CN109992690A - A kind of image search method and system - Google Patents
A kind of image search method and system Download PDFInfo
- Publication number
- CN109992690A CN109992690A CN201910180562.1A CN201910180562A CN109992690A CN 109992690 A CN109992690 A CN 109992690A CN 201910180562 A CN201910180562 A CN 201910180562A CN 109992690 A CN109992690 A CN 109992690A
- Authority
- CN
- China
- Prior art keywords
- image
- module
- feature
- vector
- target
- 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.)
- Granted
Links
Abstract
The present invention relates to a kind of image search method and systems, are related to electronic data processing field.The following steps are included: S1: obtaining n target images and extract the feature of all target images；S2: clustering the feature and constructs lexicographic tree T1；S3: according to the lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain the dictionary vector d of every target imagej；S4: query image is obtained, the dictionary vector q of the query image is calculatedj, and calculate dictionary vector qjWith dictionary vector djSimilarity sj；S5: according to the similarity sjObtain query result.This programme solves the technical issues of quick-searching for how completing millions image, the quick-searching suitable for millions image.
Description
Technical field
The present invention relates to electronic data processing field, in particular to a kind of image search method and system.
Background technique
With the development of internet, the requirement of image retrieval is also improved day by day.Existing image search method is in image
The data volume in library cannot short response time completion retrieval after improving.
Summary of the invention
The technical problem to be solved by the present invention is to how complete the quick-searching of millions image.
The technical scheme to solve the above technical problems is that a kind of image search method, comprising the following steps:
S1: obtaining n target images and extracts the feature of all target images；
S2: clustering the feature and constructs lexicographic tree T1；
S3: according to the lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain every target
The dictionary vector d of imagej；
S4: query image is obtained, the dictionary vector q of the query image is calculatedj, and calculate dictionary vector qjWith dictionary to
Measure djSimilarity sj；
S5: according to the similarity sjObtain query result.
This programme constructs lexicographic tree, simple target is calculated further according to lexicographic tree by the feature of acquisition target image
Frequency vector F in imagej, the dictionary vector d of every target image is obtained in conjunction with frequency vectorj, then obtain in the same way
Take the dictionary vector q of query imagej, the dictionary vector for comparing query image and target image obtains similarity, to be looked into
Ask result.
The beneficial effects of the present invention are: carrying out feature extraction before starting image retrieval to target image, being owned
The feature vector of target image, when inquiring image similar with query image in all target images, it is only necessary to calculate inquiry
The feature vector of image, then compare the dictionary vector q of query imagejWith the dictionary vector d of each target imagejSimilarity is obtained,
To achieve the purpose that quickly to complete quick-searching in the image of millions.
Based on the above technical solution, the present invention can also be improved as follows.
Further, step S1 is further comprising the steps of:
S11: creation includes the neural network mathematical model of at least three convolution submodule；
S12: Three Channel Color image is obtained as target image, and by the target image and is converted to 3 single channel figures
Picture；
S13: taking 3 convolution submodules as target convolution submodule, and in each target convolution submodule
Increase a pond layer below and obtains object module；
S14: 3 concatenated initial image informations of image information are obtained according to the single channel image and the object module
Matrix, and be normalized to obtain remodeling image feature information matrix；
S15: according to the remaining single channel image, respectively repeatedly step S14 is primary, obtains 3 remodeling characteristics of image letters
Cease matrix I1、I2And I3。
Beneficial effect using above-mentioned further scheme is, based on the image characteristics extraction of Three Channel Color image, to compare
More characteristics of image are extracted based on common color image, improve similarity sjReliability.
Further, step S2 is further comprising the steps of:
S21: each remodeling image feature information matrix is formed at least 3 64 Wei Te by image characteristics extraction
Sign；
S22: all features are clustered to obtain four cluster centres, respectively { μ1, μ2, μ3, μ4}；
S23: each cluster centre is clustered to obtain cluster node again, until the number of plies of lexicographic tree to be built
The class at h=4 or place again without subclass until；
S24 obtains lexicographic tree T according to the weight for the amount of images calculate node that the feature in each node covers1。
Further, step S3 is specifically included:
S31: a target image is obtained；
S32: the feature for extracting the target image obtains feature to be processed；
S33: calculating the feature to be processed at a distance from the cluster node in the object module, will be each to be processed
Feature is included into nearest dictionary tree node；
S34: the Characteristic Number to be processed in each dictionary tree node of statistics obtains frequency vector Fj；
S35: dictionary vector d is obtained according to following equationj:
dj=WT·Fj
Wherein WTFor the weight of the cluster node；
S36: repeating step S31-S35 until traversing all target images.
Beneficial effect using above-mentioned further scheme is, before retrieval and inquisition image, first by all target images
Dictionary vector djExtract, so that in the similarity for obtaining query image and each target image, it is only necessary to compare dictionary to
Measure djWith dictionary vector qj, to achieve the purpose that quick-searching.
Further, step S4 is specifically included:
S41: query image and the object module are obtained；
S42: the dictionary vector q of the query image is calculated according to the object modulej；
S43: dictionary vector q is calculated according to following equationjWith dictionary vector djSimilarity sj:
Wherein, p is the dictionary vector qjWith dictionary vector djDimension；
S44: according to the similarity sj, the image in described image library is sorted from large to small, and export preceding n images
As query result.
Another technical solution that the present invention solves above-mentioned technical problem is as follows:
A kind of image indexing system, including image input module, characteristic extracting module, lexicographic tree module and enquiry module,
Described image input module rises target image for obtaining n, and the characteristic extracting module is for extracting all target images
Feature；The lexicographic tree module is for clustering the feature and constructing lexicographic tree T1；The lexicographic tree module is also used
According to the lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain the word of every target image
Allusion quotation vector dj；Described image input module is also used to obtain query image, and the enquiry module is used for according to the lexicographic tree T1Meter
Calculate the dictionary vector q of the query imagej, and calculate dictionary vector qjWith dictionary vector djSimilarity sj, finally according to institute
State similarity sjObtain query result.
Beneficial effect using above-mentioned further scheme is, before starting image retrieval, carries out feature to target image
It extracts, obtains the feature vector of all target images, when inquiring image similar with query image in all target images, only
The feature vector of query image need to be calculated, then compares the dictionary vector q of query imagejWith the dictionary vector of each target image
djSimilarity is obtained, to achieve the purpose that quickly to complete quick-searching in the image of millions.
Further, for obtaining Three Channel Color image as target image, the feature mentions described image input module
The target image is converted to 3 for creating the neural network mathematical model including at least three convolution submodule by modulus block
A single channel image；The characteristic extracting module is also used to take 3 convolution submodules as target convolution submodule, and
Increase a pond layer behind each target convolution submodule and obtain object module, further according to single channel image described in every
3 concatenated initial image information squares of image information corresponding with single channel image described in every are obtained with the object module
Battle array, and each initial image information matrix of single channel image described in every is normalized to obtain remodeling characteristics of image
Information matrix I1、I2And I3。
Beneficial effect using above-mentioned further scheme is, based on the image characteristics extraction of Three Channel Color image, to compare
More characteristics of image are extracted based on common color image, improve similarity sjReliability.
Further, the lexicographic tree module is for mentioning each remodeling image feature information matrix by characteristics of image
It takes to form 64 dimensional feature of at least three, and all features is clustered to obtain four cluster centres, respectively { μ1, μ2,
μ3, μ4, the lexicographic tree module is also used to be clustered to obtain cluster node again to each cluster centre, until to be built
Lexicographic tree number of plies h=4 or place class again without subclass until；The lexicographic tree module is also used to according in each node
The weight of amount of images calculate node that covers of feature, obtain lexicographic tree T1。
Further, the lexicographic tree module is for obtaining all target images, and extracts the single target image
Feature obtain feature to be processed；The feature to be processed is calculated at a distance from the cluster node in the object module, it will
Each feature to be processed is included into nearest dictionary tree node；The lexicographic tree module is also used to count each dictionary
Characteristic Number to be processed in tree node, obtains frequency vector Fj；The lexicographic tree module is also used to be obtained according to following equation
The dictionary vector d of the single target imagej:
dj=WT·Fj
Wherein WTFor the weight of the cluster node.
Beneficial effect using above-mentioned further scheme is, before retrieval and inquisition image, first by all target images
Dictionary vector djExtract, so that in the similarity for obtaining query image and each target image, it is only necessary to compare dictionary to
Measure djWith dictionary vector qj, to achieve the purpose that quick-searching.
Further, the enquiry module is used to calculate the dictionary vector q of the query image according to the object modulej；
And dictionary vector q is calculated according to following equationjWith dictionary vector djSimilarity sj:
Wherein, p is the dictionary vector qjWith dictionary vector djDimension.
The advantages of additional aspect of the invention, will be set forth in part in the description, and will partially become from the following description
It obtains obviously, or practice is recognized through the invention.
Detailed description of the invention
Fig. 1 is the flow chart of the embodiment of image search method of the present invention；
Fig. 2 is the system structure diagram of the embodiment of image indexing system of the present invention；
Fig. 3 is the image characteristics extraction architecture diagram of the other embodiments of image search method of the present invention.
Specific embodiment
The principle and features of the present invention will be described below with reference to the accompanying drawings, and the given examples are served only to explain the present invention, and
It is non-to be used to limit the scope of the invention.
Embodiment is substantially as shown in Fig. 1:
Image search method in the present embodiment, comprising: S1: obtaining n target images and extracts the spy of target complete image
Sign；
S2: clustering feature and constructs lexicographic tree T1；
S3: according to lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain the word of every target image
Allusion quotation vector dj；
S4: query image is obtained, the dictionary vector q of query image is calculatedj, and calculate dictionary vector qjWith dictionary vector dj
Similarity sj；
S5: according to similarity sjObtain query result.
This programme constructs lexicographic tree, simple target is calculated further according to lexicographic tree by the feature of acquisition target image
Frequency vector F in imagej, the dictionary vector d of every target image is obtained in conjunction with frequency vectorj, then obtain in the same way
Take the dictionary vector q of query imagej, the dictionary vector for comparing query image and target image obtains similarity, to be looked into
Ask result.
Based on the above technical solution, the present invention can also be improved as follows.
Optionally, in some other embodiments, step S1 is further comprising the steps of:
S11: creation includes the neural network mathematical model of at least three convolution submodule, in the present embodiment, the nerve of creation
Network digital model can be VGG16 model, including 5 convolution modules, and 3 convolution modules next can be used as convolution
Module obtains characteristic information；
S12: Three Channel Color image is obtained as target image, and by target image and is converted to 3 single channel images；
S13: taking 3 convolution submodules as target convolution submodule, and increases behind each target convolution submodule
One pond layer obtains object module, and the pond layer in the present embodiment can be maximum pond MaxPool ing2D；
S14: obtaining the concatenated initial image information matrix of 3 image informations according to single channel image and object module, and
It is normalized to obtain remodeling image feature information matrix, can use L in the present embodiment2Place is normalized in-norm
Reason；
S15: according to remaining single channel image, respectively repeatedly step S14 is primary, obtains 3 remodeling image feature information squares
Battle array I1、I2And I3。
Based on the image characteristics extraction of Three Channel Color image, more figures are extracted compared to based on common color image
As feature, similarity s is improvedjReliability.
Optionally, in some other embodiments, step S2 is further comprising the steps of:
S21: forming 64 dimensional feature of at least three by image characteristics extraction for each remodeling image feature information matrix, this
60 64 dimensional features can be extracted in embodiment；
S22: all features are clustered to obtain four cluster centres, respectively { μ1, μ2, μ3, μ4, in the present embodiment
All characteristics of image can be carried out by hierarchical cluster using K-Means, obtained cluster centre respectively can for μ 1=α 1, α 2,
α 3, α 4, β 5, β 6, β 7, β 8, β 20, β 19, β 18, β 17, γ 10, γ 11, γ 13 }, μ 2={ α 5, α 7, α 12, α 16, β 1, β 2, β
15, β 16, γ 20, γ 19, γ 18, γ 17, γ 1, γ 2, γ 14 }, μ 3={ α 6, α 9, α 19, α 20, β 3, β 4, β 9, β 10, γ
16, γ 12, γ 4, γ 5, γ 8, γ 9, γ 15 }, μ 4={ α 8, α 10, α 11, α 13, α 14, α 15, α 17, α 18, β 11, β 12, β
13,β14,γ3,γ6,γ7}；
S23: being clustered to obtain cluster node again to each cluster centre, until the number of plies h=4 of lexicographic tree to be built
Or the class at place again without subclass until；
S24 obtains lexicographic tree T according to the weight for the amount of images calculate node that the feature in each node covers1, this
In embodiment, the weight WT of the cluster node can be calculated according to the following formula:
WT=log (N/NT) (1)
Wherein N indicates the total number of images in described image library, NTIndicate the figure covered for the feature in cluster node T
As quantity, log () is logarithmic function, N=3, N in the present embodimentT=2.
Such as utilize image T={ I1,I2,I3Building lexicographic tree, the amount of images being related in each node is such as
{A:3,B:3,C:2,D:1,E:1,H:2,F:2,I:1,G:1,J:3,K:3,N:3,L:3,O:3,R:2, S:1,P:1,M:3,T:
3,U:2}
Therefore, the weight vectors of lexicographic tree T1 nodeAre as follows:
Optionally, in some other embodiments, step S3 is specifically included:
S31: a target image is obtained；
S32: the feature for extracting target image obtains feature to be processed；
S33: calculating feature to be processed at a distance from cluster node in object module, and each feature to be processed is included into distance
In nearest dictionary tree node；
S34: the Characteristic Number to be processed in each dictionary tree node is counted, frequency vector F is obtainedj；
S35: dictionary vector d is obtained according to following equationj:
dj=WT·Fj (2)
Wherein WTFor the weight of cluster node；
S36: repeating step S31-S35 until traversing all target images.
Before retrieval and inquisition image, first by the dictionary vector d of all target imagesjIt extracts, so that being inquired
When the similarity of image and each target image, it is only necessary to compare dictionary vector djWith dictionary vector qj, to reach quick
The purpose of retrieval.
For example, as shown in Fig. 3, being equipped with image I1Characteristic point des (I1)={ α1,α2,…,α20Correspond in lexicographic tree
Node in number be respectively as follows: A:20, B:4, C:2, D:1, E:1, H:2, F:2, I:0, G:0, J:0, K:4, N:4, L:4, O:
3, R:2, S:1, P:1, M:8, T:1, U:7 } so, image I1Frequency vector F1It is as follows:
| Node | A | B | C | D | E | H | F | I | G | J | K | N | L | O | R | S | P | M | T | U |
| F1 | 20 | 4 | 2 | 1 | 1 | 2 | 2 | 0 | 0 | 0 | 4 | 4 | 4 | 3 | 2 | 1 | 1 | 8 | 1 | 7 |
Image I is obtained using formula (2)1Dictionary vector djAre as follows:
Optionally, in some other embodiments, step S4 is specifically included:
S41: query image and object module are obtained；
S42: the dictionary vector q of query image is calculated according to object modulej；
S43: dictionary vector q is calculated according to following equationjWith dictionary vector djSimilarity sj:
Wherein, p is dictionary vector qjWith dictionary vector djDimension, p=64 in the present embodiment；
S44: according to similarity sj, the image in image library is sorted from large to small, and export preceding n images as inquiry
As a result.
Another technical solution that the present invention solves above-mentioned technical problem is as follows:
A kind of image indexing system, as shown in Fig. 2, including image input module 1, characteristic extracting module 2, lexicographic tree mould
Block 3 and enquiry module 4, image input module 1 is for obtaining n target images, and characteristic extracting module 2 is for extracting target complete
The feature of image；Lexicographic tree module 3 is for clustering feature and constructing lexicographic tree T1；Lexicographic tree module 3 is also used to basis
Lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain the dictionary vector d of every target imagej；Image is defeated
Enter module 1 and be also used to obtain query image, enquiry module 4 is used for according to lexicographic tree T1Calculate the dictionary vector of query image
qj, and calculate dictionary vector qjWith dictionary vector djSimilarity sj, finally according to similarity sjObtain query result.
Before starting image retrieval, feature extraction is carried out to target image, obtains the feature vector of all target images,
When inquiring image similar with query image in all target images, it is only necessary to calculate the feature vector of query image, then right
Than the dictionary vector q of query imagejWith the dictionary vector d of each target imagejSimilarity is obtained, to reach quickly ten million
The purpose of quick-searching is completed in the image of grade.
Optionally, in some other embodiments, image input module 1 is for obtaining Three Channel Color image as target
Image, characteristic extracting module 2 is for creating the neural network mathematical model including at least three convolution submodule, and by target figure
As being converted to 3 single channel images；Characteristic extracting module 2 is also used to take 3 convolution submodules as target convolution submodule, and
Increase a pond layer behind each target convolution submodule and obtain object module, further according to every single channel image and target
Model obtains 3 concatenated initial image information matrixes of image information corresponding with every single channel image, and to every single-pass
Each initial image information matrix of road image is normalized to obtain remodeling image feature information matrix I1、I2And I3。
Based on the image characteristics extraction of Three Channel Color image, more figures are extracted compared to based on common color image
As feature, similarity s is improvedjReliability.
Optionally, in some other embodiments, lexicographic tree module 3 is for leading to each remodeling image feature information matrix
It crosses image characteristics extraction and forms 64 dimensional feature of at least three, and all features are clustered to obtain four cluster centres, respectively
{μ1, μ2, μ3, μ4, lexicographic tree module 3 is also used to be clustered to obtain cluster node again to each cluster centre, until to be built
Lexicographic tree number of plies h=4 or place class again without subclass until；Lexicographic tree module 3 is also used to according in each node
The weight for the amount of images calculate node that feature covers, obtains lexicographic tree T1。
Optionally, in some other embodiments, lexicographic tree module 3 is for obtaining all target images, and extracts single
The feature of target image obtains feature to be processed；Feature to be processed is calculated at a distance from cluster node in object module, it will be each
Feature to be processed is included into nearest dictionary tree node；Lexicographic tree module 3 is also used to count in each dictionary tree node
Characteristic Number to be processed obtains frequency vector Fj；Lexicographic tree module 3 is also used to obtain simple target image according to following equation
Dictionary vector dj:
dJ=WT·Fj
Wherein WTFor the weight of cluster node.
Before retrieval and inquisition image, first by the dictionary vector d of all target imagesjIt extracts, so that being inquired
When the similarity of image and each target image, it is only necessary to compare dictionary vector djWith dictionary vector qj, to reach quick
The purpose of retrieval.
Optionally, in some other embodiments, enquiry module 4 is used to calculate the dictionary of query image according to object module
Vector qj；And dictionary vector q is calculated according to following equationjWith dictionary vector djSimilarity sj:
Wherein, p is dictionary vector qjWith dictionary vector djDimension.
It should be noted that the various embodiments described above are product embodiments corresponding with above-mentioned each method embodiment, for this
In embodiment the explanation of each constructional device and optional embodiment can with reference in above-mentioned each method embodiment pair it should be noted that
This is repeated no more.
Reader should be understood that in the description of this specification reference term " one embodiment ", " is shown " some embodiments "
The description of example ", " specific example " or " some examples " etc. mean specific features described in conjunction with this embodiment or example, structure,
Material or feature are included at least one embodiment or example of the invention.In the present specification, above-mentioned term is shown
The statement of meaning property need not be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described
It may be combined in any suitable manner in any one or more of the embodiments or examples.In addition, without conflicting with each other, this
The technical staff in field can be by the spy of different embodiments or examples described in this specification and different embodiments or examples
Sign is combined.
It is apparent to those skilled in the art that for convenience of description and succinctly, the dress of foregoing description
The specific work process with unit is set, can refer to corresponding processes in the foregoing method embodiment, details are not described herein.
More than, only a specific embodiment of the invention, but scope of protection of the present invention is not limited thereto, and it is any to be familiar with
Those skilled in the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or substitutions,
These modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be wanted with right
Subject to the protection scope asked.
Claims (10)
1. a kind of image search method, which comprises the following steps:
S1: obtaining n target images and extracts the feature of all target images；
S2: clustering the feature and constructs lexicographic tree T1；
S3: according to the lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain every target image
Dictionary vector dj；
S4: query image is obtained, the dictionary vector q of the query image is calculatedj, and calculate dictionary vector qjWith dictionary vector dj
Similarity sj；
S5: according to the similarity sjObtain query result.
2. image search method according to claim 1, it is characterised in that: step S1 the following steps are included:
S11: creation includes the neural network mathematical model of at least three convolution submodule；
S12: Three Channel Color image is obtained as target image, and by the target image and is converted to 3 single channel images；
S13: taking 3 convolution submodules as target convolution submodule, and behind each target convolution submodule
Increase a pond layer and obtains object module；
S14: the concatenated initial image information square of 3 image informations is obtained according to the single channel image and the object module
Battle array, and be normalized to obtain remodeling image feature information matrix；
S15: according to the remaining single channel image, respectively repeatedly step S14 is primary, obtains 3 remodeling image feature information squares
Battle array I1、I2And I3。
3. image search method according to claim 2, it is characterised in that: step S2 the following steps are included:
S21: each remodeling image feature information matrix is formed into 64 dimensional feature of at least three by image characteristics extraction；
S22: all features are clustered to obtain four cluster centres, respectively { μ1, μ2, μ3, μ4}；
S23: being clustered to obtain cluster node again to each cluster centre, until the number of plies h=4 of lexicographic tree to be built
Or the class at place again without subclass until；
S24 obtains lexicographic tree T according to the weight for the amount of images calculate node that the feature in each node covers1。
4. image search method according to claim 3, it is characterised in that: step S3 is specifically included:
S31: a target image is obtained；
S32: the feature for extracting the target image obtains feature to be processed；
S33: the feature to be processed is calculated at a distance from the cluster node in the object module, by each feature to be processed
It is included into nearest dictionary tree node；
S34: the Characteristic Number to be processed in each dictionary tree node of statistics obtains frequency vector Fj；
S35: dictionary vector d is obtained according to following equationj:
dj=WT·Fj
Wherein WTFor the weight of the cluster node；
S36: repeating step S31-S35 until traversing all target images.
5. image search method according to claim 4, it is characterised in that: step S4 is specifically included:
S41: query image and the object module are obtained；
S42: the dictionary vector q of the query image is calculated according to the object modulej；
S43: dictionary vector q is calculated according to following equationjWith dictionary vector djSimilarity sj:
Wherein, p is the dictionary vector qjWith dictionary vector djDimension；
S44: according to the similarity sj, the image in described image library is sorted from large to small, and export preceding n image conducts
Query result.
6. a kind of image indexing system, it is characterised in that: including image input module, characteristic extracting module, lexicographic tree module and
Enquiry module, described image input module is for obtaining n target images, and the characteristic extracting module is for extracting described in whole
The feature of target image；The lexicographic tree module is for clustering the feature and constructing lexicographic tree T1；The lexicographic tree
Module is also used to according to the lexicographic tree T1Frequency vector F is calculatedj, further according to frequency vector FjObtain every target
The dictionary vector d of imagej；Described image input module is also used to obtain query image, and the enquiry module is used for according to
Lexicographic tree T1Calculate the dictionary vector q of the query imagej, and calculate dictionary vector qjWith dictionary vector djSimilarity sj,
Finally according to the similarity sjObtain query result.
7. image indexing system according to claim 6, it is characterised in that: described image input module is for obtaining threeway
Road color image is as target image, and the characteristic extracting module is for creating the nerve net including at least three convolution submodule
Network mathematical model, and the target image is converted into 3 single channel images；The characteristic extracting module is also used to take 3 institutes
Convolution submodule is stated as target convolution submodule, and increases a pond layer behind each target convolution submodule and obtains
To object module, 3 and every single channel image are obtained further according to single channel image described in every and the object module
The corresponding concatenated initial image information matrix of image information, and to each initial image information of single channel image described in every
Matrix is normalized to obtain remodeling image feature information matrix I1、I2And I3。
8. image indexing system according to claim 7, it is characterised in that: the lexicographic tree module is used for will be each described
Remodeling image feature information matrix forms 64 dimensional feature of at least three by image characteristics extraction, and all features are carried out
Cluster obtains four cluster centres, respectively { μ1, μ2, μ3, μ4, the lexicographic tree module is also used to in each cluster
The heart is clustered to obtain cluster node again, until the number of plies h=4 of lexicographic tree to be built or the class at place are without subclass again
Only；The weight for the amount of images calculate node that the lexicographic tree module is also used to be covered according to the feature in each node, obtains
Lexicographic tree T1。
9. image indexing system according to claim 8, it is characterised in that: the lexicographic tree module is for obtaining all institutes
Target image is stated, and the feature for extracting the single target image obtains feature to be processed；Calculate the feature to be processed and institute
Each feature to be processed is included into nearest dictionary tree node by the distance for stating the cluster node in object module；Institute
It states lexicographic tree module and is also used to count Characteristic Number to be processed in each dictionary tree node, obtain frequency vector Fj；Institute
State the dictionary vector d that lexicographic tree module is also used to obtain the single target image according to following equationj:
dj=WT·Fj
Wherein WTFor the weight of the cluster node.
10. image indexing system according to claim 9, it is characterised in that: the enquiry module is used for according to the mesh
Mark model calculates the dictionary vector q of the query imagej；And dictionary vector q is calculated according to following equationjWith dictionary vector dj's
Similarity sj:
Wherein, p is the dictionary vector qjWith dictionary vector djDimension；
The enquiry module is also used to according to the similarity sj, the image in described image library is sorted from large to small, and exports
Preceding n images are as query result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910180562.1A CN109992690B (en) | 2019-03-11 | 2019-03-11 | Image retrieval method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910180562.1A CN109992690B (en) | 2019-03-11 | 2019-03-11 | Image retrieval method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109992690A true CN109992690A (en) | 2019-07-09 |
| CN109992690B CN109992690B (en) | 2021-04-13 |
Family
ID=67129632
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910180562.1A Active CN109992690B (en) | 2019-03-11 | 2019-03-11 | Image retrieval method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109992690B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110704575A (en) * | 2019-09-21 | 2020-01-17 | 天津大学 | Dynamic self-adaptive binary hierarchical vocabulary tree image retrieval method |
| CN111199346A (en) * | 2019-12-30 | 2020-05-26 | 广东海外建设咨询有限公司 | Comprehensive management gallery supervision method and device, computer equipment and storage medium |
| CN111310712A (en) * | 2020-03-04 | 2020-06-19 | 杭州晟元数据安全技术股份有限公司 | Fast searching method based on fingerprint bag-of-words features |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013016837A1 (en) * | 2011-07-29 | 2013-02-07 | Hewlett-Packard Development Company, L.P. | Incremental image clustering |
| CN103605765A (en) * | 2013-11-26 | 2014-02-26 | 电子科技大学 | Mass image retrieval system based on cluster compactness |
| US20160125270A1 (en) * | 2005-05-09 | 2016-05-05 | Google Inc. | System And Method For Providing Objectified Image Renderings Using Recognition Information From Images |
| CN105975643A (en) * | 2016-07-22 | 2016-09-28 | 南京维睛视空信息科技有限公司 | Real-time image retrieval method based on text index |
| CN106845341A (en) * | 2016-12-15 | 2017-06-13 | 南京积图网络科技有限公司 | A kind of unlicensed vehicle identification method based on virtual number plate |
| CN107451200A (en) * | 2017-07-06 | 2017-12-08 | 西安交通大学 | Search method using Randomized Quantizing words tree and the image search method based on it |
| CN107677279A (en) * | 2017-09-26 | 2018-02-09 | 上海思岚科技有限公司 | It is a kind of to position the method and system for building figure |
| US20180276250A1 (en) * | 2009-12-07 | 2018-09-27 | Google Llc | Distributed Image Search |
| CN108986168A (en) * | 2018-06-13 | 2018-12-11 | 深圳市感动智能科技有限公司 | A kind of robot winding detection method and device combining bag of words tree-model based on depth measure study |
| CN109271546A (en) * | 2018-07-25 | 2019-01-25 | 西北大学 | The foundation of image retrieval Feature Selection Model, Database and search method |
-
2019
- 2019-03-11 CN CN201910180562.1A patent/CN109992690B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160125270A1 (en) * | 2005-05-09 | 2016-05-05 | Google Inc. | System And Method For Providing Objectified Image Renderings Using Recognition Information From Images |
| US20180276250A1 (en) * | 2009-12-07 | 2018-09-27 | Google Llc | Distributed Image Search |
| WO2013016837A1 (en) * | 2011-07-29 | 2013-02-07 | Hewlett-Packard Development Company, L.P. | Incremental image clustering |
| CN103605765A (en) * | 2013-11-26 | 2014-02-26 | 电子科技大学 | Mass image retrieval system based on cluster compactness |
| CN105975643A (en) * | 2016-07-22 | 2016-09-28 | 南京维睛视空信息科技有限公司 | Real-time image retrieval method based on text index |
| CN106845341A (en) * | 2016-12-15 | 2017-06-13 | 南京积图网络科技有限公司 | A kind of unlicensed vehicle identification method based on virtual number plate |
| CN107451200A (en) * | 2017-07-06 | 2017-12-08 | 西安交通大学 | Search method using Randomized Quantizing words tree and the image search method based on it |
| CN107677279A (en) * | 2017-09-26 | 2018-02-09 | 上海思岚科技有限公司 | It is a kind of to position the method and system for building figure |
| CN108986168A (en) * | 2018-06-13 | 2018-12-11 | 深圳市感动智能科技有限公司 | A kind of robot winding detection method and device combining bag of words tree-model based on depth measure study |
| CN109271546A (en) * | 2018-07-25 | 2019-01-25 | 西北大学 | The foundation of image retrieval Feature Selection Model, Database and search method |
Non-Patent Citations (2)
| Title |
|---|
| M.V. SUDHAMANI等: "《IMAGE: Region-Based Image Retrieval Toolbox》", 《IEEE XPLORE》 * |
| 邵杰: "《基于特定目标的大规模图像检索技术研究》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110704575A (en) * | 2019-09-21 | 2020-01-17 | 天津大学 | Dynamic self-adaptive binary hierarchical vocabulary tree image retrieval method |
| CN110704575B (en) * | 2019-09-21 | 2021-12-07 | 天津大学 | Dynamic self-adaptive binary hierarchical vocabulary tree image retrieval method |
| CN111199346A (en) * | 2019-12-30 | 2020-05-26 | 广东海外建设咨询有限公司 | Comprehensive management gallery supervision method and device, computer equipment and storage medium |
| CN111199346B (en) * | 2019-12-30 | 2023-05-05 | 广东海外建设咨询有限公司 | Utility tunnel supervision method, device, computer equipment and storage medium |
| CN111310712A (en) * | 2020-03-04 | 2020-06-19 | 杭州晟元数据安全技术股份有限公司 | Fast searching method based on fingerprint bag-of-words features |
| CN111310712B (en) * | 2020-03-04 | 2024-02-13 | 杭州晟元数据安全技术股份有限公司 | Quick searching method based on fingerprint word bag characteristics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109992690B (en) | 2021-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109992690A (en) | A kind of image search method and system | |
| DE102017011260A1 (en) | Tagging large images using a picture-with-theme embedding | |
| DE102017011262A1 (en) | Theme linking and marking for dense images | |
| Gupta et al. | Combination of global and local features using DWT with SVM for CBIR | |
| CN111339839B (en) | Intensive target detection metering method | |
| CN109964250A (en) | For analyzing the method and system of the image in convolutional neural networks | |
| CN110111325A (en) | Neuroimaging classification method, terminal and computer readable storage medium | |
| CN107679539B (en) | Single convolution neural network local information and global information integration method based on local perception field | |
| CN104951562B (en) | A kind of image search method based on VLAD dual adaptions | |
| CN104216974A (en) | Unmanned aerial vehicle aerial image matching method based on vocabulary tree blocking and clustering | |
| CN110956601B (en) | Infrared image fusion method and device based on multi-sensor mode coefficients and computer readable storage medium | |
| Wu et al. | Multiview Vehicle Tracking by Graph Matching Model. | |
| CN107391594A (en) | A kind of image search method based on the sequence of iteration vision | |
| CN113011444B (en) | Image identification method based on neural network frequency domain attention mechanism | |
| Liu et al. | Nommer: Nominate synergistic context in vision transformer for visual recognition | |
| CN114387450A (en) | Picture feature extraction method and device, storage medium and computer equipment | |
| CN109446473A (en) | Steady tensor principal component analytical method based on piecemeal | |
| CN116310452B (en) | Multi-view clustering method and system | |
| Amoda et al. | Efficient image retrieval using region based image retrieval | |
| CN106384364A (en) | LPP-ELM based objective stereoscopic image quality evaluation method | |
| CN105205496A (en) | Enhancement type sparse representation hyperspectral image classifying device and method based on space information constraint | |
| CN106570910B (en) | Based on the image automatic annotation method from coding characteristic and Neighborhood Model | |
| CN112861965A (en) | Image matching method based on multi-feature cross consistency model | |
| Wu et al. | Change detection of remote sensing images based on weighted nonnegative matrix factorization | |
| Fu et al. | Patch-based contour prior image denoising for salt and pepper noise |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |