CN108898166A - A kind of image labeling method - Google Patents
A kind of image labeling method Download PDFInfo
- Publication number
- CN108898166A CN108898166A CN201810605917.2A CN201810605917A CN108898166A CN 108898166 A CN108898166 A CN 108898166A CN 201810605917 A CN201810605917 A CN 201810605917A CN 108898166 A CN108898166 A CN 108898166A
- Authority
- CN
- China
- Prior art keywords
- class
- image
- distance
- clustering algorithm
- cluster centre
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of image labeling methods, include the following steps：Each exemplar that data are concentrated is clustered using improved FCM clustering algorithm, the data set of different semantemes is divided into different classes, obtains the cluster centre set of each class；Calculate image to each class cluster centre Euclidean distance, be calculated image to each class average distance, acquire be with the smallest class of image distance image mark class；Image nearest class of cluster centre distance into class is found, the mark word that the mark word that frequency of occurrence is most in class is image is counted.Image labeling method provided by the invention, each semantic label class is clustered using improved FCM clustering algorithm, new distance measure algorithm is used in improved FCM clustering algorithm, substantially increase the accuracy rate of image labeling, it is good to mark effect, the needs of practical application can be met well.
Description
Technical field
The invention belongs to technical field of image processing, and in particular to a kind of image labeling method.
Background technique
In recent years, as the development of computer technology, memory technology is getting faster and digital image information and internet
It is universal, a large amount of data pass through major website daily and digital product generates, and at a terrific speed on the internet
It is propagated.A large amount of random data needs are distinguished and are handled, then are identified and the storage of specification these data, so as to
More easily applied.This just gives how effectively to retrieve and process these data in heaps and establish problem.Automated graphics
With regard to very good solution, these mark the matter of semantics of images to label technology (Automatic Image Annotation, AIA),
Reflect the content of image by being automatically labeled keyword to these images.This technology passes through the figure marked
As library or other tools have trained a set of semantic relational model with image feature space, and by this model to not marking
Image carries out semantic tagger, it attempts to establish the corresponding relationship of image, semantic feature and characteristics of the underlying image, to handle image
Semantic gap problem.
In the prior art, common image labeling method is based on fuzzy C-mean algorithm (fuzzy c-means, be abbreviated as FCM)
The image labeling method of clustering algorithm, traditional FCM clustering algorithm are similarity using traditional Euclidean distance, with image spy
The similarity of the value of sign and the difference of cluster centre as objective function and similar sample, the accuracy of classification is not high, the boundary of classification
Limit is more fuzzy, makes to distinguish between class and class not enough obviously, this results in the figure based on traditional FCM clustering algorithm of the prior art
As the mark accuracy rate of mask method is not high, mark ineffective, it would be highly desirable to which those skilled in the art change the mask method
Into.
Summary of the invention
For above-mentioned problems of the prior art, it can avoid above-mentioned skill occur the purpose of the present invention is to provide one kind
The image labeling method of art defect.
In order to achieve the above-mentioned object of the invention, technical solution provided by the invention is as follows：
A kind of image labeling method, includes the following steps：
Step 1：Each exemplar that data are concentrated is clustered using improved FCM clustering algorithm, it will be different
Semantic data set is divided into different classes, obtains the cluster centre set of each class；
Step 2：Calculate image to each class cluster centre Euclidean distance, be calculated image to each class putting down
Equal distance, acquire be with the smallest class of image distance image mark class；
Step 3：It finds image nearest class of cluster centre distance into class, it is most to count frequency of occurrence in class
Mark word is the mark word of image.
Further, the improved FCM clustering algorithm includes the following steps：
The collection that step 1) obtains different exemplar central points is combined into C { (c1, c2..., cj), wherein j representative sample class
Number, the feature for then obtaining each tag class is stored in set X respectivelyij{(xi1, xi2..., xin) wherein i represent in class
Picture number, j represents the number of class, n representative sample intrinsic dimensionality；
Step 2) utilizes obtained inter- object distance dikWith between class distance dijUpdate Measure Formula
3) cluster centre function is updated using improved distance measure：
And subordinating degree function：
4) combined using center point set, sample feature set and fuzziness m and parameter θ and find out the poly- of each tag class
Class center ciAnd subordinating degree function uik, new central point is obtained using the cluster centre acquired, wherein the value model of fuzziness m
It enclosesThe reasonable value range of θ isThe termination condition of cluster, i.e. iteration error ε
=0.01, maximum number of iterations is set as 150；WhenOr when circulating beyond given number, stop following
Ring；R record recycles current number.
Further, the objective function of the improved FCM clustering algorithm is
Further, the objective function of the improved FCM clustering algorithm is
Further, the cluster centre function of the improved FCM clustering algorithm is：
Further, the subordinating degree function of the improved FCM clustering algorithm is
Further, it in the step 2, acquires and is with formula used in the smallest class of image distance to be marked：
Image labeling method provided by the invention carries out each semantic label class using improved FCM clustering algorithm
It clusters, new distance measure algorithm is used in improved FCM clustering algorithm, overcome in traditional FCM clustering algorithm and only consider
Single inter- object distance relationship is without the defect of the whole relation between considering similar foreign peoples's sample, to substantially increase image mark
The accuracy rate of note, mark effect is good, can meet the needs of practical application well.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawing and specific implementation
The present invention will be further described for example.It should be appreciated that described herein, specific examples are only used to explain the present invention, and does not have to
It is of the invention in limiting.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work premise
Under every other embodiment obtained, shall fall within the protection scope of the present invention.
Traditional fuzzy C-mean algorithm (fuzzy c-means, be abbreviated as FCM) clustering algorithm using it is traditional it is European away from
From for similarity, it is usually to utilize X=(x1, x2..., xn) and Y=(y1, y2..., yn) two vectors carry out similarity fortune
It calculates, Euclidean distance is typical similarity calculating method, and Euclidean distance calculation formula is
That formula (1) indicates is the distance between point-to-point, x in high-dimensional vectori、yiIn ∈ X, and xi、yiIt respectively indicates
I-th of seat target value of x and Y.
Traditional FCM clustering algorithm is all using the value of characteristics of image and the difference of cluster centre as objective function and similar sample
Similarity, as Dik=| | xk-ci||2.Using the similarity measure between sample as distance, expression is FCM clustering algorithm
Sample is at a distance from cluster centre, and when the degree of membership of sample and a certain class cluster centre is bigger, the distance between they are smaller,
Otherwise when sample is bigger at a distance from cluster centre, the degree of membership between them is smaller.But for FCM clustering algorithm come
Say, the degree of membership of their samples to class it is not absolute 0 and 1, the fuzzy value between 0 to 1, this result in classification standard
True property is not high, and the boundary of classification is more fuzzy, can make to distinguish between class and class to a certain extent not obvious enough.
In order to improve the accuracy of traditional fuzzy C-means clustering, present invention employs a kind of new distance measure algorithm,
Based on unilaterally considering distance between similar, caused cluster centre and degree of membership defect constrained each other are proposed a kind of by class
At a distance between class and new distance measuring method that similar distance is merged, using image, semantic label as similar sample
Between the foreign peoples's sample judgement of distance, learn to new distance measure considered not only the tightness degree of similar sample also
The sparse degree between different samples is considered, formula (2) is new distance measure formula：
Wherein,Indicate k to cluster centre ciEuclidean distance, dcij=| | ci-cj||2It indicates
It is cluster centre ciTo cjEuclidean distance.The tightness that new distance measure considers identical semantic sample simultaneously also allows for
The separation degree of different semantic samples, when mutually similar interior distanceMore hour represents tightness in similar sample class and gets over
Greatly, if class and between class distanceWhen bigger, the dispersion degree represented between foreign peoples's sample is bigger.Only when similar sample distance
With foreign peoples's sample distanceDifference be in an optimal solution in the case where, the absolute value of objective function gets to minimum, cluster
Effect is just better, and the accuracy rate of mark is just higher.Signified similar foreign peoples's sample representation sample of same label and not in text
With the sample of label.In addition, formula (3) is that joined variable element θ in the distance of foreign peoples's semanteme sample to control its class spacing
From weight, make final to estimate acquirement optimal value.
The belonging relation in cluster between point-to-point can be effectively embodied by distance measure, can effectively improve mark
Precision.New distance measure considers the distance between distance and foreign peoples's sample of similar sample simultaneously, is not merely
Using point to cluster centre distance as uniquely measure, the intervention of foreign peoples's sample distance improve it is traditional apart from thought, and
The specific gravity of similar foreign peoples's sample distance is dynamically balanced by making the difference to take absolute value, makes it to obtain better Clustering Effect, mention
Height mark accuracy rate.
Traditional FCM clustering algorithm is to determine that the point in clustering belongs to the algorithm of the degree of which class by degree of membership.
FCM clustering algorithm is a kind of clustering algorithm based on division, and fuzzy concept therein is a kind of uncertainty, what certainty referred to
Be it is non-black i.e. white, things is only divided into positive and negative two kinds of possibilities.Uncertainty is a kind of fuzzy concept, and what it was indicated is one
Thing tends to a kind of degree of possibility, is a kind of possibility.It can be described as this possibility a number from 0 to 1
Value, rather than non-zero then 1.The similarity degree of a type belonging to one sample point is called degree of membership, indicates to be subordinate to using u
Degree.Its degree of membership u meets formula (4), and the degree of membership summation in a sample database is 1.
Formula (4) is the condition formula of constraint condition.The generalized form of the objective function of traditional FCM clustering algorithm is：
In formula (5) objective function be as degree of membership at a distance from each sample to cluster centre composed by, m is to be subordinate to
The multiplier of degree can dispose the weight of sample.Lagrange's method of multipliers is used to objective function, and combines the constraint of degree of membership
Condition, first to uijIt asks local derviation that can calculate degree of membership formula to be：
Secondly to ciIt just asks local derviation that can obtain cluster centre function to be：
By above formula as it can be seen that cluster centre function and subordinating degree function interact, the relationship for including each other.Appoint
Meaning both is assigned to one of initial value, if meet condition can grey iterative generation finally work as J until objective function J tends towards stability
It tends towards stability when converging to a stable value.
The new distance measure algorithm that the present invention uses improves the defect of traditional FCM clustering algorithm, traditional FCM cluster
Algorithm is more isolated for the judgement of numerical point, does not have effective method of adjustment for the cluster centre being randomly generated.The present invention
The difference of the similar foreign peoples's sample distance used, improve only consider single inter- object distance relationship in traditional FCM clustering algorithm and
The defect of the whole relation between similar foreign peoples's sample is not accounted for.Formula (8) is improved objective function：
Formula (9) obtains new unconfined objective function by the way that Lagrange multiplier is added in objective function：
To objective function for degree of membership uikDerivation obtains：
The constraint for being 1 by the sum of degree of membership, above formula can be obtained degree of membership formula with abbreviation and be：
Then to cluster centre ViDerivation can derive that cluster centre formula is：
Wherein
If i=j
If i ≠ j
Formula (8)-(15) are the modified flow based on the improved FCM clustering algorithm of similar foreign peoples's sample, are using identical
Label and different labels are tested as similar and foreign peoples sample, and formula (8) is improved objective function, formula (11)
For the subordinating degree function obtained based on new distance measure, formula (12) is improved cluster centre function.
The invention proposes a kind of new clustering methods using grey iterative generation, calculate distance between foreign peoples's sample for the first time
When, the central point in each label is obtained using traditional clustering algorithm, as parameter required for the different between class distances of calculating.
Every class exemplar is clustered respectively followed by improved clustering algorithm, is obtained in several clusters of every class sample
The heart, after the cluster centre for obtaining every class, using the average point of the cluster centre in result as the central point of new foreign peoples's distance,
The new cluster result of new central point and each exemplar can be continuously available by clustering to every a kind of sample.
Using the distance of central point as the distance between foreign peoples's sample, stop iteration when reaching specified the number of iterations.
The present invention is exactly to be improved using the distance between the distance of similar sample and foreign peoples's sample to clustering algorithm, is led to
Crossing continuous iteration makes Clustering Effect tend towards stability.The center in class is determined by calculating the cluster centre point in similar sample
Point calculates the distance between foreign peoples's sample with central point, and by improved degree of membership formula and cluster centre formula come
Improve tradition FCM clustering algorithm.Improved FCM clustering algorithm considered not only the tightness in class, it is also contemplated that class with
Sparse degree between class.Traditional FCM clustering algorithm is only considered a little to be improved to the thought of the distance of cluster centre.Tool
Body, the realization process of improved FCM clustering algorithm includes the following steps：
1) collection for obtaining different exemplar central points first is combined into C { (c1’c2..., cj), wherein j representative sample class
Number, the feature for then obtaining each tag class is stored in set X respectivelyij{(xi1, xi2..., xin) wherein i represent in class
Picture number, j represents the number of class, n representative sample intrinsic dimensionality.
2) obtained inter- object distance d is utilizedikWith between class distance dijUpdate Measure Formula
3) cluster centre function is updated using improved distance measure：
And subordinating degree function：
4) combined using center point set, sample feature set and fuzziness m and parameter θ and find out the poly- of each tag class
Class center ciAnd subordinating degree function uik, new central point is obtained using the cluster centre acquired, wherein the value model of fuzziness m
It enclosesThe reasonable value range of θ isThe termination condition of cluster, i.e. iteration error ε
=0.01, maximum number of iterations is set as 150.WhenIt (r record recycles current number) or circulates beyond
When given number, stop circulation.
Mask method based on similar foreign peoples's sample in the present invention is in conjunction with the pass between identical semantic and different semantic samples
System, obtains new distance measure and improves the Clustering Model of traditional FCM clustering algorithm, by improved FCM clustering algorithm application
It has arrived in image labeling.
Fuzzy C-mean algorithm (FCM) clustering algorithm can calculate each sample to the degree of membership of affiliated class, that is, for data
The arbitrary sample of concentration not only belongs to certain class or is not belonging to two kinds of situations of certain class.
Traditional mask method based on cluster is clustered to entire data set, then according to cluster result to training
Image is classified, then determines class belonging to test image, finally counts the markup information in class, the as mark of test image
Word.But tradition FCM clustering algorithm belongs to a kind of unsupervised clustering algorithm, when data set is larger, classifying quality is simultaneously paid no attention to
Think.Since the boundary of characteristics of the underlying image is more fuzzy, the image of different semantic samples may be divided into one kind, it can will be identical
Semantic sample is divided into inhomogeneity, will affect mark effect in this way.
Refering to what is shown in Fig. 1, image labeling method of the invention includes the following steps：
Step 1：Each exemplar that data are concentrated is clustered using above-mentioned improved FCM clustering algorithm, it will
Different semantic data sets are divided into different classes, obtain the cluster centre set of each class
Cj{(ck1, ck2..., ckn)}；
Step 2：The low-level image feature of test image is obtained, the Euclidean distance of the cluster centre of calculating image to each class leads to
The Euclidean distance of cluster centre for crossing image to each class is calculated with the cluster centre number being currently located in sample wait mark
Infuse image to each class average distance, using formula (16) acquire with the smallest class of image distance to be marked, and wait mark
The note the smallest class of image distance is the mark class of image；
Step 3：After the mark class of image has been determined, obtained by being clustered using improved FCM clustering algorithm
Result on the basis of, find test image nearest class of cluster centre distance into class, count in class frequency of occurrence most
More mark words, the most mark word of the frequency of occurrence are the mark word of image.
Image labeling method proposed by the present invention gathers each semantic label class using improved FCM clustering algorithm
Class, rather than clustered in whole data set.Method of the invention considers the semantic information of image in test set,
And the sample of each semanteme is clustered respectively, the image in different labels is avoided at a distance of from smaller, same label
Image at a distance of from larger problem.
Embodiments of the present invention above described embodiment only expresses, the description thereof is more specific and detailed, but can not
Therefore limitations on the scope of the patent of the present invention are interpreted as.It should be pointed out that for those of ordinary skill in the art,
Without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection model of the invention
It encloses.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (7)
1. a kind of image labeling method, which is characterized in that include the following steps：
Step 1：Each exemplar that data are concentrated is clustered using improved FCM clustering algorithm, by different semantemes
Data set be divided into different classes, obtain the cluster centre set of each class；
Step 2：Calculate image to each class cluster centre Euclidean distance, be calculated image to each class average departure
From, acquire be with the smallest class of image distance image mark class；
Step 3：Image nearest class of cluster centre distance into class is found, the mark that frequency of occurrence is most in class is counted
Word is the mark word of image.
2. image labeling method according to claim 1, which is characterized in that the improved FCM clustering algorithm include with
Lower step：
The collection that step 1) obtains different exemplar central points is combined into C { (c1, c2..., cj), wherein j representative sample class is a
Number, the feature for then obtaining each tag class are stored in set X respectivelyij{(xi1, xi2..., xin) wherein i represent the figure in class
Piece number, j represent the number of class, n representative sample intrinsic dimensionality.
Step 2) utilizes obtained inter- object distance dikWith between class distance dijUpdate Measure Formula
3) cluster centre function is updated using improved distance measure：
And subordinating degree function：
4) combined using center point set, sample feature set and fuzziness m and parameter θ and found out in the cluster of each tag class
Heart ciAnd subordinating degree function uik, new central point is obtained using the cluster centre acquired, wherein the value range of fuzziness mThe reasonable value range of θ isThe termination condition of cluster, i.e. iteration error ε=
0.01, maximum number of iterations is set as 150.WhenOr when circulating beyond given number, stop circulation；r
Record recycles current number.
3. image labeling method according to claim 1, which is characterized in that the target of the improved FCM clustering algorithm
Function is
4. image labeling method according to claim 1, which is characterized in that the target of the improved FCM clustering algorithm
Function is
5. image labeling method according to claim 1, which is characterized in that the cluster of the improved FCM clustering algorithm
Central function is：
6. image labeling method according to claim 1, which is characterized in that the improved FCM clustering algorithm is subordinate to
Spending function is
7. image labeling method according to claim 1, which is characterized in that in the step 2, acquire and figure to be marked
Image distance is from formula used in a smallest class：
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810605917.2A CN108898166A (en) | 2018-06-13 | 2018-06-13 | A kind of image labeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810605917.2A CN108898166A (en) | 2018-06-13 | 2018-06-13 | A kind of image labeling method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108898166A true CN108898166A (en) | 2018-11-27 |
Family
ID=64344904
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810605917.2A Pending CN108898166A (en) | 2018-06-13 | 2018-06-13 | A kind of image labeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108898166A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147852A (en) * | 2019-05-29 | 2019-08-20 | 北京达佳互联信息技术有限公司 | Method, apparatus, equipment and the storage medium of image recognition |
| CN110990617A (en) * | 2019-11-27 | 2020-04-10 | 广东智媒云图科技股份有限公司 | Picture marking method, device, equipment and storage medium |
| CN111407269A (en) * | 2020-03-30 | 2020-07-14 | 华南理工大学 | EEG signal emotion recognition method based on reinforcement learning |
| CN111582327A (en) * | 2020-04-22 | 2020-08-25 | 郑州轻工业大学 | Intelligent terminal-oriented automatic labeling method for motion recognition training samples |
| CN112654999A (en) * | 2020-07-21 | 2021-04-13 | 华为技术有限公司 | Method and device for determining labeling information |
| CN113283242A (en) * | 2021-05-31 | 2021-08-20 | 西安理工大学 | Named entity recognition method based on combination of clustering and pre-training models |
| CN114362973A (en) * | 2020-09-27 | 2022-04-15 | 中国科学院软件研究所 | K-means and FCM clustering combined flow detection method and electronic device |
| WO2022194049A1 (en) * | 2021-03-15 | 2022-09-22 | 华为技术有限公司 | Object processing method and apparatus |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102426697A (en) * | 2011-10-24 | 2012-04-25 | 西安电子科技大学 | Image segmentation method based on genetic rough set C-mean clustering |
| CN102436583A (en) * | 2011-09-26 | 2012-05-02 | 哈尔滨工程大学 | Image segmentation method based on annotated image learning |
| CN103839261A (en) * | 2014-02-18 | 2014-06-04 | 西安电子科技大学 | SAR image segmentation method based on decomposition evolution multi-objective optimization and FCM |
| CN104077352A (en) * | 2014-05-27 | 2014-10-01 | 浙江大学 | Energy model based image semantic annotation method |
| CN104298987A (en) * | 2014-10-09 | 2015-01-21 | 西安电子科技大学 | Handwritten numeral recognition method based on point density weighting online FCM clustering |
| CN105740881A (en) * | 2016-01-22 | 2016-07-06 | 天津中科智能识别产业技术研究院有限公司 | Partially-annotated image clustering method and partially-annotated image clustering device based on matrix decomposition |
| CN106971397A (en) * | 2017-04-01 | 2017-07-21 | 郭建辉 | Based on the city high-resolution remote sensing image dividing method for improving JSEG algorithms |
-
2018
- 2018-06-13 CN CN201810605917.2A patent/CN108898166A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102436583A (en) * | 2011-09-26 | 2012-05-02 | 哈尔滨工程大学 | Image segmentation method based on annotated image learning |
| CN102426697A (en) * | 2011-10-24 | 2012-04-25 | 西安电子科技大学 | Image segmentation method based on genetic rough set C-mean clustering |
| CN103839261A (en) * | 2014-02-18 | 2014-06-04 | 西安电子科技大学 | SAR image segmentation method based on decomposition evolution multi-objective optimization and FCM |
| CN104077352A (en) * | 2014-05-27 | 2014-10-01 | 浙江大学 | Energy model based image semantic annotation method |
| CN104298987A (en) * | 2014-10-09 | 2015-01-21 | 西安电子科技大学 | Handwritten numeral recognition method based on point density weighting online FCM clustering |
| CN105740881A (en) * | 2016-01-22 | 2016-07-06 | 天津中科智能识别产业技术研究院有限公司 | Partially-annotated image clustering method and partially-annotated image clustering device based on matrix decomposition |
| CN106971397A (en) * | 2017-04-01 | 2017-07-21 | 郭建辉 | Based on the city high-resolution remote sensing image dividing method for improving JSEG algorithms |
Non-Patent Citations (4)
| Title |
|---|
| 刘璐 等: "基于类内类间距离的模糊C-均值聚类分割算法", 《计算机工程与设计》 * |
| 王浩 等: "基于距离测度学习的AP聚类图像标注", 《计算机工程与应用》 * |
| 赵洪宋 等: "基于遗传算法和FCM的图像自动标注", 《计算机与数字工程》 * |
| 鲍泓 等: "自动图像标注技术研究进展", 《计算机科学》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147852A (en) * | 2019-05-29 | 2019-08-20 | 北京达佳互联信息技术有限公司 | Method, apparatus, equipment and the storage medium of image recognition |
| US11263483B2 (en) | 2019-05-29 | 2022-03-01 | Beijing Dajia Internet Information Technology Co., Ltd. | Method and apparatus for recognizing image and storage medium |
| CN110990617A (en) * | 2019-11-27 | 2020-04-10 | 广东智媒云图科技股份有限公司 | Picture marking method, device, equipment and storage medium |
| CN110990617B (en) * | 2019-11-27 | 2024-04-19 | 广东智媒云图科技股份有限公司 | Picture marking method, device, equipment and storage medium |
| CN111407269A (en) * | 2020-03-30 | 2020-07-14 | 华南理工大学 | EEG signal emotion recognition method based on reinforcement learning |
| CN111582327A (en) * | 2020-04-22 | 2020-08-25 | 郑州轻工业大学 | Intelligent terminal-oriented automatic labeling method for motion recognition training samples |
| CN112654999B (en) * | 2020-07-21 | 2022-01-28 | 华为技术有限公司 | Method and device for determining labeling information |
| CN112654999A (en) * | 2020-07-21 | 2021-04-13 | 华为技术有限公司 | Method and device for determining labeling information |
| CN114362973A (en) * | 2020-09-27 | 2022-04-15 | 中国科学院软件研究所 | K-means and FCM clustering combined flow detection method and electronic device |
| CN114362973B (en) * | 2020-09-27 | 2023-02-28 | 中国科学院软件研究所 | K-means and FCM clustering combined flow detection method and electronic device |
| WO2022194049A1 (en) * | 2021-03-15 | 2022-09-22 | 华为技术有限公司 | Object processing method and apparatus |
| CN113283242A (en) * | 2021-05-31 | 2021-08-20 | 西安理工大学 | Named entity recognition method based on combination of clustering and pre-training models |
| CN113283242B (en) * | 2021-05-31 | 2024-04-26 | 西安理工大学 | Named entity recognition method based on combination of clustering and pre-training model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898166A (en) | A kind of image labeling method | |
| Qin et al. | Research progress on semi-supervised clustering | |
| CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
| CN104899253B (en) | Towards the society image across modality images-label degree of correlation learning method | |
| Shu et al. | P-odn: Prototype-based open deep network for open set recognition | |
| Fang et al. | Adaptive core fusion-based density peak clustering for complex data with arbitrary shapes and densities | |
| Li et al. | Adaptive metric learning for saliency detection | |
| Niu et al. | Knowledge-based topic model for unsupervised object discovery and localization | |
| CN110807086B (en) | Text data labeling method and device, storage medium and electronic equipment | |
| Li et al. | Fusing semantic aspects for image annotation and retrieval | |
| CN103778206A (en) | Method for providing network service resources | |
| CN110830291B (en) | Node classification method of heterogeneous information network based on meta-path | |
| Wang et al. | Spherere: Distinguishing lexical relations with hyperspherical relation embeddings | |
| Hui et al. | Inter-class angular loss for convolutional neural networks | |
| CN116882414B (en) | Automatic comment generation method and related device based on large-scale language model | |
| Ren et al. | Uncertainty-guided Boundary Learning for Imbalanced Social Event Detection | |
| Cao et al. | The research on medical image classification algorithm based on PLSA-BOW model | |
| Papapanagiotou et al. | Improving concept-based image retrieval with training weights computed from tags | |
| Zhuang et al. | Non-exhaustive learning using gaussian mixture generative adversarial networks | |
| Song et al. | Iterative 3D shape classification by online metric learning | |
| Wang et al. | CenterMatch: A Center Matching Method for Semi-supervised Facial Expression Recognition | |
| Chen et al. | Multi-modal multi-layered topic classification model for social event analysis | |
| Li et al. | Non-Co-Occurrence Enhanced Multi-Label Cross-Modal Hashing Retrieval Based on Graph Convolutional Network | |
| Xu et al. | Improving classification accuracy on uncertain data by considering multiple subclasses | |
| Yan et al. | A potential-based density estimation method for clustering using decision graph |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181127 |
|
| WD01 | Invention patent application deemed withdrawn after publication |