CN102663418B - An image set modeling and matching method based on regression model - Google Patents
An image set modeling and matching method based on regression model Download PDFInfo
- Publication number
- CN102663418B CN102663418B CN201210076886.9A CN201210076886A CN102663418B CN 102663418 B CN102663418 B CN 102663418B CN 201210076886 A CN201210076886 A CN 201210076886A CN 102663418 B CN102663418 B CN 102663418B
- Authority
- CN
- China
- Prior art keywords
- matrix
- regression model
- classification
- sample
- image collection
- 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.)
- Active
Links
Images
Abstract
The invention discloses an image set modeling and a matching method based on regression model, comprising the following steps of organizing training image sets, establishing characteristic-category regression model, projecting the tested image set and classifying same. For image set with any form of sample distribution, the method studies bilinear regression model by utilizing sample characteristics and corresponding category mark and establishes essential semantic association between the set sample and the category thereof. In addition, it is required for the sets marked as unknown category only to perform linear regression of each sample and synthesis of category response of each are required to acquire category output of the entire set. The method has the advantages of visualization, high efficiency, simple computing, no any prior assumption to the form of distribution and scale of the set sample, and good tolerance of possible noise data in the set.
Description
Technical field
The present invention relates to technical field of computer vision, particularly a kind of modeling of the image collection based on regression model and matching process.
Background technology
For a long time, object classification is an important topic in computer vision research, and the training and testing of sorter is all that the sample of the single or little amount based on object carries out conventionally.Along with the universal development of the hardware technologies such as video camera and mass-memory unit, at a lot of new application scenarioss, such as, in the task such as video monitoring, video frequency searching, can get the great amount of images data of object, thereby for the training and testing stage of rear end classification problem provides a large amount of samples, these samples conventionally carry out modeling with the pattern of image collection and represent.In this class identification problem, each set comprises the multiple image patterns that belong to same object classification conventionally, and these image patterns have been contained object apparent changing pattern widely, such as the variation at attitude visual angle, non-rigid deformation, illumination variation etc.
The difficult point of the classification problem based on image collection is, how effectively to portray with modeling set in the distribution of multiple image, and the information of utilizing multisample to provide according to built model generalization.In recent years, typical way mainly contains two classes, respectively from parameter type and two angles of nonparametric formula to image collection modeling, the former utilizes probability distribution function to carry out the sample distribution of presentation video set conventionally, and then adopts the tolerance such as K-L divergence to calculate two similarities between probability distribution function.Latter is modeled as linear subspaces or more general non-linearity manifold according to the essential change pattern of sample in image collection, based on this modeling pattern, the problem of sets match classification just can be converted into the problem of subspace or stream shape coupling, thereby adopts various possible similarity measurement function and matching strategy on subspace or stream shape to classify.In general, this of current employing two class sets are built modeling method jointly all has hypothesis to a certain degree to the form of sample distribution in image collection, and the source of gathering sample in practical problems is normally diversified, when larger with the sample distribution form difference of model hypothesis, classifying quality just has very large deviation.
Summary of the invention
The present invention is intended at least solve the technical matters existing in prior art, has proposed to special innovation a kind of modeling and matching process of the image collection based on regression model.
In order to realize above-mentioned purpose of the present invention, the invention provides a kind of image collection modeling and matching process based on regression model, it comprises the steps:
S1: tissue training's image collection;
S2: set up feature-classification regression model;
S3: test pattern set is carried out to projection;
S4: set is classified to test pattern.
The present invention is directed to the image collection with arbitrary sample distribution form, adopt sample characteristics and corresponding classification mark study Assessment of Bilinear Regression model, set up the essential semantic association between set sample and its classification; For the set of unknown classification mark, only its each sample need be carried out linear regression then the classification response of comprehensive each sample can obtain the classification output of unitary set.The method is intuitively efficient, calculate easy, the distribution form of its pair set sample and set sample scale all without any a priori assumption, the noise data that may exist in pair set has good tolerance.
Additional aspect of the present invention and advantage in the following description part provide, and part will become obviously from the following description, or recognize by practice of the present invention.
Accompanying drawing explanation
Above-mentioned and/or additional aspect of the present invention and advantage accompanying drawing below combination is understood becoming the description of embodiment obviously and easily, wherein:
Fig. 1 the present invention is based on the image collection modeling of regression model and the process flow diagram of matching process;
Fig. 2 is the schematic diagram of partial least-square regression method of the present invention.
Embodiment
Describe embodiments of the invention below in detail, the example of described embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or has the element of identical or similar functions from start to finish.Below by the embodiment being described with reference to the drawings, be exemplary, only for explaining the present invention, and can not be interpreted as limitation of the present invention.
The invention provides a kind of image collection modeling and matching process based on regression model, as shown in Figure 1, it comprises the steps:
S1: tissue training's image collection;
S2: set up feature-classification regression model;
S3: test pattern set is carried out to projection;
S4: set is classified to test pattern.
The method of tissue training of the present invention image collection is: given m the training image set with class label, being configured to respectively deposit observation sample characteristics represents vectorial prediction matrix X and indicates vectorial response matrix Y for depositing observation sample class, wherein, m is greater than 1 positive integer.The character representation vector of the corresponding observation sample of every a line of prediction matrix X, this character representation vector can be but be not limited to gradation of image feature, wavelet character.The corresponding category attribute indication of observing sample of every a line of response matrix Y is vectorial.Particularly, suppose that given classification number is c, so the classification indication vector of each behavior c dimension of response matrix Y, if j element of a classification indication vector is 1, all the other elements are 0 entirely, are [0, ..., 1 ..., 0], show that current observation sample belongs to j class, wherein, c is greater than 1 positive integer, and j is the positive integer that is less than or equal to c.
The method that the present invention sets up feature-classification regression model is: adopt partial least-square regression method study to observe the linear regression model (LRM) between prediction matrix X and the response matrix Y of image collection: Y=XB
pLS.As shown in Figure 2, the concrete solution procedure of this model is as follows:
First, according to prediction matrix X and the response matrix Y of structure, the objective function of definition partial least squares regression:
X＝TP
T+E
Y＝UQ
T+F
Wherein, T is the semantic expressiveness matrix that predicted data X is corresponding, and U is the semantic expressiveness matrix that response data Y is corresponding, and P is the load matrix that X is corresponding, and Q is the load matrix that Y is corresponding, and E is the residual matrix that X is corresponding, and F is the residual matrix that Y is corresponding.
Then, adopt iterative optimization method to maximize the covariance between the semantic space variable T of target and U, in the present embodiment, the method that can adopt but be not limited to NIPALS solves the covariance between the semantic space variable T of target and U, wherein, is linear relationship between variable T and U, there is U=TD+H, wherein, D is diagonal matrix, and H is residual matrix.
Finally, utilize the variable T and the U that optimize gained, derive the relation between prediction matrix X and response matrix Y: Y=X (P
t)
+dQ
t, obtain linear regression model (LRM) B
pLS=(P
t)
+dQ
t.By this linear regression model (LRM), the sample characteristics that can set up different classes of observed data represents the essential semantic association between its corresponding category attribute label, thereby classifies for the sample projection of test phase.
The method that the present invention carries out projection to test set is: an image collection of given unknown classification, according to the make of training image set, represents that using each image pattern characteristic of correspondence vector is as observing image prediction matrix X
ta row vector, utilize the linear regression model (LRM) B of training stage study gained
pLS, obtain the classification that test pattern sample set is corresponding and indicate vectorial response matrix: Y
t=X
tb
pLS.
The method that the present invention classifies to test set is: by the classification of each single sample in integration test image collection, indicate vectorial ballot to obtain.Specifically there are two kinds of ballot modes:
The first ballot mode is: first according to classification corresponding to each single sample indication vector, determine classification under it, carry out maximum number ballot afterwards according to corresponding sample number of all categories.Concrete operations are: recording responses matrix Y
tthe corresponding location label of greatest member in each row vector, the frequency occurring according to each label is afterwards carried out descending sequence, and the highest label of the frequency is to category label that should test pattern set.
The second ballot mode is: first, by cumulative the classification indication vector of all single samples classification indication vector that obtains general image set, vote afterwards according to this classification indication vector for different classes of response.Concrete operations are: cumulative matrix Y
tall row vectors obtain an entirety and vector, record should and vector in the corresponding location label of greatest member value, be the category label of given test pattern set.
In the description of this instructions, the description of reference term " embodiment ", " some embodiment ", " example ", " concrete example " or " some examples " etc. means to be contained at least one embodiment of the present invention or example in conjunction with specific features, structure, material or the feature of this embodiment or example description.In this manual, the schematic statement of above-mentioned term is not necessarily referred to identical embodiment or example.And specific features, structure, material or the feature of description can be with suitable mode combination in any one or more embodiment or example.
Although illustrated and described embodiments of the invention, those having ordinary skill in the art will appreciate that: in the situation that not departing from principle of the present invention and aim, can carry out multiple variation, modification, replacement and modification to these embodiment, scope of the present invention is limited by claim and equivalent thereof.
Claims (6)
1. image collection modeling and the matching process based on regression model, is characterized in that, comprises the steps:
S1: tissue training's image collection;
S2: set up feature-classification regression model, wherein, the method for setting up feature-classification regression model is: adopt the linear regression model (LRM) between partial least-square regression method study prediction matrix X and response matrix Y: Y=XB
pLS, described linear regression model (LRM) Y=XB
pLScomprise following solution procedure:
S11: according to prediction matrix X and the response matrix Y of structure, the objective function of definition partial least squares regression:
X=TP
T+E
Y=UQ
T+F
Wherein, T is the semantic expressiveness matrix that predicted data X is corresponding, and U is the semantic expressiveness matrix that response data Y is corresponding, and P is the load matrix that X is corresponding, and Q is the load matrix that Y is corresponding, and E is the residual matrix that X is corresponding, and F is the residual matrix that Y is corresponding,
S12: the covariance in employing iterative optimization method maximization target semantic space between variable T and U wherein, is linear relationship between T and U, has U=TD+H, and wherein, D is diagonal matrix, and H is residual matrix,
S13: accumulate Spatial Semantics variable T and U in utilization optimization gained, derive the linear regression model (LRM) between prediction matrix X and response matrix Y: Y=X (P
t)
+dQ
t, obtain B
pLS=(P
t)
+dQ
t;
S3: test pattern set is carried out to projection;
S4: set is classified to test pattern.
2. image collection modeling and the matching process based on regression model as claimed in claim 1, it is characterized in that, the method of described tissue training image collection is: given m the training image set with class label, being configured to respectively deposit observation sample characteristics represents vectorial prediction matrix X and indicates vectorial response matrix Y for depositing observation sample class, wherein, the character representation vector of the corresponding observation sample of every a line of prediction matrix X, the corresponding classification indication of observing sample of every a line of response matrix Y is vectorial, described m is greater than 1 positive integer.
3. image collection modeling and the matching process based on regression model as claimed in claim 1, it is characterized in that, the method of test set being carried out to projection is: an image collection of given unknown classification, according to the make of training image set, each image pattern characteristic of correspondence is represented to vector is as observing image prediction matrix X
ta row vector, utilize the linear regression model (LRM) of training stage study gained to obtain B
pLS, obtain the classification that test pattern sample set is corresponding and indicate vectorial response matrix: Y
t=X
tb
pLS.
4. image collection modeling and the matching process based on regression model as claimed in claim 1, is characterized in that, the method that test set is classified is: in integration test image collection, the classification of each single sample indicates vectorial ballot to obtain.
5. image collection modeling and the matching process based on regression model as claimed in claim 4, it is characterized in that, described ballot mode is: according to classification corresponding to each single sample indication vector, determine classification under it, carry out maximum number ballot afterwards according to corresponding sample number of all categories.
6. image collection modeling and the matching process based on regression model as claimed in claim 4, it is characterized in that, described ballot mode is: by cumulative the classification indication vector of all single samples classification indication vector that obtains general image set, according to described classification indication vector, for different classes of response, vote.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210076886.9A CN102663418B (en) | 2012-03-21 | 2012-03-21 | An image set modeling and matching method based on regression model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210076886.9A CN102663418B (en) | 2012-03-21 | 2012-03-21 | An image set modeling and matching method based on regression model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102663418A CN102663418A (en) | 2012-09-12 |
| CN102663418B true CN102663418B (en) | 2014-04-23 |
Family
ID=46772902
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210076886.9A Active CN102663418B (en) | 2012-03-21 | 2012-03-21 | An image set modeling and matching method based on regression model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102663418B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104424373B (en) * | 2013-08-26 | 2017-05-31 | 中国科学院亚热带农业生态研究所 | A kind of fine expression of space variable correlation |
| CN105928901B (en) * | 2016-07-11 | 2019-06-07 | 上海创和亿电子科技发展有限公司 | A kind of near-infrared quantitative model construction method that qualitative, quantitative combines |
| WO2018064800A1 (en) * | 2016-10-08 | 2018-04-12 | Nokia Technologies Oy | Apparatus, method and computer program product for distance estimation between samples |
| CN110209878B (en) * | 2018-08-02 | 2022-09-20 | 腾讯科技（深圳）有限公司 | Video processing method and device, computer readable medium and electronic equipment |
| CN109740018B (en) * | 2019-01-29 | 2021-03-02 | 北京字节跳动网络技术有限公司 | Method and device for generating video label model |
| CN109961032B (en) * | 2019-03-18 | 2022-03-29 | 北京字节跳动网络技术有限公司 | Method and apparatus for generating classification model |
| CN113434716B (en) * | 2021-07-02 | 2024-01-26 | 泰康保险集团股份有限公司 | Cross-modal information retrieval method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551864A (en) * | 2009-05-13 | 2009-10-07 | 西安电子科技大学 | Image classification method based on feature correlation of frequency domain direction |
| CN102270347A (en) * | 2011-08-05 | 2011-12-07 | 上海交通大学 | Target detection method based on linear regression model |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8442309B2 (en) * | 2009-06-04 | 2013-05-14 | Honda Motor Co., Ltd. | Semantic scene segmentation using random multinomial logit (RML) |
-
2012
- 2012-03-21 CN CN201210076886.9A patent/CN102663418B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551864A (en) * | 2009-05-13 | 2009-10-07 | 西安电子科技大学 | Image classification method based on feature correlation of frequency domain direction |
| CN102270347A (en) * | 2011-08-05 | 2011-12-07 | 上海交通大学 | Target detection method based on linear regression model |
Non-Patent Citations (2)
| Title |
|---|
| 基于偏最小二乘的人脸超分辨率重构;胡宇等;《北京理工大学学报》;20100930;第30卷(第9期);第1098页-1101页 * |
| 胡宇等.基于偏最小二乘的人脸超分辨率重构.《北京理工大学学报》.2010,第30卷(第9期),第1098页-1101页. |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102663418A (en) | 2012-09-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102663418B (en) | An image set modeling and matching method based on regression model | |
| CN102187336B (en) | Method for identifying an object in a video archive | |
| US10013634B2 (en) | Multi-sensor visual analytics | |
| CN102163290B (en) | Method for modeling abnormal events in multi-visual angle video monitoring based on temporal-spatial correlation information | |
| Kim et al. | Color–texture segmentation using unsupervised graph cuts | |
| CN109508671B (en) | Video abnormal event detection system and method based on weak supervision learning | |
| KR20180118596A (en) | Semi-automatic labeling of data sets | |
| CN102955902B (en) | Method and system for evaluating reliability of radar simulation equipment | |
| CN104298679A (en) | Application service recommendation method and device | |
| Zhang et al. | Online change-point detection in sparse time series with application to online advertising | |
| CN111105628A (en) | Parking lot portrait construction method and device | |
| CN110019891A (en) | Image storage method, image search method and device | |
| CN110458096A (en) | A kind of extensive commodity recognition method based on deep learning | |
| CN109416836A (en) | Information processing equipment, information processing method and information processing system | |
| CN102722732B (en) | Image set matching method based on data second order static modeling | |
| CN108932509A (en) | A kind of across scene objects search methods and device based on video tracking | |
| CN108154183A (en) | A kind of objective classification method based on part and depth characteristic set | |
| CN109948703A (en) | Gene image procossing estimation method, system, medium and equipment based on deep learning | |
| Muthumanickam et al. | Supporting exploration of eye tracking data: Identifying changing behaviour over long durations | |
| Reif et al. | Meta2-features: Providing meta-learners more information | |
| CN106682582A (en) | Compressed sensing appearance model-based face tracking method and system | |
| CN105468887B (en) | Data analysis system and method | |
| Yildirim et al. | Evaluating salient object detection in natural images with multiple objects having multi‐level saliency | |
| CN110110795A (en) | Image classification method and device | |
| CN109376473A (en) | The calculation method and device of vehicle control device, vehicle and its course continuation mileage |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |