CN106127247B - Image classification method based on the more example support vector machines of multitask - Google Patents
Image classification method based on the more example support vector machines of multitask Download PDFInfo
- Publication number
- CN106127247B CN106127247B CN201610466376.0A CN201610466376A CN106127247B CN 106127247 B CN106127247 B CN 106127247B CN 201610466376 A CN201610466376 A CN 201610466376A CN 106127247 B CN106127247 B CN 106127247B
- Authority
- CN
- China
- Prior art keywords
- packet
- multitask
- image
- support vector
- distance
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Abstract
The invention discloses a kind of image classification methods based on the more example support vector machines of multitask.This method comprises: establishing T learning tasks for T group image；The image of T learning tasks is carried out how instantiating；For one class packet of picture construction of each classification in T task；Establish the example in class packet to more example packets Euclidean distance formula；Example distance vector of the building class packet to more example packets；Establish class packet to more example packets weighted euclidean distance formula；The distance for constraining more example packets to generic is less than the distance to other classifications；Establish the optimization problem of the more example support vector machines of multitask；Converting optimization problem is traditional single task list example support vector machines problem；Solve support vector machines optimization problem.The present invention relates to a kind of methods for optimizing weighted euclidean distance formula, by establishing the more example support vector machines problems concerning study of multitask example images, so that optimal dissolve ideal weight, to improve the performance of Image Classifier.
Description
Technical field
The present invention relates to Image Classfication Technology fields, more particularly to the image based on the more example support vector machines of multitask
Classification method.
Background technique
With the progress of information technology and the sustainable development of social networks, the figure of magnanimity is had existed above internet
Picture, and the amount of images newly uploaded on internet daily also exponentially rises, and the scene that image is included is also more and more richer
Richness, although social network sites have obtained permanent development, the picture of magnanimity is not fully utilized but on website, and
Can all there are a large amount of new images to upload to above website daily, how identify not labeled image, and Accurate classification arrives
It is most of Internet company all in a problem of research with preferably site for service user in corresponding classification.
On the one hand, due to that may include various background elements when shooting image, then it will lead to image
It include not only a scene, if such as single example support vector machines may using traditional single example image recognition methods
Lead to misclassification.For example, at the zoo photographed when, different plant species may be photographed same image simultaneously, such as
The animals such as people, horse, bird all may be in same image.
On the other hand, due to the opening of internet and the diversity of capture apparatus, the photo of the same person may
It appears in above different social network sites, perhaps as captured by distinct device or from different video institute editings, this
A little pictures, which are mixed, to be identified, it is clear that is unreasonable；Furthermore it in order to improve the performance of Image Classifier, needs a large amount of
Markd image carry out the training of classifier, if lack of training samples, will lead to the performance decline of classifier, from
And influence the effect of image classification.The image classification of early stage is classified by way of handmarking, but this side
The artificial of method successfully can be very high, perhaps also feasible under a small amount of image, but generates speed with the present image in internet,
It is then less desirable.
Summary of the invention
Although there are many quantity in the same type of image marked face on the internet, not due to source mode
Together, for example, the equipment of shooting or the social network sites of storage are different, these pictures, which are mixed, to carry out the training of classifier and is
Unreasonable, but training is grouped according to source formation, then it can be potentially encountered lack of training samples so as to cause classification
The problems such as accuracy decline of device, it is possible to using the form of multitask, several groups picture be trained simultaneously, and utilized
The correlation of every group of picture improves the performance of every group of picture classifier.And since image contains multiple scenes, image is seen
It is handled at single example, then can neglect the correlation of multiple scenes, multi-instance learning method can be used at this time, one
A image regards multiple examples as.
Image classification method based on the more example support vector machines of multitask of the invention includes the following steps:
(1) image of several groups is obtained, and guarantees that the quantity of every group of image is few, as unit of group, establishes several
Learning tasks, and in the form of handmarking, carry out the manual sort of image.
(2) all images of all learning tasks, more sample datas are converted to.
(3) in each multi-instance learning task, an associated more example packets are constructed for each image category, this is more
Example packet in the present invention be known as class packet, and establish the example in class packet to more example packets Euclidean distance formula.
(4) building class packet to more example packets example distance vector, so that the weighting for establishing class packet to more example packets is European
Range formula.
(5) constraint is established, guarantees that more example packets will be far smaller than to the distance of other classifications to the distance of generic.
(6) optimization problem of the more example support vector machines of multitask is established.
(7) the more example support vector machines optimization problems of the multitask of switch process (6) are a similar single task list example
The optimization problem of support vector machines.
(8) the support vector machines optimization problem of solution procedure (7), can obtain the weight of optimization, to train one
A Image Classifier based on the more example support vector machines of multitask, carries out the classification of image.
Detailed description of the invention
Fig. 1 is the flow chart of the Web page classification method of the invention based on maximum spacing multitask multi-instance learning.
Specific embodiment
Image classification method based on the more example support vector machines of multitask of the invention includes the following steps:
The first step, obtains the image of several groups, and guarantees that the quantity of every group of image is few, as unit of group, if establishing
Dry learning tasks, and in the form of handmarking, carry out the manual sort of image.For example, if there is T group image, then
T Image Classifier learning tasks are established, and since the amount of images of T task is all few, handmarking can be carried out.
All images of all learning tasks are converted to more sample datas by second step.Since image contains multiple fields
Scape, and when classification, it is only necessary to one of key scenes, so whole image is converted to a single example at this time
Classify, the correlation of multiple scenes may be neglected, classifying quality is caused to be deteriorated, so at this time can be using showing more
Example learning method carries out image classification.Before multi-instance learning method, need to carry out more sample datas to image, it can
With using classical image cutting method, such as the compartmentalization for Blobworld System, the Lai Jinhang image that the present invention uses, this
When to each image-region carry out feature extraction, so that the image-region be made to be converted to an example.One image contains multiple
Region can then be converted to multiple examples, and an image is properly termed as example packet more than one at this time.
Third step constructs an associated more example packets in each multi-instance learning task for each image category,
More example packets in the present invention be known as class packet, and establish the example in class packet to more example packets Euclidean distance formula.No
As traditional more exemplary methods, the present invention does not pay close attention to the distance between image and image directly, but all of each classification
Image is combined, and establishes more example packets of a class rank, referred to as class packet, and the example established in class packet to showing more
The Euclidean distance formula of example packet, as follows:
In above formula, exampleIt is class packet CktJ-th of example,It is more example packet BitCenter.nktIt is class packet Ckt
Example number.
4th step, the example distance vector of building class packet to more example packets, to establish the weighting of class packet to more example packets
Euclidean distance formula.In the third step, can in the hope of each class packet example to more example packets apart from size, with this apart from size
For vector element, class packet is established to the example distance vector of more example packets, then k-th of class of t-th of task is clipped to i-th more shows
The example distance vector of example packetIt is as follows:
Establish one and example distance vectorThe weight vector w of equal lengthkt, which is defined as follows:
By example distance vectorWith weight vector wktWant to multiply, then the weighting of available class packet to more example packets is European
Range formula:
5th step, establish constraint, guarantee more example packets to generic distance will far smaller than arrive other classifications away from
From.Establish following constraint:
In above formula, Pt(Bit) it is more example packet BitAffiliated category set, Nt(Bit) be and more example packet BitUnrelated class
Do not gather,For error term, which ensure that classification n to more example packet BitDistance be greater than classification p to more example packets
BitDistance.
6th step establishes the optimization problem of the more example support vector machines of multitask.In t-th of task, all categories
Weight vector form a vector wt, it is as follows:
Correspondingly, one isometric vector of buildingVectorByWith-Composition, the other positions of the vector
Filling 0, it is possible to the constraints conversion established in the 5th step be following form:
Based on the constraint, wtBe converted to the form of multi-task learning, i.e. wt=w0+vt, w0It is considered as that all tasks are total
The public weight coefficient enjoyed, and vtIt is the weight coefficient that each task then exclusively enjoys, establishes the more example branch of a multitask thus
The optimization problem of vector machine is held, as follows:
In above formula, CwFor controlling error termSize, regularization parameter γ0And γ1For controlling multi-instance learning
Similitude between task.If γ0It is intended to infinity, then it is not that each multi-instance learning task, which trains the classifier come,
It is relevant.Opposite, if γ1It is intended to infinity, then it is identical that all multi-instance learning tasks, which train the classifier come,
Or it is similar.
7th step, the more example support vector machines optimization problems of multitask for turning the 6th step are a similar single task list example
The optimization problem of support vector machines.In order to use the numerical technologies such as quadratic programming solve the more examples of the multitask support to
Amount machine problem, needs the problem to be converted to the form of a similar traditional support vector machine optimization problem, therefore establishes two
Vector is as follows:
According to two above vector, the more example support vector machines of multitask of the 6th step can be converted to the support of standard
Vector machine optimization problem form is as follows:
8th step solves the support vector machines optimization problem of the 7th step, the weight of optimization can be obtained, to train
One Image Classifier based on the more example support vector machines of multitask, carries out the classification of image.
In the case where not departing from spirit of that invention or necessary characteristic, the present invention can be embodied in other specific forms.It answers
The specific embodiment various aspects are considered merely as illustrative and not restrictive.Therefore, scope of the invention such as appended claims
It is shown as indicated above shown in range.Change in all equivalent meanings and range for falling in claim should be regarded as
It falls in the scope of claim.
Claims (4)
1. a kind of image classification method based on the more example support vector machines of multitask, which comprises the steps of:
The first step, the image for obtaining several groups establish several learning tasks, and as unit of group with the shape of handmarking
Formula carries out the manual sort of image；
Second step, all images all learning tasks, are converted to more sample datas；
Third step, in each multi-instance learning task, for each image category construct an associated more example packet set,
More example packet collection are collectively referred to as class packet, and establish the example in class packet to more example packets Euclidean distance formula；
The example distance vector of 4th step, building class packet to more example packets, so that the weighting for establishing class packet to more example packets is European
Range formula；
5th step establishes constraint, guarantees that more example packets are less than the distance of other classifications to the distance of generic；
6th step, the optimization problem for establishing the more example support vector machines of multitask；
7th step, convert the more example support vector machines optimization problems of multitask of the 6th step into a single task list example support to
The optimization problem of amount machine；
8th step, the support vector machines optimization problem for solving the 7th step, can obtain the weight of optimization, to train one
Based on the Image Classifier of the more example support vector machines of multitask, the classification of image is carried out；
In 6th step, the optimization problem of the more example support vector machines of multitask is established；In t-th of task, all classes
Other weight vector forms a vector wt, it is as follows:
Correspondingly, one isometric vector of buildingVectorByWithComposition, the other positions filling of the vector
0, it is possible to the constraints conversion established in the 5th step be following form:
Based on the constraint, wtBe converted to the form of multi-task learning, i.e. wt=w0+vt, w0It is considered as all task sharings
Public weight coefficient, and vtIt is the weight coefficient that each task is exclusively enjoyed, establishes the more example supporting vectors of a multitask thus
The optimization problem of machine is as follows:
In above formula, T is task number, CwFor controlling error termSize, regularization parameter γ0And γ1It is more for controlling
Similitude between learn-by-example task, if γ0It is intended to infinity, then each multi-instance learning task trains point come
Class device is incoherent；Opposite, if γ1It is intended to infinity, then all multi-instance learning tasks train the classification come
Device is same or similar；
In 7th step, the more example support vector machines optimization problems of multitask for turning the 6th step are a single task list example branch
The optimization problem of vector machine is held, solves the more example supporting vectors of the multitask to use the numerical technologies such as quadratic programming
Machine problem, needs the problem to be converted to the form of a similar traditional support vector machine optimization problem, thus establish two to
It measures as follows:
In above formula,According to two above vector, the more example support vector machines of multitask of the 6th step can be converted
It is as follows for the support vector machines optimization problem form of standard:
2. the image classification method according to claim 1 based on the more example support vector machines of multitask, which is characterized in that
In third step, in each multi-instance learning task, an associated more example packets are constructed for each image category, this shows
Example packet be known as class packet, and establish the example in class packet to more example packets Euclidean distance formula specifically: each classification
All images are combined, and establish more example packets of a class rank, referred to as class packet, and the example established in class packet to showing more
The Euclidean distance formula of example packet, as follows:
Wherein, exampleIt is class packet CktJ-th of example,It is more example packet BitCenter, nktIt is class packet CktExample
Number.
3. the image classification method according to claim 2 based on the more example support vector machines of multitask, which is characterized in that
In 4th step, building class packet to more example packets example distance vector, thus the weighting for establishing class packet to more example packets it is European away from
From formula, in the third step, can in the hope of each class packet example to more example packets apart from size, using this apart from size as vector
Element establishes class packet to the example distance vector of more example packets, then k-th of class of t-th of task is clipped to the packet of example more than i-th
Example distance vectorIt is as follows:
Establish one and example distance vectorThe weight vector w of equal lengthkt, which is defined as follows:
By example distance vectorWith weight vector wktIt is multiplied, then weighted euclidean distance of the available class packet to more example packets
Formula:
4. the image classification method according to claim 3 based on the more example support vector machines of multitask, which is characterized in that
In 5th step, constraint is established, guarantees that more example packets are less than the distance of other classifications to the distance of generic, is established following
Constraint:
In above formula, Pt(Bit) it is more example packet BitAffiliated category set, Nt(Bit) be and more example packet BitUnrelated classification collection
It closes,For error term, which ensure that classification n to more example packet BitDistance be greater than classification p to more example packet Bit's
Distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610466376.0A CN106127247B (en) | 2016-06-21 | 2016-06-21 | Image classification method based on the more example support vector machines of multitask |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610466376.0A CN106127247B (en) | 2016-06-21 | 2016-06-21 | Image classification method based on the more example support vector machines of multitask |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106127247A CN106127247A (en) | 2016-11-16 |
| CN106127247B true CN106127247B (en) | 2019-07-09 |
Family
ID=57268191
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610466376.0A Active CN106127247B (en) | 2016-06-21 | 2016-06-21 | Image classification method based on the more example support vector machines of multitask |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106127247B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183752A (en) * | 2020-12-01 | 2021-01-05 | 南京智谷人工智能研究院有限公司 | End-to-end multi-example learning method based on automatic example selection |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392256A (en) * | 2017-07-31 | 2017-11-24 | 广东工业大学 | A kind of image-recognizing method and system |
| CN108664986B (en) * | 2018-01-16 | 2020-09-04 | 北京工商大学 | Based on lpNorm regularized multi-task learning image classification method and system |
| CN108537290A (en) * | 2018-04-25 | 2018-09-14 | 攀枝花学院 | Stellar spectra classification method based on data distribution characteristics and fuzzy membership function |
| CN109165673B (en) * | 2018-07-18 | 2021-08-31 | 广东工业大学 | Image classification method based on metric learning and multi-example support vector machine |
| CN109919165B (en) * | 2019-03-18 | 2021-07-06 | 广东工业大学 | Similarity-based multi-instance dictionary learning classification method and device |
| CN110008365B (en) * | 2019-04-09 | 2023-02-07 | 广东工业大学 | Image processing method, device and equipment and readable storage medium |
| CN110175657B (en) * | 2019-06-05 | 2021-10-01 | 广东工业大学 | Image multi-label marking method, device, equipment and readable storage medium |
| CN110414621B (en) * | 2019-08-06 | 2022-03-22 | 广东工业大学 | Classifier construction method and device based on multi-instance learning |
| CN112052840B (en) * | 2020-10-10 | 2023-02-03 | 苏州科达科技股份有限公司 | Picture screening method, system, equipment and storage medium |
| CN112766161B (en) * | 2021-01-20 | 2022-12-02 | 西安电子科技大学 | Hyperspectral target detection method based on integrated constraint multi-example learning |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945372A (en) * | 2012-10-18 | 2013-02-27 | 浙江大学 | Classifying method based on multi-label constraint support vector machine |
| CN103116893A (en) * | 2013-03-15 | 2013-05-22 | 南京大学 | Digital image labeling method based on multi-exampling multi-marking learning |
| CN104778457A (en) * | 2015-04-18 | 2015-07-15 | 吉林大学 | Video face identification algorithm on basis of multi-instance learning |
| CN105046269A (en) * | 2015-06-19 | 2015-11-11 | 鲁东大学 | Multi-instance multi-label scene classification method based on multinuclear fusion |
| CN105069473A (en) * | 2015-08-05 | 2015-11-18 | 广东工业大学 | Multi-instance weighted packet learning method for online uncertain image recognition |
| CN105117429A (en) * | 2015-08-05 | 2015-12-02 | 广东工业大学 | Scenario image annotation method based on active learning and multi-label multi-instance learning |
| CN105404877A (en) * | 2015-12-08 | 2016-03-16 | 商汤集团有限公司 | Human face attribute prediction method and apparatus based on deep study and multi-task study |
-
2016
- 2016-06-21 CN CN201610466376.0A patent/CN106127247B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102945372A (en) * | 2012-10-18 | 2013-02-27 | 浙江大学 | Classifying method based on multi-label constraint support vector machine |
| CN103116893A (en) * | 2013-03-15 | 2013-05-22 | 南京大学 | Digital image labeling method based on multi-exampling multi-marking learning |
| CN104778457A (en) * | 2015-04-18 | 2015-07-15 | 吉林大学 | Video face identification algorithm on basis of multi-instance learning |
| CN105046269A (en) * | 2015-06-19 | 2015-11-11 | 鲁东大学 | Multi-instance multi-label scene classification method based on multinuclear fusion |
| CN105069473A (en) * | 2015-08-05 | 2015-11-18 | 广东工业大学 | Multi-instance weighted packet learning method for online uncertain image recognition |
| CN105117429A (en) * | 2015-08-05 | 2015-12-02 | 广东工业大学 | Scenario image annotation method based on active learning and multi-label multi-instance learning |
| CN105404877A (en) * | 2015-12-08 | 2016-03-16 | 商汤集团有限公司 | Human face attribute prediction method and apparatus based on deep study and multi-task study |
Non-Patent Citations (3)
| Title |
|---|
| Deep Convolutional Neural Networks for Multi-Instance Multi-Task Learning;Tao Zeng 等;《2015 IEEE International Conference on Data Mining》;20151231;第579-588页 |
| 基于SVM的多示例多标签主动学习;李杰龙 等;《计算机工程与设计》;20160131;第37卷(第1期);第254-258页 |
| 基于密度聚类和多示例学习的图像分类方法;陈涛 等;《吉林大学学报（工学版）》;20140731;第44卷(第4期);第1126-1134页 |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112183752A (en) * | 2020-12-01 | 2021-01-05 | 南京智谷人工智能研究院有限公司 | End-to-end multi-example learning method based on automatic example selection |
| CN112183752B (en) * | 2020-12-01 | 2021-02-19 | 南京智谷人工智能研究院有限公司 | End-to-end multi-example learning method based on automatic example selection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106127247A (en) | 2016-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106127247B (en) | Image classification method based on the more example support vector machines of multitask | |
| Fang et al. | Perceptual quality assessment of smartphone photography | |
| CN108140032B (en) | Apparatus and method for automatic video summarization | |
| CN109558942B (en) | Neural network migration method based on shallow learning | |
| CN110147700B (en) | Video classification method, device, storage medium and equipment | |
| US8923570B2 (en) | Automated memory book creation | |
| CN110533097A (en) | A kind of image definition recognition methods, device, electronic equipment and storage medium | |
| CN106096661B (en) | The zero sample image classification method based on relative priority random forest | |
| CN109002857A (en) | A kind of transformation of video style and automatic generation method and system based on deep learning | |
| CN111444966A (en) | Media information classification method and device | |
| CN110796098B (en) | Method, device, equipment and storage medium for training and auditing content auditing model | |
| CN104112143A (en) | Weighted hyper-sphere support vector machine algorithm based image classification method | |
| CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
| CN111709477A (en) | Method and tool for garbage classification based on improved MobileNet network | |
| JP2017084340A (en) | Tag processing method and tag processing device | |
| CN111143617A (en) | Automatic generation method and system for picture or video text description | |
| CN106445977A (en) | Picture pushing method and device | |
| CN109165673A (en) | Image classification method based on metric learning and more example support vector machines | |
| CN109636764A (en) | A kind of image style transfer method based on deep learning and conspicuousness detection | |
| CN112418082A (en) | Plant leaf identification system and method based on metric learning and depth feature learning | |
| CN106250396B (en) | Automatic image label generation system and method | |
| CN110019827A (en) | A kind of corpus library generating method, device, equipment and computer storage medium | |
| WO2011096010A1 (en) | Pattern recognition device | |
| CN111444255A (en) | Training method and device of data model | |
| CN113657473A (en) | Web service classification method based on transfer learning |
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 |