method and system of feature based space decomposition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104463137A CN104463137A (en) | 2015-03-25 |
| CN104463137B true CN104463137B (en) | 2017-07-18 |
Family
ID=52909153
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410802645.7A Expired - Fee Related CN104463137B (en) | 2014-12-18 | 2014-12-18 | The abnormal face image detecting method and system of feature based space decomposition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104463137B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106547852A (en) * | 2016-10-19 | 2017-03-29 | 腾讯科技（深圳）有限公司 | Abnormal deviation data examination method and device, data preprocessing method and system |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109144988B (en) * | 2018-08-07 | 2021-02-23 | 东软集团股份有限公司 | Abnormal data detection method and device |
| CN112200075A (en) * | 2020-10-09 | 2021-01-08 | 西安西图之光智能科技有限公司 | Face anti-counterfeiting method based on anomaly detection |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101848160A (en) * | 2010-05-26 | 2010-09-29 | 钱叶魁 | Method for detecting and classifying all-network flow abnormity on line |
| CN102542246A (en) * | 2011-03-29 | 2012-07-04 | 广州市浩云安防科技股份有限公司 | Abnormal face detection method for ATM (Automatic Teller Machine) |
| CN103761516A (en) * | 2014-02-14 | 2014-04-30 | 重庆科技学院 | ATM abnormal human face detection method based on video monitoring |
-
2014
- 2014-12-18 CN CN201410802645.7A patent/CN104463137B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101848160A (en) * | 2010-05-26 | 2010-09-29 | 钱叶魁 | Method for detecting and classifying all-network flow abnormity on line |
| CN102542246A (en) * | 2011-03-29 | 2012-07-04 | 广州市浩云安防科技股份有限公司 | Abnormal face detection method for ATM (Automatic Teller Machine) |
| CN103761516A (en) * | 2014-02-14 | 2014-04-30 | 重庆科技学院 | ATM abnormal human face detection method based on video monitoring |
Non-Patent Citations (5)
| Title |
|---|
| "Kernel PCA for Novelty Detection";Heiko Hoffmann;《Pattern Recognition》;20061231;全文 * |
| "单张图像对每人多张注册图像的人脸识别";韩兆翠 等;《大连理工大学学报》;20121130;第52卷(第6期);全文 * |
| "基于减量学习的鲁棒稀疏最小二乘支持向量回归机";高润鹏 等;《探测与控制学报》;20111231;第33卷(第6期);全文 * |
| "异常人脸的定位与识别";袁宝华;《中国优秀硕士学位论文全文数据库 信息科技辑》;20051115(第7期);全文 * |
| 李权 等."基于KPCA 的多变量时间序列数据异常检测方法研究".《计算机测量与控制》.2011, * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106547852A (en) * | 2016-10-19 | 2017-03-29 | 腾讯科技（深圳）有限公司 | Abnormal deviation data examination method and device, data preprocessing method and system |
| CN106547852B (en) * | 2016-10-19 | 2021-03-12 | 腾讯科技（深圳）有限公司 | Abnormal data detection method and device, and data preprocessing method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104463137A (en) | 2015-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI543102B (en) | Method and system of cause analysis and correction for manufacturing data | |
| DE102017128693A1 (en) | Feature and limit setting for threat detection in an industrial plant control system | |
| JP5301717B1 (en) | Equipment condition monitoring method and apparatus | |
| CN108537215A (en) | A kind of flame detecting method based on image object detection | |
| WO2011043108A1 (en) | Equipment status monitoring method, monitoring system, and monitoring program | |
| CN109657547A (en) | A kind of abnormal trajectory analysis method based on associated model | |
| CN112528519A (en) | Method, system, readable medium and electronic device for engine quality early warning service | |
| CN103473540B (en) | The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality | |
| US20170261403A1 (en) | Abnormality detection procedure development apparatus and abnormality detection procedure development method | |
| Li et al. | A supervised clustering and classification algorithm for mining data with mixed variables | |
| CN102282516A (en) | Abnormality detecting method and abnormality detecting system | |
| CN103390154A (en) | Face recognition method based on extraction of multiple evolution features | |
| KR20190072652A (en) | Information processing apparatus and information processing method | |
| CN106845526A (en) | A kind of relevant parameter Fault Classification based on the analysis of big data Fusion of Clustering | |
| CN104463137B (en) | The abnormal face image detecting method and system of feature based space decomposition | |
| US7552035B2 (en) | Method to use a receiver operator characteristics curve for model comparison in machine condition monitoring | |
| CN113139600A (en) | Intelligent power grid equipment anomaly detection method and system based on federal learning | |
| JP2009068946A (en) | Flaw sorting apparatus, flaw sorting method and program | |
| CN108171214A (en) | A kind of video monitoring abnormality recognition method and system based on deep learning | |
| CN114925536A (en) | Airborne system PHM testability modeling and diagnosis strategy optimization method and device | |
| CN117421582A (en) | Equipment health analysis method based on multi-source data driving | |
| CN114443338A (en) | Sparse negative sample-oriented anomaly detection method, model construction method and device | |
| CN116720079A (en) | Wind driven generator fault mode identification method and system based on multi-feature fusion | |
| KR102486463B1 (en) | Method and Apparatus for Real Time Fault Detection Using Time series data According to Degradation | |
| CN113722230B (en) | Integrated evaluation method and device for vulnerability mining capability of fuzzy test tool |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| 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: 20170718 Termination date: 20211218 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |