CN103778146B - Image clustering device and method - Google Patents
Image clustering device and method Download PDFInfo
- Publication number
- CN103778146B CN103778146B CN201210406382.9A CN201210406382A CN103778146B CN 103778146 B CN103778146 B CN 103778146B CN 201210406382 A CN201210406382 A CN 201210406382A CN 103778146 B CN103778146 B CN 103778146B
- Authority
- CN
- China
- Prior art keywords
- subset
- image
- cluster
- image clustering
- goal
- 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
- 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
-
- 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
Abstract
The embodiment of the present invention provides a kind of image clustering device and method, and described image clustering method includes:Multiple images are carried out with the cluster of view-based access control model feature to obtain first set;Multiple images are carried out with the cluster of link structure to obtain second set;By visual signature information and link structure information fusion first set and second set, to obtain the result of image clustering.By the embodiment of the present invention, the accuracy of cluster result, the more consistent class of generative semantics can be improved further.
Description
Technical field
The present invention relates to image processing field, particularly to a kind of image clustering device and method.
Background technology
With digital camera and the popularization with camera function mobile phone, the acquisition of image becomes more and more easier.Additionally, mutually
The fast development of networking and web epigraph share becoming more and more popular of website, and the quantity of image is being in just explosive growth, quickly
Browsing and search for required image therefore becomes to waste time and energy.Currently rely primarily on the label of image to assist fast browsing, but
Label has polysemy, ambiguousness and inaccuracy etc. in itself and limits, and therefore can not solve this problem well.
The image self-organizing of image content-based is extremely important, and it can assistant images browse effectively.Image clustering
(Image Clustering)It is a kind of effective ways of the image self-organizing realizing image content-based, it in some way will
Similar image Rapid Combination is together.ClustTour is that be recently proposed a kind of carries out phylogenetic group to sight spot image in city
The method knitted, it is utilized respectively label information and visual information two similar diagrams of structure of image first, then in this two phases
Obtain final cluster result like a figure clustering method is applied on figure.
But, inventor finds, in prior art(Such as ClustTour)Only considered the link structure between image,
It with only the clustering method based on figure it is therefore desirable to the label information of image is it is impossible to lift cluster result further.
It is listed below for understanding the present invention and the beneficial document of routine techniquess, incorporated them into herein by quoting
In, as herein illustrated completely.
【List of references 1】S.Papadopoulos,C.Zigkolis,S.Kapiris,Y.Kompatsiaris and
A.Vakali.ClustTour:City Exploration by use of Hybrid Photo Clustering,In
Proceedings ofACM Multimedia,1617-1620,2010.
【List of references 2】X.W.Xu,N.Yuruk,Z.D.Feng and T.A.J.Schweiger.SCAN:
AStructural Clustering Algorithm for Networks,Proceedings of the 13th ACM
SIGKDDinternational conference on Knowledge discovery and data mining,824-
833,2007.
Content of the invention
The embodiment of the present invention provides a kind of image clustering device and method it is therefore intended that improving cluster result further
Accuracy, the more consistent class of generative semantics.
One side according to embodiments of the present invention, provides a kind of image clustering device, and described image clustering apparatus include:
Multiple images are carried out the cluster of view-based access control model feature to obtain first set by the first cluster cell;
Second cluster cell, carries out the cluster of link structure to obtain second set to the plurality of image;
Integrated unit, by visual signature information and first set and second set described in link structure information fusion, comes
Obtain the result of image clustering.
Other side according to embodiments of the present invention, provides a kind of image polymerization, described image clustering method bag
Include:
Multiple images are carried out with the cluster of view-based access control model feature to obtain first set;
The plurality of image is carried out with the cluster of link structure to obtain second set;
By visual signature information and first set and second set described in link structure information fusion, to obtain image and to gather
The result of class.
The beneficial effects of the present invention is:By merging the cluster of view-based access control model feature and gathering based on link structure information
Class, can improve the accuracy of cluster result, the more consistent class of generative semantics further.
With reference to explanation hereinafter and accompanying drawing, disclose in detail only certain exemplary embodiments of this invention, specify the former of the present invention
Reason can be in adopted mode.It should be understood that embodiments of the present invention are not so limited in scope.In appended power
In the range of the spirit and terms that profit requires, embodiments of the present invention include many changes, modifications and are equal to.
The feature describing for a kind of embodiment and/or illustrating can be in same or similar mode one or more
Use in individual other embodiment, combined with the feature in other embodiment, or substitute the feature in other embodiment.
It should be emphasized that term "comprises/comprising" refers to the presence of feature, one integral piece, step or assembly herein when using, but simultaneously
It is not excluded for the presence of one or more further features, one integral piece, step or assembly or additional.
Brief description
Fig. 1 is a composition schematic diagram of the image clustering device of the embodiment of the present invention 1;
Fig. 2 is a composition schematic diagram of the image clustering device of the embodiment of the present invention 2;
Fig. 3 is a composition schematic diagram of the integrated unit of the embodiment of the present invention 2;
Fig. 4 is a composition schematic diagram of the first updating block of the embodiment of the present invention 2;
Fig. 5 is a flow chart of the image clustering method of the embodiment of the present invention 3;
Fig. 6 is a flow chart of the image clustering method of the embodiment of the present invention 4;
Fig. 7 is another schematic diagram of the image clustering method of the embodiment of the present invention 4;
Fig. 8 is the flow chart that yield in the second subset is updated of the embodiment of the present invention 4.
Specific embodiment
Referring to the drawings, by description below, the aforementioned and further feature of the present invention will be apparent from.In description
In accompanying drawing, specifically disclose only certain exemplary embodiments of this invention, which show the portion that wherein can adopt the principle of the present invention
Divide embodiment it will thus be appreciated that the invention is not restricted to described embodiment, on the contrary, the present invention includes falling into appended power
Whole modifications in the range of profit requirement, modification and equivalent.
Embodiment 1
The embodiment of the present invention provides a kind of image clustering device, and Fig. 1 is the one of the image clustering device of the embodiment of the present invention
Individual composition schematic diagram.As shown in figure 1, this image clustering device 100 includes:First cluster cell 101, the second cluster cell 102
With integrated unit 103.
Wherein, the first cluster cell 101 carries out the cluster of view-based access control model feature to obtain first set to multiple images;The
Two cluster cells 102 carry out the cluster of link structure to obtain second set to multiple images;Integrated unit 103 is special by vision
Reference breath and link structure information fusion first set and second set, to obtain the result of image clustering.
In the present embodiment, multiple images can be given first, this image can be geographical marking image.For example, it is possible to
Give N number of image I={ (x1,g1),(x2,g2),…,(xn,gn), Ii=(xi,gi), wherein xiIt is a d dimensional feature vector, it
Represent the primitive character of i-th image, giIt is an e dimensional feature vector, it represents the additional information of image, for example, can be
GPS information.The purpose of the present invention is to be divided into m class to this N number of image so that as similar as possible between the image of each apoplexy due to endogenous wind.
In the present embodiment, the first cluster cell 101 can be clustered with view-based access control model feature.Can be using tradition cluster
Method, such as k-means, agglomerative clustering etc. cluster to image in Image Visual Feature, obtain view-based access control model feature
Cluster result, the image wherein with larger vision similarity will be classified as a class.
In the present embodiment, the second cluster cell 102 can build k nearest neighbor based on Image Visual Feature to multiple images
(KNN, K-Nearest Neighbor)Figure, and structuring cluster is carried out on KNN figure to obtain based on link structure information
Cluster result.
For example, it is possible to be primarily based on Image Visual Feature scheme to all picture construction KNN, concrete building process can be as
Under:Calculate the distance of each image and other all images, k before selection according to Image Visual Feature1The image of individual minimum range
K as image1Neighbour's image, each image is considered as one node of in figure, each image and its k1Individual neighbour's image is connected
Connect the side of formation figure, the vision similarity between image determines the weight on side.
Then, using a kind of structuring clustering algorithm, such as SCAN on KNN figure2, image is clustered, thus
To the cluster result based on link structure information, if wherein two images have enough common concatenated image, then they
A class will be classified as.
In the present embodiment, integrated unit 103 can merge first set and second set to obtain the knot of image clustering
Really.Due to considering the link structure between the visual signature of image and image, can be for different modalities information using not
Same clustering method, thus obtains good cluster result.
It should be noted that the above cluster only to view-based access control model feature and being illustrated based on the cluster of link structure
Property explanation.But the invention is not restricted to this, for example can also be using other clustering algorithms.Or, when being clustered based on link structure
It is not limited to KNN figure, other structures can also be built.Specific embodiment can be determined as the case may be.
From above-described embodiment, by merging the cluster of view-based access control model feature and the cluster based on link structure information,
The accuracy of cluster result, the more consistent class of generative semantics can be improved further.
Embodiment 2
On the basis of embodiment 1, the embodiment of the present invention provides a kind of image clustering device again, same as Example 1
Content repeats no more.
Fig. 2 is another composition schematic diagram of the image clustering device of the embodiment of the present invention.As shown in Fig. 2 this image gathers
Class device 200 includes:First cluster cell 201, the second cluster cell 202 and integrated unit 203.As shown in Fig. 2 this image gathers
Class device 200 can also include:Taxon 204, this taxon 204 clusters to multiple images according to classification information,
To screen to multiple images.
In the present embodiment, this classification information can be additional information of images, for example, can be GPS information, but the present invention
Not limited to this, can also be other classification informations.Taxon 204 can be gathered to image based on additional information of images
Class, and using cluster result to image filtering.
For example, it is possible to first by traditional clustering method as k-means, meanshift etc. on additional information of images such as
GPS information clusters to image, then according to certain predetermined principle filters to each class of cluster result, such as to having
The class of less image or have more inclined GPS location class carry out filter delete.Filter remaining image and will be used for follow-up gathering
Class is processed, and can input the first cluster cell 201 and the second cluster cell 202.
In the present embodiment, whole image clustering apparatus can be broadly divided into three parts:(1)Based on additional information of images such as
GPS information is clustered to image and is based on cluster result carries out image filtering;(2)View-based access control model feature and link structure information
Image is clustered;(3)Merge the cluster result of view-based access control model feature and the cluster result based on link structure.Below to tool
How body merges is schematically illustrated.
Fig. 3 is a composition schematic diagram of the integrated unit of the embodiment of the present invention, as shown in figure 3, integrated unit 203 is permissible
Including select unit 301 and processing unit 302.Wherein, one of first set and second set are gathered by select unit 301
As goal set, using another set as source set;Element during source is gathered by processing unit 302 is added to goal set
In, or the element in goal set is updated according to the element in the set of source.
Specifically, processing unit 302 can include:First computing unit 3021, combining unit 3022 and first update list
Unit 3023.Wherein, the first computing unit 3021, for one of source set the first subset, calculates this first subset and object set
The degree of overlapping of each subset closed;There is not the subset that degree of overlapping is more than predetermined threshold value in combining unit 3022 in goal set
When, the first subset is added in goal set;There is degree of overlapping more than default in goal set in the first updating block 3023
During the yield in the second subset of threshold value, yield in the second subset is updated.
In the specific implementation, can from first set and second set optionally one of cluster result as object set
Close Cd, another cluster result is as source set Cs.Source set C can be analyzedsEach of element csi(I.e. one in cluster result
Individual class), to determine whether this element is directly appended to goal set CdIn, still it is used for updating goal set CdIn some units
Element.Concrete analysis process can be as follows:
Calculate csiWith goal set CdEach of element cdjDegree of overlapping, degree of overlapping Overlap of two of which element
(csi,cdj) computing formula can be as(1):
If there is not the element that degree of overlapping is more than predetermined threshold value thr in goal set, by csiIt is added in goal set;
If there is certain element c in goal setdj, it is with element csiDegree of overlapping be more than predetermined threshold value thr, then utilize element csiTo mesh
Element c in mark setdjIt is updated.
Fig. 4 is a composition schematic diagram of the first updating block of the embodiment of the present invention, as shown in figure 4, first updates list
Unit 3023 can include:First signal generating unit 401, the second updating block 402 and the first replacement unit 403.Wherein, the first generation
Unit 401 is using the common factor of the first subset and yield in the second subset as the 3rd subset;Second updating block 402 is based on cluster measured value pair
3rd subset is updated;First replacement unit 403 replaces yield in the second subset with the 3rd subset after renewal.
Specifically, the second updating block 402 can include:Second signal generating unit 4021, the second computing unit 4022 and
Two replacement units 4023.Wherein, the second signal generating unit 4021 for be not belonging to the 3rd subset and belong to the first subset or second son
Each element of collection, forms a 4th new subset after increasing to the 3rd subset;Second computing unit 4022 for each
Four subsets calculate cluster measured value, to obtain the 4th subset of the cluster measured value with optimum;Second replacement unit
4023, when the cluster measured value of the 4th subset is better than the cluster measured value of the 3rd subset, replace the 3rd subset with the 4th subset.
In the specific implementation, the second signal generating unit 4021, the second computing unit 4022 and the second replacement unit 4023 are permissible
It is performed a plurality of times, repeats said process, the cluster measured value until not having new set is better than the cluster measured value of the 3rd subset.
In the present embodiment, image clustering device 200 can also include:3rd computing unit(In figure is not shown), the 3rd
Computing unit, for some set of multiple images, calculates the cluster measured value of this set.Wherein, cluster measured value can wrap
Include:Overall Vision correlation, local visual correlation, overall situation link correlation, in local link correlation one of them or
A combination thereof.
It is combined into any sort c with this collection below to be schematically illustrated.Two image I can be definedi, IjVisual signature similar
Spend and be
Wherein σ=mean (‖ xi-xj‖2, 1≤i ≠ j≤n) be all images two-by-two on calculate average visual feature
Distance.
Each image, as a figure node, for each image i, can calculate its feature with other images by (2)
Similarity, finds out front k2The individual image with maximum characteristic similarity, is designated as Nvk2(i), and by this image and this k2Individual image is even
Connect, form the side of figure, thus constructing k neighbour figure, the weight on the side of figure is by formula(3)Calculate.
In one embodiment, can be calculated as below for overall Vision correlation:According to Image Visual Feature, calculate
All vision similarity are averaging as overall Vision correlation by the vision similarity between any two images in class c, such as public
Formula(4).
Distgv(c)=mean(Sv,ij),i,j∈c (4)
In another embodiment, can be calculated as below for local visual correlation:According to Image Visual Feature, right
Each image in class c, finds out the k2 image most like to it from c and to calculate it similar to the vision of this k2 image
All vision similarity are averaging as local visual correlation by degree, such as formula(5).
Distlv(c)=mean(Sv,ij),i∈c,j∈NNvk2(i) (5)
In another embodiment, can be calculated as below for overall situation link correlation:All images are based on and regard
Feel feature construction k neighbour figure, find such side on k neighbour's figure, the image that it connects, all in class c, is asked to these side rights again
With obtain the first value preset;Find out such side on k neighbour's figure, its connect at least one image of image in class c, to these
The weight summation on side obtains the second value preset;Overall situation link correlation, such as formula are provided divided by the second value preset by the first value preset
(6).
In the specific implementation, the 3rd computing unit can include:4th computing unit, the 5th computing unit and the 6th calculating
Unit.Wherein, the 4th computing unit finds the image of connection all in this set one or more on the k nearest neighbor figure building
Side, the weight summation to these sides obtains the first value preset;5th computing unit finds out the image of connection at least on k nearest neighbor figure
One image, one or more side in this set, the weight summation to these sides obtains the second value preset;6th computing unit
By the first value preset divided by the second value preset to obtain overall situation link correlation.
In another embodiment, can be calculated as below for local link correlation:Obtain any two images in class c
Link weight, all-links weight is averaging as local link correlation.Wherein it is possible to be based on all images regard
Feel feature construction k neighbour figure, the link weight of two images is the weight on the side connecting this two width image on figure, if do not deposited on figure
Connect this two width image on side, link weight is 0.Can be as formula(7).
Distll(c)=mean(Wij),i,j∈c (7)
In the specific implementation, the 3rd computing unit can include:7th computing unit and the 8th computing unit.Wherein,
Seven computing units obtain the link weight of any two image in this set;8th computing unit is averaging to all-links weight
Value is to obtain local link correlation.
In the present embodiment, two kinds of cluster results can be merged in the way of certain is consistent:The cluster of view-based access control model feature
With the cluster based on link information, therefore two category informations can be considered simultaneously, thus generating the class of more semantic congruence.And
And, four kinds of different cluster measured values can be considered simultaneously, can more efficiently assess cluster, gather for as one man merging two kinds
Class result provides basis very well.Additionally, the cluster measured value of combination can also be used for other application as to all kinds of in cluster result
It is ranked up.
Embodiment 3
The embodiment of the present invention provides a kind of image polymerization, corresponding to the image polymerization in embodiment 1, identical
Content repeats no more.Fig. 5 is a flow chart of the image clustering method of the embodiment of the present invention, as shown in figure 5, this image clustering
Method includes:
Multiple images are carried out the cluster of view-based access control model feature to obtain first set by step 501;
Multiple images are carried out the cluster of link structure to obtain second set by step 502;
Step 503, by visual signature information and link structure information fusion first set and second set, to obtain figure
Result as cluster.
In the present embodiment, step 502 carries out the cluster of link structure and specifically may be used with obtaining second set to multiple images
To include:KNN figure is built to multiple images based on Image Visual Feature, and structuring is carried out on KNN figure and cluster to obtain base
Cluster result in link structure information.But the invention is not restricted to this.
From above-described embodiment, by merging the cluster of view-based access control model feature and the cluster based on link structure information,
The accuracy of cluster result, the more consistent class of generative semantics can be improved further.
Embodiment 4
The embodiment of the present invention provides a kind of image polymerization, corresponding to the image polymerization in embodiment 2, identical
Content repeats no more.Fig. 6 is a flow chart of the image clustering method of the embodiment of the present invention, as shown in fig. 6, this image clustering
Method includes:
Multiple images are clustered to screen by step 601 according to classification information;
Multiple images after screening are carried out the cluster of view-based access control model feature to obtain first set by step 602;
Multiple images after screening are carried out the cluster of link structure to obtain second set by step 603;
Step 604, by visual signature information and link structure information fusion first set and second set, to obtain figure
Result as cluster.
Fig. 7 is another schematic diagram of the image clustering method of the embodiment of the present invention, as shown in fig. 7, image can be passed through
Additional information(Such as GPS information)Screened, filtered out noise image, so that cluster result is more accurate.
In the present embodiment, concrete by visual signature information and link structure information fusion first set and second set
Can include:Using the set of one of first set and second set as goal set, using another set as source set;
Element during source is gathered is added in goal set, or updates the element in goal set according to the element in the set of source.
In the present embodiment, the element in source being gathered is added in goal set, or according to the element in the set of source
The element updating in goal set specifically can include:For source set each of the first subset, calculate the first subset and
The degree of overlapping of each subset of goal set;When there is not the subset that degree of overlapping is more than predetermined threshold value in goal set, by the
One subset is added in goal set;When there is the yield in the second subset that degree of overlapping is more than predetermined threshold value in goal set, to second
Subset is updated.
In the specific implementation, yield in the second subset is updated including:The common factor of the first subset and yield in the second subset is made
For the 3rd subset;Based on cluster measured value, the 3rd subset is updated;Replace yield in the second subset with the 3rd subset after updating.
Fig. 8 is the flow chart that yield in the second subset is updated of the embodiment of the present invention.As shown in figure 8, entering to yield in the second subset
Row updates and specifically can include:
Step 801, using the common factor of the first subset and yield in the second subset as the 3rd subset;
Step 802, for being not belonging to the 3rd subset and belonging to each element of the first subset or yield in the second subset, increases to
Form a 4th new subset after 3rd subset;
In the present embodiment, generate if there are the 4th new subset, then execution step 803;There is no the 4th new subset life
Become, then execution step 805.
Step 803, calculates cluster measured value for each the 4th new subset, to obtain the cluster measured value with optimum
The 4th subset;And judge the cluster measured value of the 4th subset whether better than the cluster measured value of the 3rd subset;
In the present embodiment, if the cluster measured value of the 4th subset is better than the cluster measured value of the 3rd subset, hold
Row step 804;Otherwise execution step 805.
Step 804, replaces the 3rd subset with the 4th subset;Then execution step 802.
Step 805, replaces yield in the second subset with the 3rd subset after updating.
That is, in the specific implementation, can be by csiAnd cdjCommon image forms class ccom.Can be for
Each belongs to csi∪cdAnd it is not belonging to ccomImage, be added to ccomIn, generate a new class.Without new class life
Become, then can use class ccomReplace cdj;Otherwise for each new class, calculate it and cluster measured value, select the cluster with optimum
One new class of measured value.Judge whether the cluster measured value of this new class is better than ccomCluster measured value, if be better than
ccom, then replaced with this new class and update class ccom, and repeat preceding step and regenerate new class, otherwise can use class ccomReplace
cdj.
In the present embodiment, image clustering method can also include:For a set of multiple images, calculate this set
Cluster measured value;Wherein this cluster measured value includes:Overall Vision correlation, local visual correlation, overall situation link are related
One of them in value, local link correlation or a combination thereof.
In one embodiment, calculate overall situation link correlation to specifically include:Find connection on the k nearest neighbor figure building
One or more side all in this set for the image, the weight summation to these sides obtains the first value preset;K nearest neighbor figure is looked for
Go out at least one image of image of connection one or more side in this set, the weight summation to these sides obtains second
Value preset;By the first value preset divided by the second value preset to obtain overall situation link correlation.
In another embodiment, calculate local link correlation to specifically include:Obtain any two figure in this set
The link weight of picture;All-links weight is averaged to obtain local link correlation.
From above-described embodiment, two kinds of cluster results can be merged in the way of certain is consistent:View-based access control model feature
Cluster and the cluster based on link information, therefore two category informations can be considered simultaneously, thus generating the class of more semantic congruence.
And it is possible to consider four kinds of different cluster measured values simultaneously, can more efficiently assess cluster, for as one man merging two kinds
Cluster result provides basis very well.Additionally, the cluster measured value of combination can also be used for other application as to each in cluster result
Class is ranked up.
The apparatus and method more than present invention can be realized by hardware it is also possible to be realized by combination of hardware software.The present invention
It is related to such computer-readable program, when this program is performed by logical block, this logical block can be made to realize above
Described device or component parts, or make this logical block realize various methods mentioned above or step.The invention still further relates to
For storing the storage medium of procedure above, such as hard disk, disk, CD, DVD, flash memory etc..
Above in association with specific embodiment, invention has been described, it will be appreciated by those skilled in the art that this
A little descriptions are all exemplary, are not limiting the scope of the invention.Those skilled in the art can be according to the present invention
Spirit and principle various variants and modifications are made to the present invention, these variants and modifications are also within the scope of the invention.
With regard to including the embodiment of above example, following remarks are also disclosed:
(Remarks 1)A kind of image clustering device, described image clustering apparatus include:
Multiple images are carried out the cluster of view-based access control model feature to obtain first set by the first cluster cell;
Second cluster cell, carries out the cluster of link structure to obtain second set to the plurality of image;
Integrated unit, by visual signature information and first set and second set described in link structure information fusion, comes
Obtain the result of image clustering.
(Remarks 2)Image clustering device according to remarks 1, wherein, described image clustering apparatus also include:
Taxon, clusters to the plurality of image according to classification information, to screen to the plurality of image.
(Remarks 3)Image clustering device according to remarks 1 or 2, wherein, described second cluster cell is regarded based on image
Feel that feature is schemed to the plurality of picture construction KNN, and structuring is carried out on described KNN figure and cluster to obtain based on link structure
The cluster result of information.
(Remarks 4)Image clustering device according to any one of remarks 1 to 3, wherein, described integrated unit includes:
Select unit, one of described first set and described second set are gathered as goal set, will be another
Individual set is as source set;
Processing unit, the element during described source is gathered is added in described goal set, or is gathered according to described source
In element update described goal set in element.
(Remarks 5)Image clustering device according to remarks 4, wherein, described processing unit includes:
First computing unit, for one of described source set the first subset, calculates described first subset and described mesh
The degree of overlapping of each subset of mark set;
Combining unit, when there is not the subset that degree of overlapping is more than predetermined threshold value in described goal set, by described first
Subset is added in goal set;
First updating block, when there is the yield in the second subset that degree of overlapping is more than predetermined threshold value in described goal set, to institute
State yield in the second subset to be updated.
(Remarks 6)Image clustering device according to remarks 5, wherein, described first updating block includes:
First signal generating unit, using the common factor of described first subset and described yield in the second subset as the 3rd subset;
Second updating block, is updated to described 3rd subset based on cluster measured value;
First replacement unit, replaces described yield in the second subset with described 3rd subset after updating.
(Remarks 7)Image clustering device according to remarks 6, wherein, described second updating block includes:
Second signal generating unit, for being not belonging to described 3rd subset and belong to the every of described first subset or yield in the second subset
Individual element, forms a 4th new subset after increasing to described 3rd subset;
Second computing unit, calculates cluster measured value for the 4th subset each described, to obtain the cluster with optimum
One the 4th subset of measured value;
Second replacement unit, is better than the cluster measured value of described 3rd subset in the cluster measured value of described 4th subset
When, replace described 3rd subset with described 4th subset.
(Remarks 8)Image clustering device according to any one of remarks 1 to 7, wherein, described image clustering apparatus also wrap
Include:
3rd computing unit, for a set of the plurality of image, calculates the cluster measured value of described set;Described
Cluster measured value includes:In overall Vision correlation, local visual correlation, overall situation link correlation, local link correlation
One of them or a combination thereof.
(Remarks 9)Image clustering device according to remarks 8, wherein, described 3rd computing unit includes:
4th computing unit, finds the image of connection one or many all in described set on the k nearest neighbor figure building
Bar side, the weight summation to described one or more side obtains the first value preset;
5th computing unit, at least one image of image finding out connection on described k nearest neighbor figure is in described set
One or more side, the second value preset is obtained to the summation of the weight on described one or more side;
6th computing unit, by described first value preset divided by described second value preset to obtain described overall situation link correlation.
(Remarks 10)Image clustering device according to remarks 8, wherein, described 3rd computing unit includes:
7th computing unit, obtains the link weight of any two image in described set;
8th computing unit, averages to all-links weight to obtain described local link correlation.
(Remarks 11)A kind of image polymerization, described image clustering method includes:
Multiple images are carried out with the cluster of view-based access control model feature to obtain first set;
The plurality of image is carried out with the cluster of link structure to obtain second set;
By visual signature information and first set and second set described in link structure information fusion, to obtain image and to gather
The result of class.
(Remarks 12)Image clustering method according to remarks 11, wherein, described image clustering method also includes:
According to classification information, the plurality of image is clustered, to screen to the plurality of image;
And, the multiple images after screening are carried out with the cluster of view-based access control model feature obtaining first set, and to sieve
Multiple images after choosing carry out the cluster of link structure to obtain second set.
(Remarks 13)Image clustering method according to remarks 11 or 12, wherein, links to the plurality of image
The cluster of structure is specifically included with obtaining second set:
Based on Image Visual Feature, the plurality of picture construction KNN is schemed, and carry out structuring cluster on described KNN figure
To obtain the cluster result based on link structure information.
(Remarks 14)Image clustering method according to any one of remarks 11 to 13, wherein, by visual signature information
Specifically include with first set and second set described in link structure information fusion:
Using the set of one of described first set and described second set as goal set, using another set as
Source is gathered;
Element during described source is gathered is added in described goal set, or according to the element in the set of described source more
The newly element in described goal set.
(Remarks 15)Image clustering method according to remarks 14, wherein, the element during described source is gathered is added to
In described goal set, or specifically included according to the element in the element described goal set of renewal in the set of described source:
For one of described source set the first subset, calculate every height of described first subset and described goal set
The degree of overlapping of collection;
When there is not the subset that degree of overlapping is more than predetermined threshold value in described goal set, described first subset is added to
In goal set;When there is the yield in the second subset that degree of overlapping is more than predetermined threshold value in described goal set, to described yield in the second subset
It is updated.
(Remarks 16)Image clustering method according to remarks 15, wherein, is updated specifically to described yield in the second subset
Including:
Using the common factor of described first subset and described yield in the second subset as the 3rd subset;
Based on cluster measured value, described 3rd subset is updated;
Replace described yield in the second subset with described 3rd subset after updating.
(Remarks 17)Image clustering method according to remarks 16, wherein, based on cluster measured value to the described 3rd son
Collection is updated specifically including:
For being not belonging to described 3rd subset and belonging to each element of described first subset or yield in the second subset, increase to
Form a 4th new subset after described 3rd subset;
Cluster measured value is calculated for the 4th subset each described, to obtain one of the cluster measured value with optimum the
Four subsets;
When the cluster measured value of described 4th subset is better than the cluster measured value of described 3rd subset, with the described 4th son
Described 3rd subset replaced by collection.
(Remarks 18)Image clustering method according to any one of remarks 11 to 17, wherein, described image clustering method
Also include:
For a set of the plurality of image, calculate the cluster measured value of described set;Described cluster measured value bag
Include:Overall Vision correlation, local visual correlation, overall situation link correlation, in local link correlation one of them or
A combination thereof.
(Remarks 19)Image clustering method according to remarks 18, wherein, calculates described overall situation link correlation concrete
Including:
The image of connection one or more side all in described set is found on the k nearest neighbor figure building, to described one
The weight summation of bar or multiple summits obtains the first value preset;
At least one image of image of connection in described set one or more is found out on described k nearest neighbor figure
Side, the weight summation to described one or more side obtains the second value preset;
By described first value preset divided by described second value preset to obtain described overall situation link correlation.
(Remarks 20)Image clustering method according to remarks 18, wherein, calculates described local link correlation concrete
Including:
Obtain the link weight of any two image in described set;
All-links weight is averaged to obtain described local link correlation.
(Remarks 21)A kind of computer-readable program, wherein when executing described program in a computer, described program makes
Obtain image clustering method as described in any one of remarks 11 to 20 for the computer execution.
(Remarks 22)A kind of storage medium of the computer-readable program that is stored with, wherein said computer-readable program makes
Image clustering method as described in any one of remarks 11 to 20 for the computer execution.
Claims (8)
1. a kind of image clustering device, described image clustering apparatus include:
Multiple images are carried out the cluster of view-based access control model feature to obtain first set by the first cluster cell;
Second cluster cell, carries out the cluster of link structure to obtain second set to the plurality of image;And
Integrated unit, by visual signature information and first set and second set described in link structure information fusion, to obtain
The result of image clustering,
Wherein, described integrated unit includes:
Select unit, one of described first set and described second set is gathered as goal set, another is collected
Cooperate to gather for source;And
Processing unit, the element during described source is gathered is added in described goal set, or according in the set of described source
Element updates the element in described goal set,
And, described processing unit includes:
First computing unit, for one of described source set the first subset, calculates described first subset and described object set
The degree of overlapping of each subset closed;
Combining unit, when there is not the subset that degree of overlapping is more than predetermined threshold value in described goal set, by described first subset
It is added in goal set;And
First updating block, when there is the yield in the second subset that degree of overlapping is more than predetermined threshold value in described goal set, to described the
Two subsets are updated.
2. image clustering device according to claim 1, wherein, described image clustering apparatus also include:
Taxon, clusters to the plurality of image according to classification information, to screen to the plurality of image.
3. image clustering device according to claim 1, wherein, described first updating block includes:
First signal generating unit, using the common factor of described first subset and described yield in the second subset as the 3rd subset;
Second updating block, is updated to described 3rd subset based on cluster measured value;
First replacement unit, replaces described yield in the second subset with described 3rd subset after updating.
4. image clustering device according to claim 3, wherein, described second updating block includes:
Second signal generating unit, for being not belonging to described 3rd subset and belong to each unit of described first subset or yield in the second subset
Element, forms a 4th new subset after increasing to described 3rd subset;
Second computing unit, calculates cluster measured value for the 4th subset each described, to obtain the cluster measurement with optimum
One the 4th subset of value;
Second replacement unit, when the cluster measured value of described 4th subset is better than the cluster measured value of described 3rd subset, uses
Described 4th subset replaces described 3rd subset.
5. the image clustering device according to any one of Claims 1-4, wherein, described image clustering apparatus also include:
3rd computing unit, for a set of the plurality of image, calculates the cluster measured value of described set;Described cluster
Measured value includes:Overall Vision correlation, local visual correlation, overall situation link correlation, its in local link correlation
One of or a combination thereof.
6. image clustering device according to claim 5, wherein, described 3rd computing unit includes:
4th computing unit, finds the image of connection one or more side all in described set on the k nearest neighbor figure building,
Weight summation to described one or more side obtains the first value preset;
5th computing unit, finds out at least one image of image of connection in described set on described k nearest neighbor figure
Bar or multiple summits, the weight summation to described one or more side obtains the second value preset;
6th computing unit, by described first value preset divided by described second value preset to obtain described overall situation link correlation.
7. image clustering device according to claim 5, wherein, described 3rd computing unit includes:
7th computing unit, obtains the link weight of any two image in described set;
8th computing unit, averages to all-links weight to obtain described local link correlation.
8. a kind of image clustering method, described image clustering method includes:
Multiple images are carried out with the cluster of view-based access control model feature to obtain first set;
The plurality of image is carried out with the cluster of link structure to obtain second set;And
By visual signature information and first set and second set described in link structure information fusion, to obtain image clustering
As a result,
Wherein, included by visual signature information and first set and second set described in link structure information fusion:
Using the set of one of described first set and described second set as goal set, using another set as source collection
Close;And
Element during described source is gathered is added in described goal set, or updates institute according to the element in the set of described source
State the element in goal set,
And, the element in the described set by described source is added in described goal set, or according in the set of described source
The process that element updates the element in described goal set includes:
For one of described source set the first subset, calculate each subset of described first subset and described goal set
Degree of overlapping;
When there is not the subset that degree of overlapping is more than predetermined threshold value in described goal set, described first subset is added to target
In set;And
When there is the yield in the second subset that degree of overlapping is more than predetermined threshold value in described goal set, described yield in the second subset is carried out more
Newly.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210406382.9A CN103778146B (en) | 2012-10-23 | 2012-10-23 | Image clustering device and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210406382.9A CN103778146B (en) | 2012-10-23 | 2012-10-23 | Image clustering device and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103778146A CN103778146A (en) | 2014-05-07 |
CN103778146B true CN103778146B (en) | 2017-03-01 |
Family
ID=50570389
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210406382.9A Active CN103778146B (en) | 2012-10-23 | 2012-10-23 | Image clustering device and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103778146B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105654102A (en) * | 2014-11-10 | 2016-06-08 | 富士通株式会社 | Data processing device and data processing method |
CN105956631A (en) * | 2016-05-19 | 2016-09-21 | 南京大学 | On-line progressive image classification method facing electronic image base |
CN106997371B (en) * | 2016-10-28 | 2020-06-23 | 华数传媒网络有限公司 | Method for constructing single-user intelligent map |
CN108805148B (en) * | 2017-04-28 | 2022-01-11 | 富士通株式会社 | Method of processing image and apparatus for processing image |
CN109978006B (en) * | 2019-02-25 | 2021-02-19 | 北京邮电大学 | Face image clustering method and device |
CN110348521A (en) * | 2019-07-12 | 2019-10-18 | 创新奇智(重庆)科技有限公司 | Image procossing clustering method and its system, electronic equipment |
CN112070144A (en) * | 2020-09-03 | 2020-12-11 | Oppo广东移动通信有限公司 | Image clustering method and device, electronic equipment and storage medium |
CN113326880A (en) * | 2021-05-31 | 2021-08-31 | 南京信息工程大学 | Unsupervised image classification method based on community division |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
CN101706806A (en) * | 2009-11-11 | 2010-05-12 | 北京航空航天大学 | Text classification method by mean shift based on feature selection |
CN102004944A (en) * | 2009-08-27 | 2011-04-06 | Sap股份公司 | Planogram compliance using automated item-tracking |
CN102509107A (en) * | 2011-10-13 | 2012-06-20 | 西北工业大学 | Local and global consistence classifying method based on sparse decomposition of figure 10 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8121358B2 (en) * | 2009-03-06 | 2012-02-21 | Cyberlink Corp. | Method of grouping images by face |
WO2012140315A1 (en) * | 2011-04-15 | 2012-10-18 | Nokia Corporation | Method, apparatus and computer program product for providing incremental clustering of faces in digital images |
-
2012
- 2012-10-23 CN CN201210406382.9A patent/CN103778146B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101295360A (en) * | 2008-05-07 | 2008-10-29 | 清华大学 | Semi-supervision image classification method based on weighted graph |
CN102004944A (en) * | 2009-08-27 | 2011-04-06 | Sap股份公司 | Planogram compliance using automated item-tracking |
CN101706806A (en) * | 2009-11-11 | 2010-05-12 | 北京航空航天大学 | Text classification method by mean shift based on feature selection |
CN102509107A (en) * | 2011-10-13 | 2012-06-20 | 西北工业大学 | Local and global consistence classifying method based on sparse decomposition of figure 10 |
Non-Patent Citations (3)
Title |
---|
一种选择性加权聚类融合算法;樊晓平 等;《计算机工程与应用》;20110729;第48卷(第22期);195-200 * |
旋转网格 : 一种新的聚类融合方法;曹巧玲 等;《计算机科学》;20110715;第38卷(第7期);157-161 * |
聚类融合算法研究及其应用;翁芳菲;《中国优秀硕士学位论文全文数据库 信息科技辑》;20090815;第2.2节,第2.4节,图2.2、2.4、2.7 * |
Also Published As
Publication number | Publication date |
---|---|
CN103778146A (en) | 2014-05-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103778146B (en) | Image clustering device and method | |
CN109948641A (en) | Anomaly groups recognition methods and device | |
CN110322356B (en) | Medical insurance abnormity detection method and system based on HIN mining dynamic multi-mode | |
Ma et al. | Decomposition-based multiobjective evolutionary algorithm for community detection in dynamic social networks | |
CN106845536B (en) | Parallel clustering method based on image scaling | |
CN111428323A (en) | Method for identifying group of key nodes by using generalized discount degree and k-shell in complex network | |
CN112800115B (en) | Data processing method and data processing device | |
CN110334157B (en) | Cloud computing management system | |
CN103400299B (en) | Method for detecting network overlapped communities based on overlapped point identification | |
CN105335368A (en) | Product clustering method and apparatus | |
Ding et al. | Community detection by propagating the label of center | |
CN106780058A (en) | The group dividing method and device of dynamic network | |
CN109509509A (en) | Protein complex method for digging based on dynamic weighting protein-protein interaction network | |
Liu et al. | Do we really need graph neural networks for traffic forecasting? | |
Wu et al. | HY-DBSCAN: A hybrid parallel DBSCAN clustering algorithm scalable on distributed-memory computers | |
CN112182306A (en) | Uncertain graph-based community discovery method | |
Wu et al. | Community detection with topological structure and attributes in information networks | |
CN110113215A (en) | One kind being based on time-weighted overlapping community detection algorithm | |
CN103902547A (en) | Increment type dynamic cell fast finding method and system based on MDL | |
Karakostas | Bridging the gap between multi-objective optimization and spatial planning: a new post-processing methodology capturing the optimum allocation of land uses against established transportation infrastructure | |
CN106354886A (en) | Method for screening nearest neighbor by using potential neighbor relation graph in recommendation system | |
CN110275895A (en) | It is a kind of to lack the filling equipment of traffic data, device and method | |
Cheng et al. | Extracting the abstraction pyramid from complex networks | |
Jia et al. | Multilayer social network overlapping community detection algorithm based on trust relationship | |
CN113436674B (en) | Incremental community detection method-TSEIA based on TOPSIS seed expansion |
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 |