CN108182188A - A kind of image search method and device based on rarefaction representation - Google Patents
A kind of image search method and device based on rarefaction representation Download PDFInfo
- Publication number
- CN108182188A CN108182188A CN201611123413.4A CN201611123413A CN108182188A CN 108182188 A CN108182188 A CN 108182188A CN 201611123413 A CN201611123413 A CN 201611123413A CN 108182188 A CN108182188 A CN 108182188A
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- input picture
- module
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5862—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
Landscapes
- Engineering & Computer Science (AREA)
- Library & Information Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of image search method and device based on rarefaction representation, the method comprising the steps of S1, input picture collection pre-processes the input picture in image set；Step S2 using group's sparse features selection strategy, selects input picture and the characteristic information of image data base, forms characteristics of image library；Step S3, the feature in the feature and image data base of input picture carry out specific measurement and compare, calculate similarity, obtain first matching result；Step S4 exports the image similar to input picture according to the size of similarity.Compared with the prior art the beneficial effects of the present invention are：The feature of extraction employs color, texture and the direction character of image, can more accurately express the true content of image, improves the performance of image retrieval.A kind of feature is proposed using group's sparse features selection strategy and selects excellent method, optimal characteristics can independently be selected to carry out characteristic matching, improves the precision of image indexing system.
Description
Technical field
The present invention relates to image retrieval technologies field more particularly to a kind of image search methods and dress based on rarefaction representation
It puts.
Background technology
With the rapid development of digital audio-effect processing and Internet technology, there is approximate number on internet with 10,000,000,000 image, and
And all increase daily with millions of speed.How to design one kind and user's needs are fast and effeciently retrieved from mass picture
The method of picture have huge realistic meaning, this is also the content of field of image search concern.Traditional text based
Image retrieval is born in the seventies in last century, and image is analyzed by artificial, and label character is carried out to picture material, into
And the search problem of image can be converted into technology maturation, efficient text retrieval problem.But traditional retrieval side
Not only descriptive power is limited for method, but also there is also subjectivities to the description of image.These limitations have been drawn later based on content
Image retrieval, this technology using some specific feature extraction algorithm abstract images low-level image feature formed a feature
Library, and then the feature for extracting query image is matched to find the most similar image with feature database.Based on content
Image retrieval does not need to artificial subjective analysis, and feature extraction and matching is realized automatically by computer, substantially increases image inspection
The efficiency of rope.
Although this technology has obtained extensive research, each major company and enterprise also all have developed the image retrieval system of oneself
System.But characteristic information included in a sub-picture is very abundant, the simple one or several kinds of feature combinations of extraction
It can not effectively represent image, and also bring for excessively extracting the redundancy of characteristic strip and choose to image indexing system
War.
In view of drawbacks described above, creator of the present invention is finally obtained the present invention by prolonged research and experiment.
Invention content
It is above-mentioned to overcome the purpose of the present invention is to provide a kind of image search method based on rarefaction representation and device
Technological deficiency.
To achieve the above object, the technical solution adopted by the present invention is：
On the one hand a kind of image search method based on rarefaction representation is provided, this method includes the following steps：
Step S1, input picture collection, the input picture concentrated to described image pre-process；
Step S2 using group's sparse features selection strategy, selects the input picture and the feature letter of image data base
Breath forms characteristics of image library；
Step S3, the feature in the feature of the input picture and described image database carry out specific measurement ratio
Compared with calculating similarity, obtain first matching result；
Step S4 exports the image similar to the input picture according to the size of the similarity.
Preferably, the step S1 includes the following steps：
Step S11, input picture collection；
Step S12 carries out size normalization to all input pictures that described image is concentrated；
Step S13 carries out image recovery processing to blurred picture in described image database；
Step S14, the input picture that described image concentration is not carried out to semantic tagger carry out semantic tagger so that every width institute
Stating input picture has corresponding keyword message；
Step S15 carries out conspicuousness detection, and shown using Remanent Model is composed to the image in described image database
Work property region.
Preferably, the step S2 includes the following steps：
Step S21, according to the keyword message of image data base by piece image in described image database and remaining
Image is compared, and if there is being more than 2/3rds keyword messages, then generates a similar sequence of the piece image
Row, become nearest sequence；If identical without keyword, a dissimilar sequence of the piece image is generated, into
For farthest sequence；
Step S22, by the salient region of image in the nearest sequence and the farthest sequence respectively with described first
The salient region of width image carries out Euclidean distance comparison, and m closest image is selected to form phase with the piece image
It is seemingly right, it selects m ranges dissimilar right from farthest image and piece image composition, obtains about the piece image
All similar pairs and dissmilarity it is right；
Step S23, to the second width image in described image database, piece image repeats above operation to the end successively, most
All similar pairs of described image database and dissmilarity are obtained eventually to image；
Step S24, by similar to assigning 1, dissmilarity obtains a similarity measurement vector Y to assigning -1；
Gained image is divided into two image columns L1 and L2 to row, feature is then extracted respectively, for same by step S25
Category feature connects to form eigenmatrix A and B, finally subtracts each other to form feature difference matrix X；
Step S26 according to the feature difference matrix X and the similarity measurement vector Y, utilizes the sparse logistic regression of group
Model calculates weight vectors w；
Step S27, it is to select that the feature in feature difference matrix X position is corresponded to according to non-zero group in the weight vectors w
The feature gone out, and then form characteristics of image library.
Preferably, the step S3 includes the following steps：
The Euclidean distance between the feature in the feature and described image database of the input picture is obtained in step S31；
The Euclidean distance obtained is normalized step S32.
Preferably, the step S1 includes the following steps：
Step S41 according to the color of the input picture, texture and the distance in direction, calculates final similarity；
Step S42 exports similar image according to the size of the final similarity.
Another aspect provides a kind of image retrieving apparatus based on rarefaction representation, which includes：
Input picture pretreatment unit, for input picture collection, the input picture concentrated to described image pre-processes；
Feature selection unit for using group's sparse features selection strategy, selects the input picture and image data
The characteristic information in library forms characteristics of image library；
Compare computing unit, carried out for the feature in the feature according to the input picture and described image database special
Fixed measurement compares, and calculates similarity, obtains first matching result；
Elementary area is exported, the image similar to the input picture is exported according to the size of the similarity.
Preferably, the input picture pretreatment unit includes：
Image input module, for input picture collection；
Image normalization module, all input pictures for being concentrated to described image carry out size normalization；
Image repair module, for carrying out image recovery processing to blurred picture in described image database；
Linguistic indexing of pictures module, the input picture for described image concentration not carried out to semantic tagger carry out semantic mark
Note so that input picture has corresponding keyword message described in every width；
Saliency detection module, for notable to the image progress in described image database using Remanent Model is composed
Property detection, and obtain salient region.
Preferably, the feature selection unit includes：
Sequence generating module, for the keyword message according to image data base by the first width figure in described image database
As being compared with remaining image, if there is being more than 2/3rds keyword messages, then the one of the piece image is generated
A similar sequences become nearest sequence；If identical without keyword, a dissmilarity of the piece image is generated
Sequence becomes farthest sequence；
Apart from comparison module, for by the salient region of image in the nearest sequence and the farthest sequence respectively and
The salient region of the piece image carries out Euclidean distance comparison, selects m closest image and the first width figure
It is similar right as forming, it selects m ranges dissimilar right with piece image composition from farthest image, obtains about described first
All similar pairs of width image and dissmilarity it is right；
Repetitive operation module, for successively to the second width image in described image database to the end piece image repeat with
Upper operation finally obtains all similar pairs of described image database and dissmilarity to image；
Similarity measurement vector forms module, dissimilar to imparting -1 for will be similar to assigning 1, obtain one it is similar
Property measuring vector Y；
Feature difference matrix forms module, for gained image to be divided into two image columns L1 and L2 to row, then distinguishes
Feature is extracted, connects to form eigenmatrix A and B for same class feature, finally subtracts each other to form feature difference matrix X；
Weight vector computation module, for according to the feature difference matrix X and the similarity measurement vector Y, utilizing
Group is sparse, and Logic Regression Models calculate weight vectors；
Characteristics of image library forms module, for corresponding to X, feature difference matrix according to non-zero group in the weight vectors w
The feature put is the feature selected, and then forms characteristics of image library.
Preferably, the relatively computing unit includes：
Euclidean distance computing module, for the feature that is obtained in the feature of the input picture and described image database it
Between Euclidean distance；
Normalized module, for the Euclidean distance obtained to be normalized.
Preferably, the output elementary area includes：
Final similarity calculation module for color, texture and the distance in direction according to the input picture, calculates
Go out final similarity；
Image module is exported, for exporting similar image according to the size of the final similarity.
Compared with the prior art the beneficial effects of the present invention are：A kind of image search method and dress based on rarefaction representation
It puts, superiority is embodied in：
(1) feature of extraction employs color, texture and the direction character of image, can more accurately express image
True content improves the performance of image retrieval.
(2) a kind of feature is proposed using group's sparse features selection strategy and selects excellent method, can independently select optimal spy
Sign carries out characteristic matching, improves the precision of image indexing system.
(3) conspicuousness detection has been carried out to image, has eliminated adverse effect of the contextual factor to retrieval.
(4) traditional content-based image retrieval is combined, using sparse representation theory, greatly increases characteristics of image
Extraction and matched efficiency.
(5) in image to there is no the image of keyword to carry out meaning automatic marking those in the process so that the skill
Art has more extensive adaptability.
Description of the drawings
Fig. 1 is a kind of flow chart of the image search method based on rarefaction representation provided by the invention；
Fig. 2 is the flow chart of step S1；
Fig. 3 is the flow chart of step S2；
Fig. 4 is the flow chart of step S3；
Fig. 5 is the flow chart of step S4；
Fig. 6 is a kind of functional block diagram of the image retrieving apparatus based on rarefaction representation provided by the invention；
Fig. 7 is the functional block diagram of input picture pretreatment unit；
Fig. 8 is characterized the functional block diagram of selecting unit；
Fig. 9 is the functional block diagram for comparing computing unit；
Figure 10 is the functional block diagram for exporting elementary area.
Specific embodiment
For ease of further understanding the technology contents of the present invention, the invention will be further described below in conjunction with the accompanying drawings.
Embodiment one
As shown in Figure 1, for a kind of flow chart of the image search method based on rarefaction representation provided by the invention, this method
Include the following steps：
Step S1, input picture collection, the input picture concentrated to described image pre-process.
Step S2 using group's sparse features selection strategy, selects the input picture and the feature letter of image data base
Breath forms characteristics of image library.
Step S3, the feature in the feature of the input picture and described image database carry out specific measurement ratio
Compared with calculating similarity, obtain first matching result.
Step S4 exports the image similar to the input picture according to the size of the similarity.
As shown in Fig. 2, the flow chart for step S1, step S1 includes the following steps：
Step S11, input picture collection.
Step S12 to all input pictures that described image is concentrated, including training image and test image, carries out size
Size normalization.
Step S13 carries out image recovery processing, specifically, being filtered using wiener to blurred picture in described image database
Wave method carries out image repair.
Step S14, the input picture that described image concentration is not carried out to semantic tagger carry out semantic tagger so that every width institute
Stating input picture has corresponding keyword message, specifically, carrying out linguistic indexing of pictures using across media correlation models.
Step S15 carries out conspicuousness detection, and shown using Remanent Model is composed to the image in described image database
Work property region.
As shown in figure 3, the flow chart for step S2, step S2 includes the following steps：
Step S21, according to the keyword message of image data base by piece image in described image database and remaining
Image is compared, and if there is being more than 2/3rds keyword messages, then generates a similar sequence of the piece image
Row, become nearest sequence；If identical without keyword, a dissimilar sequence of the piece image is generated, into
For farthest sequence.
Step S22, by the salient region of image in the nearest sequence and the farthest sequence respectively with described first
The salient region of width image carries out Euclidean distance comparison, and m closest image is selected to form phase with the piece image
It is seemingly right, it selects m ranges dissimilar right from farthest image and piece image composition, obtains about the piece image
All similar pairs and dissmilarity it is right.
Step S23, to the second width image in described image database, piece image repeats above operation to the end successively, most
All similar pairs of described image database and dissmilarity are obtained eventually to image.
Step S24, by similar to assigning 1, dissmilarity obtains a similarity measurement vector Y to assigning -1.
Gained image is divided into two image columns L1 and L2 to row, feature is then extracted respectively, for same by step S25
Category feature connects to form eigenmatrix A and B, finally subtracts each other to form feature difference matrix X.
Step S26 according to the feature difference matrix X and the similarity measurement vector Y, utilizes the sparse logistic regression of group
ModelCalculate weight vectors wT, wherein xiFor X's
I-th row, yi∈ { -1,1 }, c ∈ R represent intercept, and λ is regularization coefficient,Represent m of feature difference matrix X
The nonoverlapping crowd of G that different characteristic is formediNorm.
Step S27, it is to select that the feature in feature difference matrix X position is corresponded to according to non-zero group in the weight vectors w
The feature gone out, and then form characteristics of image library.
As shown in figure 4, the flow chart for step S3, step S3 includes the following steps：
The Euclidean distance between the feature in the feature and described image database of the input picture is obtained in step S31.
Step S32 the Euclidean distance obtained is normalized, D (x, y)new=D (x, y)/δ (D (x,
Y)), wherein δ (D (x, y)) represents the standard deviation of D (x, y), and D (x, y) represents the Euclidean distance between two vectors.
As shown in figure 5, the flow chart for step S4, step S4 includes the following steps：
Step S41 according to the color of the input picture, texture and the distance in direction, is utilized
D (x, y)=ω1D1(x,y)+ω2D2(x,y)+ω3D3(x, y) calculates final similarity.Wherein Di(x,y)
(i=1,2,3) the distance between the distance between color of image feature vector, texture feature vector and direction character are represented respectively
The distance between vector, ωi(i=1,2,3) corresponding weight coefficient is represented respectively.
Step S42 exports similar image according to the size of the final similarity.
Embodiment two
As shown in fig. 6, for a kind of functional block diagram of the image retrieving apparatus based on rarefaction representation provided by the invention.The dress
Put including：Input picture pretreatment unit 1, feature selection unit 2 compare computing unit 3 and output elementary area 4.Input figure
As pretreatment unit 1, for input picture collection, the input picture concentrated to described image pre-processes.Feature selection unit
2, for using group's sparse features selection strategy, the input picture and the characteristic information of image data base are selected, forms figure
As feature database.Compare computing unit 3, carried out for the feature in the feature according to the input picture and described image database
Specific measurement compares, and calculates similarity, obtains first matching result.Elementary area 4 is exported, according to the big of the similarity
The small output image similar to the input picture.
As shown in fig. 7, the functional block diagram for input picture pretreatment unit 1, input picture pretreatment unit 1 includes：Figure
As input module 11, image normalization module 12, image repair module 13, linguistic indexing of pictures module 14 and saliency inspection
Survey module 15.Image input module 11, for input picture collection.Image normalization module 12, for what is concentrated to described image
All input pictures including training image and test image, carry out size normalization.Image repair module 13, for pair
Blurred picture carries out image recovery processing in described image database.Specifically, carry out image repair using Wiener Filter Method.Figure
As semantic tagger module 14, the input picture for described image concentration not carried out to semantic tagger carries out semantic tagger so that
Input picture has corresponding keyword message described in every width.Specifically, carry out image, semantic mark using across media correlation models
Note.Saliency detection module 15, for using compose Remanent Model in described image database image carry out conspicuousness
Detection, and obtain salient region.
As shown in figure 8, being characterized the functional block diagram of selecting unit 2, feature selection unit 2 includes：Sequence generating module 21,
Apart from comparison module 22, repetitive operation module 23, similarity measurement vector form module 24, feature difference matrix forms module
25th, weight vector computation module 26 and characteristics of image library form module 27.Sequence generating module 21, for according to image data base
Keyword message piece image in described image database is compared with remaining image, if there is being more than 2/3rds
Keyword message, then a similar sequences of the piece image are generated, become nearest sequence；If there is no keyword phase
With, then a dissimilar sequence of the piece image is generated, becomes farthest sequence.Apart from comparison module 22, for inciting somebody to action
The salient region of the image salient region with the piece image respectively in the nearest sequence and the farthest sequence
Euclidean distance comparison is carried out, selects m closest image and the piece image composition similar right, selects m ranges from farthest
Image and piece image composition it is dissimilar right, obtain about all similar pairs of the piece image and dissimilar
It is right.Repetitive operation module 23, for successively to the second width image in described image database to the end piece image repeat more than
Operation finally obtains all similar pairs of described image database and dissmilarity to image.Similarity measurement vector forms module
24, for by similar, to assigning 1, dissmilarity to obtain a similarity measurement vector Y to assigning -1.Feature difference matrix is formed
Module 25 for gained image to be divided into two image columns L1 and L2 to row, then extracts feature, for same category feature respectively
Series connection forms eigenmatrix A and B, finally subtracts each other to form feature difference matrix X.Weight vector computation module 26, for according to institute
Feature difference matrix X and the similarity measurement vector Y are stated, weight vectors w is calculated using the sparse Logic Regression Models of group.Figure
As feature database formation module 27, for corresponding to the spy in feature difference matrix X position according to non-zero group in the weight vectors w
Sign is the feature selected, and then forms characteristics of image library.
As shown in figure 9, the functional block diagram to compare computing unit 3, compares computing unit 3 and includes：Euclidean distance calculates mould
Block 31 and normalized module 32.Euclidean distance computing module 31, for the feature of the input picture and the figure to be obtained
As the Euclidean distance between the feature in database.Normalized module 32, for the Euclidean distance obtained to be carried out
Normalized.
As shown in Figure 10, the functional block diagram for output elementary area 4, output elementary area 4 include：Final similarity meter
Calculate module 41 and output image module 42.Final similarity calculation module 41, for according to the input picture color, line
Reason and the distance in direction, calculate final similarity.Image module 42 is exported, for according to the big of the final similarity
Small output similar image.
A kind of image search method and device based on rarefaction representation provided by the invention, superiority are embodied in：
(1) feature of extraction employs color, texture and the direction character of image, can more accurately express image
True content improves the performance of image retrieval.
(2) a kind of feature is proposed using group's sparse features selection strategy and selects excellent method, can independently select optimal spy
Sign carries out characteristic matching, improves the precision of image indexing system.
(3) conspicuousness detection has been carried out to image, has eliminated adverse effect of the contextual factor to retrieval.
(4) traditional content-based image retrieval is combined, using sparse representation theory, greatly increases characteristics of image
Extraction and matched efficiency.
(5) in image to there is no the image of keyword to carry out meaning automatic marking those in the process so that the skill
Art has more extensive adaptability.
The foregoing is merely presently preferred embodiments of the present invention, is merely illustrative for the purpose of the present invention, and not restrictive
's.Those skilled in the art understands, many changes can be carried out to it in the spirit and scope limited in the claims in the present invention,
It changes or even equivalent, but falls in protection scope of the present invention.
Claims (10)
1. a kind of image search method based on rarefaction representation, which is characterized in that this method includes the following steps：
Step S1, input picture collection, the input picture concentrated to described image pre-process；
Step S2 using group's sparse features selection strategy, selects the input picture and the characteristic information of image data base, structure
Into characteristics of image library；
Step S3, the feature in the feature of the input picture and described image database carry out specific measurement and compare,
Similarity is calculated, obtains first matching result；
Step S4 exports the image similar to the input picture according to the size of the similarity.
A kind of 2. image search method based on rarefaction representation according to claim 1, which is characterized in that the step S1
Include the following steps：
Step S11, input picture collection；
Step S12 carries out size normalization to all input pictures that described image is concentrated；
Step S13 carries out image recovery processing to blurred picture in described image database；
Step S14, the input picture that described image concentration is not carried out to semantic tagger carry out semantic tagger so that defeated described in every width
Entering image has corresponding keyword message；
Step S15 carries out conspicuousness detection, and obtain conspicuousness using Remanent Model is composed to the image in described image database
Region.
A kind of 3. image search method based on rarefaction representation according to claim 2, which is characterized in that the step S2
Include the following steps：
Step S21, according to the keyword message of image data base by piece image and remaining image in described image database
It is compared, if there is being more than 2/3rds keyword messages, then generates a similar sequences of the piece image, into
For nearest sequence；If identical without keyword, a dissimilar sequence of the piece image is generated, becomes farthest
Sequence；
Step S22, by the salient region of image in the nearest sequence and the farthest sequence respectively with the first width figure
The salient region of picture carries out Euclidean distance comparison, selects m closest image and the piece image composition similar right,
It selects m ranges dissimilar right with piece image composition from farthest image, obtains about all of the piece image
Similar pair and dissmilarity it is right；
Step S23, to the second width image in described image database, piece image repeats above operation to the end successively, final
To all similar pairs of described image database and dissmilarity to image；
Step S24, by similar to assigning 1, dissmilarity obtains a similarity measurement vector Y to assigning -1；
Gained image is divided into two image columns L1 and L2 to row, then extracts feature respectively, for same class spy by step S25
Sign series connection forms eigenmatrix A and B, finally subtracts each other to form feature difference matrix X；
Step S26 according to the feature difference matrix X and the similarity measurement vector Y, utilizes the sparse Logic Regression Models of group
Calculate weight vectors w；
Step S27, it is what is selected that the feature in feature difference matrix X position is corresponded to according to non-zero group in the weight vectors w
Feature, and then form characteristics of image library.
A kind of 4. image search method based on rarefaction representation according to claim 3, which is characterized in that the step S3
Include the following steps：
The Euclidean distance between the feature in the feature and described image database of the input picture is obtained in step S31；
The Euclidean distance obtained is normalized step S32.
A kind of 5. image search method based on rarefaction representation according to claim 4, which is characterized in that the step S1
Include the following steps：
Step S41 according to the color of the input picture, texture and the distance in direction, calculates final similarity；
Step S42 exports similar image according to the size of the final similarity.
6. a kind of image retrieving apparatus based on rarefaction representation, which is characterized in that the device includes：
Input picture pretreatment unit, for input picture collection, the input picture concentrated to described image pre-processes；
Feature selection unit for using group's sparse features selection strategy, selects the input picture and image data base
Characteristic information forms characteristics of image library；
Compare computing unit, carried out for the feature in the feature according to the input picture and described image database specific
Measurement compares, and calculates similarity, obtains first matching result；
Elementary area is exported, the image similar to the input picture is exported according to the size of the similarity.
A kind of 7. image retrieving apparatus based on rarefaction representation according to claim 6, which is characterized in that the input figure
As pretreatment unit includes：
Image input module, for input picture collection；
Image normalization module, all input pictures for being concentrated to described image carry out size normalization；
Image repair module, for carrying out image recovery processing to blurred picture in described image database；
Linguistic indexing of pictures module, the input picture for described image concentration not carried out to semantic tagger carry out semantic tagger,
So that input picture has corresponding keyword message described in every width；
Saliency detection module, for using compose Remanent Model in described image database image carry out conspicuousness inspection
It surveys, and obtains salient region.
A kind of 8. image retrieving apparatus based on rarefaction representation according to claim 7, which is characterized in that the feature choosing
Unit is selected to include：
Sequence generating module, for the keyword message according to image data base by piece image in described image database with
Remaining image is compared, and if there is being more than 2/3rds keyword messages, then generates a phase of the piece image
Like sequence, become nearest sequence；If identical without keyword, a dissimilar sequence of the piece image is generated
Row, become farthest sequence；
Apart from comparison module, for by the salient region of the nearest sequence and image in the farthest sequence respectively with it is described
The salient region of piece image carries out Euclidean distance comparison, selects m closest image and the piece image group
Into similar right, select m ranges dissimilar right from farthest image and piece image composition, obtain about the first width figure
All similar pairs of picture and dissmilarity it is right；
Repetitive operation module, for piece image to repeat above grasp to the end to the second width image in described image database successively
Make, finally obtain all similar pairs of described image database and dissmilarity to image；
Similarity measurement vector forms module, for by similar, to assigning 1, dissmilarity to obtain a similarity measurements to assigning -1
Measure vector Y；
Feature difference matrix forms module, for gained image to be divided into two image columns L1 and L2 to row, then extracts respectively
Feature connects for same class feature to form eigenmatrix A and B, finally subtracts each other to form feature difference matrix X；
Weight vector computation module, it is dilute using group for according to the feature difference matrix X and the similarity measurement vector Y
Thin Logic Regression Models calculate weight vectors；
Characteristics of image library forms module, for corresponding to feature difference matrix X position according to non-zero group in the weight vectors w
Feature be the feature selected, and then form characteristics of image library.
9. a kind of image retrieving apparatus based on rarefaction representation according to claim 8, which is characterized in that described relatively to count
Unit is calculated to include：
Euclidean distance computing module, for being obtained between the feature in the feature of the input picture and described image database
Euclidean distance；
Normalized module, for the Euclidean distance obtained to be normalized.
A kind of 10. image retrieving apparatus based on rarefaction representation according to claim 9, which is characterized in that the output
Elementary area includes：
Final similarity calculation module for color, texture and the distance in direction according to the input picture, calculates most
Whole similarity；
Image module is exported, for exporting similar image according to the size of the final similarity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611123413.4A CN108182188A (en) | 2016-12-08 | 2016-12-08 | A kind of image search method and device based on rarefaction representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611123413.4A CN108182188A (en) | 2016-12-08 | 2016-12-08 | A kind of image search method and device based on rarefaction representation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108182188A true CN108182188A (en) | 2018-06-19 |
Family
ID=62544843
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611123413.4A Pending CN108182188A (en) | 2016-12-08 | 2016-12-08 | A kind of image search method and device based on rarefaction representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108182188A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800818A (en) * | 2019-01-25 | 2019-05-24 | 宝鸡文理学院 | A kind of image meaning automatic marking and search method and system |
| CN116188822A (en) * | 2023-04-28 | 2023-05-30 | 青岛尘元科技信息有限公司 | Image similarity judging method, device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810252A (en) * | 2014-01-21 | 2014-05-21 | 南京信息工程大学 | Image retrieval method based on group sparse feature selection |
| US9082040B2 (en) * | 2011-05-13 | 2015-07-14 | Microsoft Technology Licensing, Llc | Identifying visual contextual synonyms |
| CN106204415A (en) * | 2015-05-04 | 2016-12-07 | 南京邮电大学 | A kind of novel method for registering images |
-
2016
- 2016-12-08 CN CN201611123413.4A patent/CN108182188A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9082040B2 (en) * | 2011-05-13 | 2015-07-14 | Microsoft Technology Licensing, Llc | Identifying visual contextual synonyms |
| CN103810252A (en) * | 2014-01-21 | 2014-05-21 | 南京信息工程大学 | Image retrieval method based on group sparse feature selection |
| CN106204415A (en) * | 2015-05-04 | 2016-12-07 | 南京邮电大学 | A kind of novel method for registering images |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109800818A (en) * | 2019-01-25 | 2019-05-24 | 宝鸡文理学院 | A kind of image meaning automatic marking and search method and system |
| CN116188822A (en) * | 2023-04-28 | 2023-05-30 | 青岛尘元科技信息有限公司 | Image similarity judging method, device, electronic equipment and storage medium |
| CN116188822B (en) * | 2023-04-28 | 2023-08-18 | 青岛尘元科技信息有限公司 | Image similarity judging method, device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tonioni et al. | A deep learning pipeline for product recognition on store shelves | |
| US8935246B2 (en) | Identifying textual terms in response to a visual query | |
| Rossetto et al. | Deep learning-based concept detection in vitrivr | |
| CN111914107B (en) | Instance retrieval method based on multi-channel attention area expansion | |
| EP3166020A1 (en) | Method and apparatus for image classification based on dictionary learning | |
| Rossetto et al. | IMOTION—a content-based video retrieval engine | |
| Niu et al. | Spatial-DiscLDA for visual recognition | |
| Abdul-Rashid et al. | Shrec’18 track: 2d image-based 3d scene retrieval | |
| CN110008365B (en) | Image processing method, device and equipment and readable storage medium | |
| Wi et al. | Poster-based multiple movie genre classification using inter-channel features | |
| Cai et al. | BTBUFood-60: Dataset for object detection in food field | |
| Li et al. | Multi-label pattern image retrieval via attention mechanism driven graph convolutional network | |
| Sitaula et al. | Hdf: hybrid deep features for scene image representation | |
| Jin et al. | The Open Brands Dataset: Unified brand detection and recognition at scale | |
| CN108182188A (en) | A kind of image search method and device based on rarefaction representation | |
| CN108717436B (en) | Commodity target rapid retrieval method based on significance detection | |
| Costache et al. | Categorization based relevance feedback search engine for earth observation images repositories | |
| Zhao et al. | Big transfer learning for fine art classification | |
| Wang et al. | A saliency detection based unsupervised commodity object retrieval scheme | |
| Ulges et al. | Tubetagger-youtube-based concept detection | |
| Wang et al. | Occluded person re-identification based on differential attention siamese network | |
| Seddati et al. | Triplet networks feature masking for sketch-based image retrieval | |
| Meng et al. | A multi-level weighted representation for person re-identification | |
| Kuo et al. | Feature learning with rank-based candidate selection for product search | |
| Ma et al. | Scene image retrieval with siamese spatial attention pooling |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180619 |
|
| RJ01 | Rejection of invention patent application after publication |