CN105426447B - A kind of related feedback method based on the learning machine that transfinites - Google Patents
A kind of related feedback method based on the learning machine that transfinites Download PDFInfo
- Publication number
- CN105426447B CN105426447B CN201510757225.6A CN201510757225A CN105426447B CN 105426447 B CN105426447 B CN 105426447B CN 201510757225 A CN201510757225 A CN 201510757225A CN 105426447 B CN105426447 B CN 105426447B
- Authority
- CN
- China
- Prior art keywords
- image
- data
- result
- training
- feature
- 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
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 18
- 239000003550 marker Substances 0.000 claims description 8
- 239000012141 concentrate Substances 0.000 claims description 2
- 235000013399 edible fruits Nutrition 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 4
- 230000003993 interaction Effects 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- 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
Abstract
A kind of related feedback method based on the learning machine that transfinites, the secondary query image of present invention input one；Image is retrieved, search result is obtained, allows user that result is marked；SIFT feature, Color feature and LBP feature are extracted respectively to labeled image；Utilize three kinds of feature three fundamental classifiers of training；The image for retrieving picture library is respectively put into three base classifiers, is voted according to prediction result, each unmarked picture of pair is marked automatically；Re -training updates classifier；Classify to picture library picture；It returns the result.The present invention establishes on the basis of transfiniting learning machine, by the query intention for introducing the mankind, carry out human-computer interaction, effectively learning data is enriched using unmarked picture library image, the precision of image feedback can be made to greatly improve, and processing speed is well controlled, and the expression of image in a computer is made more to meet understanding of the mankind to image, semantic, and the present invention is made to have good feedback effects.
Description
Technical field
The invention belongs to field of image processing, the related feedback method that is related in image retrieval, and in particular to one kind is based on
Transfinite the related feedback method of learning machine.
Background technique
Current social has stepped into the mainly huge data age based on multimedia information data, wherein again with number
Image information data is the most prominent.Compared with other multimedia data, the content in image data is richer, and expression is more
Intuitively, certainty and have become in daily life information and share very main form.In face of increasing
Image information data, excavate the voluminous amount of information hidden deeply in image data, effectively so as in large-scale picture number
According to quickly and accurately finding out user's image information actually required in library.This trend has been increasingly becoming computer view
The Main Topics of the related fieldss such as feel and multimedia data information retrieval.But as the low-level image feature of image and high level
Between semanteme there is huge spread, this image data information retrieval based on content still cannot obtain satisfied knot
Fruit.Relevant feedback is to improve the effective way of content-based image retrieval effect, and a comprehensive feedback can be mentioned greatly
High retrieval precision.However, data are essential for relevant feedback, but use and manually manually mark a large amount of numbers
According to being an intolerable lengthy procedure.Due to the specific demand to relevant feedback, has label if can make full use of
And very rare data, and a large amount of unlabelled data, this is a kind of ideal in the way of data.
In recent years, applying for learning machine of transfiniting is widely used in different field, but is rarely used for phase
Close feedback.Since the learning machine that transfinites is with very good nicety of grading and processing time, and precision is high, speed is simultaneously fastly
Evaluate the vital factor of relevant feedback performance.
The invention proposes an image retrieval based on the learning machine relevant feedback scheme that transfinites with Gaussian kernel, with
Phase overcomes the limitation of above-mentioned time and precision, and this method is cascaded using three Weak Classifiers, to form powerful classifier, uses
To learn different features to extract a small amount of handmarking's data characteristics, then using in its next automatic terrestrial reference note image data base
A large amount of data untagged.It can see from experiment, using the method for Gaussian kernel, nicety of grading with higher, simultaneously
The processing time is largely also declined.Meanwhile it has used the data of a small amount of handmarking as training data, and
Unlabelled data have been used while the frame for combining coorinated training.Therefore, how image retrieval is effectively utilized
Characteristic is an emphasis in image retrieval research field.
Summary of the invention
At this several years recently, with the explosive swift and violent growth of the image data volume of digital form, this needs was efficiently
And effective method, to allow user to scan for by so a large amount of set.Therefore under such conditions, based on interior
The image retrieval technologies of appearance become more and more popular, and many systems are by this trend development in this many decades.Therefore exist
In the present invention, the related feedback method based on the learning machine that transfinites is proposed, the method can organically combine image retrieval
With the feature in relevant feedback, and application these features construct feedback method.
A kind of related feedback method based on the learning machine that transfinites, it is characterised in that the following steps are included:
Step 1, the secondary query image of input one；
Step 2, image is retrieved, obtains search result, allow user that result is marked；
Step 3, SIFT feature, Color feature and LBP feature are extracted respectively to labeled image；
Step 4, three kinds of feature three fundamental classifiers of training are utilized；
Step 5, the image for retrieving picture library is respectively put into three base classifiers, is voted according to prediction result,
Each unmarked picture of pair is marked automatically；
Step 6, re -training updates classifier；
Step 7, classify to picture library picture；
Step 8, it returns the result.
The method have the advantages that:
The present invention establishes on the basis of transfiniting learning machine, by introducing the query intention of the mankind, carries out human-computer interaction, has
Effect enriches learning data using unmarked picture library image, the precision of image feedback can be made to greatly improve, and handle speed
Degree is well controlled, and so that the expression of image in a computer is more met understanding of the mankind to image, semantic, makes tool of the present invention
There are good feedback effects.
Detailed description of the invention
Fig. 1 is the overview flow chart of invention the method.
Fig. 2 is the flow chart of voting process in this method.
Fig. 3 is to mark the flow chart without mark data in this method automatically.
Specific embodiment
The present invention is described further With reference to embodiment.
The process of related feedback method of the present invention based on the learning machine that transfinites is as shown in Figure 1, comprising the following steps:
Step 1, the secondary query image of input one；
Step 2, image is retrieved, obtains search result, allow user that result is marked；
Step 2.1, in searching system, after what user needed to mark is the result of primary retrieval or once feeds back
As a result.Picture after user's mark is as data set L, and the not marked picture of user is as data set U, and user does not have at this time
Labeled data set is entire picture library data set D-L.It include positive example P and negative example N, therefore L in labeled data set
=P ∪ N.
Step 3, SIFT feature, Color feature and LBP feature are extracted respectively to labeled image；
Step 4, three kinds of feature three fundamental classifiers of training are utilized；
Step 4.1, by the enlightenment of coorinated training algorithm, respectively to have mark training set L and without mark training set U extract three kinds
Feature is respectively trained three base classifiers with these three features, can increase data in this way to the otherness of classifier group, effectively
Reduction uncertainty as brought by single data Expressive Features, prediction effect is enhanced with this.
Step 5, the image for retrieving picture library is respectively put into three base classifiers, is voted according to prediction result,
Each unmarked picture of pair is marked automatically；
Step 5.1, entire data untagged collection U is put into the classifier group being made of three base classifiers and is carried out in advance
It surveys, voting mechanism concentrates corresponding i as shown in Fig. 2, providing data untagged after the prediction of each base classifierthPredictive marker,
Predictive marker is denoted as Yin, all predictive marker collection ψ obtained are predicted according to three kinds of different characteristics are as follows:
Step 6, re -training updates classifier, automatic to mark data untagged collection；
Step 6.1, after poll closing, the L_temp of all data untaggeds with predictive marker result is obtained, L_ is used
Temp carrys out re -training classifier group, to promote the classification capacity of classifier device group.After updated classifier group, then will be without mark
Note data set U, which is put into classifier group, carries out prediction classification, as shown in Figure 3.
Step 6.2, as a result, obtaining two group data sets for marking with positive example and marking with negative example.With positive example
The data set of label is denoted as P*, and the data set with negative example label is denoted as N*.
Step 7, it returns the result.
Step 8, decide whether to carry out lower side feedback operation to user, if it is skip to step 2.1.
Claims (1)
1. a kind of related feedback method based on the learning machine that transfinites, it is characterised in that: method includes the following steps:
Step 1, the secondary query image of input one；
Step 2, image is retrieved, obtains search result, allow user that result is marked；
Step 2.1, in searching system, what user needed to mark be primary retrieval the either primary feedback of result after knot
Fruit；Picture after user's mark is as data set L, and the not marked picture of user is as data set U, and user does not have at this time
Labeled data set is entire picture library data set D-L；It include positive example P and negative example N, therefore L=P in labeled data set
∪N；
Step 3, SIFT feature, Color feature and LBP feature are extracted respectively to labeled image；
Step 4, three kinds of feature three fundamental classifiers of training are utilized；
Step 4.1, by the enlightenment of coorinated training algorithm, respectively to have mark training set L and without mark training set U extract three kinds of spies
Sign, is respectively trained three base classifiers with these three features, can increase data in this way to the otherness of classifier group, effectively
Reduce as brought by single data Expressive Features uncertainty, prediction effect is enhanced with this；
Step 5, the image for retrieving picture library is respectively put into three base classifiers, is voted according to prediction result, to every
One secondary unmarked picture is marked automatically；
Step 5.1, entire data untagged collection U is put into the classifier group being made of three base classifiers and is predicted, often
Data untagged, which is provided, after one base classifier prediction concentrates corresponding ithPredictive marker, predictive marker is denoted as Yin, according to three kinds
Different characteristic predicts all predictive marker collection ψ obtained are as follows:
Step 6, re -training updates classifier, automatic to mark data untagged collection；
Step 6.1, after poll closing, the L_temp of all data untaggeds with predictive marker result is obtained, L_temp is used
Carry out re -training classifier group, to promote the classification capacity of classifier device group；After updated classifier group, then by unmarked number
It is put into classifier group according to collection U and carries out prediction classification；
Step 6.2, as a result, obtaining two group data sets for marking with positive example and marking with negative example；It is marked with positive example
Data set be denoted as P*, with negative example label data set be denoted as N*；
Step 7, it returns the result；
Step 8, decide whether to carry out lower side feedback operation to user, if it is skip to step 2.1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510757225.6A CN105426447B (en) | 2015-11-09 | 2015-11-09 | A kind of related feedback method based on the learning machine that transfinites |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510757225.6A CN105426447B (en) | 2015-11-09 | 2015-11-09 | A kind of related feedback method based on the learning machine that transfinites |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105426447A CN105426447A (en) | 2016-03-23 |
| CN105426447B true CN105426447B (en) | 2019-02-01 |
Family
ID=55504659
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510757225.6A Active CN105426447B (en) | 2015-11-09 | 2015-11-09 | A kind of related feedback method based on the learning machine that transfinites |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105426447B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106779091B (en) * | 2016-12-23 | 2019-02-12 | 杭州电子科技大学 | A kind of periodic vibration signal localization method based on transfinite learning machine and arrival distance |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1851703A (en) * | 2006-05-10 | 2006-10-25 | 南京大学 | Active semi-monitoring-related feedback method for digital image search |
| CN102024030A (en) * | 2010-11-30 | 2011-04-20 | 辽宁师范大学 | Multi-classifier integration method based on maximum expected parameter estimation |
| CN102436589A (en) * | 2010-09-29 | 2012-05-02 | 中国科学院电子学研究所 | Complex object automatic recognition method based on multi-category primitive self-learning |
| CN104484666A (en) * | 2014-12-17 | 2015-04-01 | 中山大学 | Advanced image semantic parsing method based on human-computer interaction |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9082047B2 (en) * | 2013-08-20 | 2015-07-14 | Xerox Corporation | Learning beautiful and ugly visual attributes |
-
2015
- 2015-11-09 CN CN201510757225.6A patent/CN105426447B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1851703A (en) * | 2006-05-10 | 2006-10-25 | 南京大学 | Active semi-monitoring-related feedback method for digital image search |
| CN102436589A (en) * | 2010-09-29 | 2012-05-02 | 中国科学院电子学研究所 | Complex object automatic recognition method based on multi-category primitive self-learning |
| CN102024030A (en) * | 2010-11-30 | 2011-04-20 | 辽宁师范大学 | Multi-classifier integration method based on maximum expected parameter estimation |
| CN104484666A (en) * | 2014-12-17 | 2015-04-01 | 中山大学 | Advanced image semantic parsing method based on human-computer interaction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105426447A (en) | 2016-03-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107423376B (en) | Supervised deep hash rapid picture retrieval method and system | |
| Spolaôr et al. | A systematic review on content-based video retrieval | |
| CN102663015B (en) | Video semantic labeling method based on characteristics bag models and supervised learning | |
| CN106407406B (en) | text processing method and system | |
| WO2015185019A1 (en) | Semantic comprehension-based expression input method and apparatus | |
| US20150074112A1 (en) | Multimedia Question Answering System and Method | |
| US10482146B2 (en) | Systems and methods for automatic customization of content filtering | |
| CN108228569B (en) | Chinese microblog emotion analysis method based on collaborative learning under loose condition | |
| CN108334610A (en) | A kind of newsletter archive sorting technique, device and server | |
| CN106126605B (en) | Short text classification method based on user portrait | |
| CN104360993B (en) | A kind of method from content needed for Text Feature Extraction | |
| CN105426533B (en) | A kind of image search method merging space constraint information | |
| CN106569996B (en) | A kind of Sentiment orientation analysis method towards Chinese microblogging | |
| CN105404677A (en) | Tree structure based retrieval method | |
| CN111611366A (en) | Intention recognition optimization processing method, device, equipment and storage medium | |
| CN105426447B (en) | A kind of related feedback method based on the learning machine that transfinites | |
| Medida et al. | An optimized E-lecture video retrieval based on machine learning classification | |
| Ragnarsdóttir et al. | Exquisitor: breaking the interaction barrier for exploration of 100 million images | |
| CN111223014B (en) | Method and system for online generation of subdivision scene teaching courses from a large number of subdivision teaching contents | |
| CN107656760A (en) | Data processing method and device, electronic equipment | |
| US11328218B1 (en) | Identifying subjective attributes by analysis of curation signals | |
| Rostami et al. | Proposing a method to classify texts using data mining | |
| CN105426490A (en) | Tree structure based indexing method | |
| CN107861943B (en) | Method for quickly extracting useful data from document set | |
| CN111723208A (en) | Conditional classification tree-based legal decision document multi-classification method and device and terminal |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20160323 Assignee: Zhihua Yunjing Henan Digital Technology Co.,Ltd. Assignor: Beijing University of Technology Contract record no.: X2024980000226 Denomination of invention: A Correlation Feedback Method Based on Overlimit Learning Machine Granted publication date: 20190201 License type: Common License Record date: 20240105 Application publication date: 20160323 Assignee: Luoyang Lexiang Network Technology Co.,Ltd. Assignor: Beijing University of Technology Contract record no.: X2024980000083 Denomination of invention: A Correlation Feedback Method Based on Overlimit Learning Machine Granted publication date: 20190201 License type: Common License Record date: 20240104 |
|
| EE01 | Entry into force of recordation of patent licensing contract |