CN103336835B - Image retrieval method based on weight color-sift characteristic dictionary - Google Patents
Image retrieval method based on weight color-sift characteristic dictionary Download PDFInfo
- Publication number
- CN103336835B CN103336835B CN201310294385.2A CN201310294385A CN103336835B CN 103336835 B CN103336835 B CN 103336835B CN 201310294385 A CN201310294385 A CN 201310294385A CN 103336835 B CN103336835 B CN 103336835B
- Authority
- CN
- China
- Prior art keywords
- image
- color
- sift
- feature
- weights
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 65
- 239000013598 vector Substances 0.000 claims description 34
- 239000000284 extract Substances 0.000 claims description 23
- 238000000605 extraction Methods 0.000 claims description 12
- 238000005119 centrifugation Methods 0.000 claims description 9
- 230000008878 coupling Effects 0.000 claims description 8
- 238000010168 coupling process Methods 0.000 claims description 8
- 238000005859 coupling reaction Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000007689 inspection Methods 0.000 claims description 2
- 238000013519 translation Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 6
- 230000002452 interceptive effect Effects 0.000 description 3
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 241000406668 Loxodonta cyclotis Species 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 241000894007 species Species 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an image retrieval method based on a weight color-sift characteristic dictionary. The image retrieval method based on the weight color-sift characteristic dictionary comprises the following steps that training images are selected in images to be retrieved randomly, the edges of the training images are extracted, the color-sift characteristics of edge points of all the training images are extracted, and a characteristic dictionary is constructed according to the color-sift characteristics, an image which needs to be retrieved is input and the color-sift characteristics of the retrieval image and the edge points of the image to be retrieved are extracted, and weight histogram characteristics of the retrieval image and the image to be retrieved are extracted based on the characteristic dictionary; similarity matching based on the weight histogram characteristics is conducted on the retrieval image and the images to be retrieved in the database based on the weight histogram characteristics; whether all the images to be retrieved in the database are all traversed is detected,, the result is matched according to similarity and the image searching result is displayed if all the images to be retrieved in the database are all traversed, and the similarity matching is conducted again if all the images to be retrieved in the database are not traversed. The image retrieval method based on the weight color-sift characteristic dictionary improves accuracy and callback rate during searching of a large-scaled image database, has dimension invariance, translation invariance and rotation invariance, and has the advantages that the image retrieval method based on the color-sift characteristic dictionary has locality, particularity, multi-amount property and high-efficiency property.
Description
Technical field
The invention belongs to image retrieval technologies field, specifically a kind of image based on weights color-sift characteristics dictionary
Search method, based on picture material, the process with realizing, image being analyzed based on image characteristics extraction and retrieving it is allowed to
User input one or more picture, to search other pictures with same or similar content.
Background technology
Image is a kind of similitude, vividness the description to objective objects or description.Image is objective right in other words
A kind of expression of elephant, it contains and is described object for information about.It is the topmost information source of people.According to statistics, one
The information about 75% that people obtains is derived from vision.As the saying goes " it is better to see once than hear a hundred times ", " very clear ", all reflect image and exist
Effect unique in information transmission.How rapidly and accurately to extract the step that picture material is image retrieval most critical.SIFT is
David G Lowe2004 summarize invariant technology characteristic detection method, proposition a kind of to metric space, image scaling,
Rotation even affine constant image local feature describes operator.So-called image content-based retrieval, searches from image library
Image containing specific objective, also includes retrieving the video segment containing specific objective from continuous video image.It is distinguished
In traditional image retrieval means, the present invention proposes a kind of image retrieval side based on weights color-sift characteristics dictionary
Method, has merged weights color-sift characteristics dictionary technology, such that it is able to provide more effective retrieval method.Can be applicable to numeral
Library, medical diagnosis, image classification, WEB related application, public safety and crime survey etc.
The patent application " a kind of image search method based on sketch feature extraction " that university of Tsing-Hua University proposes（Patent Shen
Please numbers 201110196051.2, publication number 201110196051.2）Disclose a kind of image retrieval based on sketch feature extraction
Method, is related to field of image search.Methods described includes step：Extract training feature vector, obtain feature lexicon；Extract input
Characteristic vector, obtains input feature value collection, carries out counting operation to feature lexicon, obtains input feature vector frequency vector, and then
Obtain Intersted word and non-Intersted word；Extract searching characteristic vector, obtain searching characteristic vector collection, and then retrieved
Characteristic frequency vector；And then obtain interest retrieval character frequency vector, non-interest retrieval character frequency vector, interest input feature vector
Frequency vector and non-interest input feature vector frequency vector；And then calculate the similarity that input sketch retrieves sketch with each, export
Retrieval result.The method has efficiency and the degree of accuracy that good user interactivity improves image retrieval, but intrinsic dimensionality
Larger it is contemplated that interest and two kinds of features of non-interest, lead to the recall precision when being applied to large database relatively low, speed is relatively
Slowly.
" a kind of combination user evaluates the interactive image retrieval side with mark for the patent application that Communication University of China proposes
Method "（Number of patent application 201310128036.3, publication number CN103164539A）Disclose a kind of evaluation and mark with reference to user
Interactive image retrieval method, belong to multimedia information retrieval field.This process employs physical features based on image and
The integrated retrieval method that text combines, it is allowed to user carries out text message description to query image in retrieving, or
The keyword that selection system provides, by carrying out " satisfaction " or the relevant evaluation of " being unsatisfied with ", image retrieval system to retrieval result
System is automatically satisfied with image and carries out text mark to the correlation of user's mark, forms high-layer semantic information；Constantly making with user
With this system can generate abundant semantic information database.In view of different user to same picture, same user's different time
Difference to same picture text marking, the present invention combines the credible of user during generative semantics information database
Degree.When entering line retrieval, integrated retrieval mode that the query image that there is semantic information is combined using feature based and text
Enter line retrieval, improve the degree of accuracy of retrieval result.Although the method combines, user evaluates and the interactive image of mark is examined
Rope mode, obtains high-layer semantic information, improves the accuracy rate of real-time retrieval, but when being applied to large-scale image data base,
Because the similarity measure of all kinds of images is numerous and diverse, process manually numerous and diverse mark and improve computation complexity, reduce the inspection of image
Rope efficiency, leads to the readjustment rate of return retrieval set and recall ratio not high.
Content of the invention
The present invention is directed to above-mentioned the deficiencies in the prior art, proposes a kind of image based on weights color-sift characteristics dictionary
Search method, effectiveness of retrieval, speed and readjustment rate when improve application large database.
A kind of image search method based on weights color-sift characteristics dictionary, it includes,
Randomly select training image in image to be retrieved, extract the edge of described training image, and extract all training
The color-sift feature of image border point, and with described color-sift feature construction characteristics dictionary；
Input image to be retrieved simultaneously extracts the color-sift feature retrieving image and image border to be retrieved point, is based on
Described characteristics dictionary is to retrieval image and image zooming-out weights histogram feature to be retrieved；To be checked in retrieval image database
Rope image carries out the similitude coupling based on weights histogram feature；
Detect whether to travel through the image whole to be retrieved in all databases, if so, then according to similitude matching result, show
Show image searching result, if it is not, then re-starting similitude coupling.
On the basis of technique scheme, image to be retrieved randomly selects training image and includes in image to be retrieved
In database, l is randomly selected for every class image and open image composition training image database.
On the basis of technique scheme, the edge extracting described training image includes：Extract described training image
Edge includes randomly selecting a training image in training image database and does greyscale transformation, and passes through Steerable Filter
Processed, selection two-dimensional Gaussian function is filter kernel function, obtains the energy function W on 2L direction of each pixelσ
(x, y, θ), and the notable pixel in edge of image is extracted through threshold decision, wherein L represents the number in direction, x and y represents picture
The coordinate value of vegetarian refreshments, θ is the value in direction, and scope is 0～2 π, is spaced apart π/L.
On the basis of technique scheme, the color-sift feature extracting all training image marginal points includes：Root
According to the notable pixel in the edge of selected training image to former training coloured image respectively in redness-R, green-G and blueness-B three
Individual passage extracts color-sift feature, obtain redness-R, the green-G of the notable pixel in each edge of this image and blueness-
Color-sift feature dec of tri- passages of Br(e)、decg(e) and decbE (), e represents e-th marginal point of this image, e=
1,2 ..., E, E are the sum of the notable pixel in all edges of this image.
On the basis of technique scheme, the described color-sift feature bag extracting all training image marginal points
Include：Color-sift feature extraction to the notable pixel in all edges of width training image each in training image database, time
Go through all images in training image database, its feature is followed successively by dec in three Color Channelsr,m(em)、decg,m(em) and
decb,m(em), m=1,2 ..., M, M are training image Database size, emRepresent that m opens the e of training imagemIndividual edge picture
Vegetarian refreshments, em=1,2 ..., Em, EmOpen the sum of the notable pixel in all edges of training image for m.
On the basis of technique scheme, construction feature dictionary step includes all edges to whole training images and shows
Write the redness-R passage green-G passage of pixel and color-sift feature dec of blueness-channel Br,m(em), decg,m(em)
And decb,m(em), by K-means cluster calculation, take K cluster centre obtain redness-R passage, green-G passage and blueness-
The w row of channel B, the two dimensional character dictionary cod of K rowr、codgAnd codb, wherein K is the number of K-means cluster centre, i.e. word
The size of allusion quotation.
On the basis of technique scheme, described weights histogram feature X contains redness-R, green-G and blueness-B
Three-channel weights histogram feature, carries out the similitude based on weights histogram feature retrieval image and image to be retrieved
Join, calculate 2 norm similarity distances and obtain Di(X,X′i).
On the basis of technique scheme, to retrieval Image Coding be calculated retrieval image based on color-sift
The weights histogram feature of characteristics dictionary, comprises the following steps：
1）Color-sift feature sec of the notable pixel in retrieval image all edge on red channel RrO (), calculates
secrO () corresponds to red channel two dimensional character dictionary codrK cluster centre Euclidean distance, when chosen distance value is minimum
Cluster centre as the affiliated center of this marginal point, single order is carried out to all edges significant point and obtains retrieval figure away from statistical computation
As corresponding to two dimensional character dictionary cod in red channelrFrequency histogram hisr；
2）Assume to fall in k-th cluster centre qkIndividual edge significant point, the q to cluster to k-th cluster centrekIndividual edge
Significant point calculates centrifugation weights li (k) for this cluster centre for this significant point, calculates all significant points of this cluster centre maximum
Centrifugation weights obtain weight vector α (k) of this cluster centre, the frequency histogram his to retrieval imagerWith corresponding weights to
Amount α (k) is done matrix point multiplication operation and is obtained retrieving the weight vector hst of imager, i.e. corresponding element multiplication, hstrFor one K dimension row to
Amount, k represents k-th cluster centre, and value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size；
3) do equally in the calculating of red channel R in green channel G, blue channel B, finally give image green channel G
Weight vector hstgWeight vector hst with blue channel Bb, conformity calculation is carried out to the weight vector in three passages and obtains
The weights histogram feature X based on color-sift characteristics dictionary of retrieval image.
On the basis of technique scheme, the described q to cluster to k-th cluster centrekIndividual edge significant point meter
Calculate centrifugation weights li (k) for this cluster centre for this significant point, calculate the maximum centrifugation power of all significant points of this cluster centre
It is worth to weight vector α (k) of this cluster centre, calculated using equation below：
α (k)=max (li (u, k)),
Wherein, u represents to fall in u-th edge significant point of k-th cluster centre, and value is 1,2 ..., qk, k represents kth
Individual cluster centre, value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size, qkRepresent to fall in k-th cluster centre
Edge significant point sum.
With respect to prior art, use direction tunable filter of the present invention carries out edge extracting, can effectively judge
The sensing of each pixel point edge principal direction, then just can fast and accurately extract the edge picture of image by threshold determination
Vegetarian refreshments information, can fast and accurately carry out the feature extraction of next step by extracting the image edge pixels point information obtaining,
Improve the speed being applied to retrieve when real time human-machine interaction and large-scale image data base and accuracy.Employ extraction edge side
The search strategy combining to pixel color-sift feature and encoder dictionary, extracts the color- of the notable pixel in edge
Sift feature simultaneously calculates weights histogram feature, to image by the encoder dictionary based on weights color-sift feature construction
Represent more there is typicalness, can significantly more efficient represent image feature difference, when being applied in retrieving, carry
High it is applied to accuracy rate during large-scale Image-Database Retrieval and readjustment rate.Employ the color- based on tri- passages of RGB
The image search method of sift characteristics dictionary, it is a kind of multiple dimensioned image retrieval algorithm, piece image is converted into multiple
The set of feature, then be compared and obtain a result and then realize image by calculating the Euclidean distance between two width image feature vectors
Search function. experimental result illustrates that this algorithm has yardstick, translation, rotational invariance, can carry out applications well.It is simultaneously
A kind of to metric space, image scaling, rotation image local feature describes operator.It has locality, different property, volume and height
The features such as effect property.Sift feature extraction algorithm can process and translates between two width images, rotates, in the case of affine transformation
Matching problem, improves accuracy rate and the readjustment rate of image retrieval.
Brief description
Fig. 1 is the flow chart of the present invention.
It is embodied as measure
Below in conjunction with the accompanying drawings invention is described further.
Embodiment 1
The realization of the image search method based on weights color-sift characteristics dictionary of the present invention with reference to Fig. 1, be given as
Lower specific embodiment：
Step 1：In image data base to be retrieved, l is randomly selected for every class image and open image composition training image data
Storehouse, this example uses Corel-1000 image data base, needs to retrieve same type in Corel-1000 image data base
Image, image library includes 10 class images, and each class includes 100 images, and in this example, l value is 10, and 10 class altogether is chosen altogether
100 training images.
Step 2：Randomly select a training image and do greyscale transformation in training image database, by the adjustable filter in direction
Ripple device is processed, and selection two-dimensional Gaussian function is filter kernel function, chooses suitable wave filter sliding window size, obtains
Energy function W on 2L direction of each pixelσ(x, y, θ), extracts the notable pixel in edge of image through threshold decision,
L represents the number in direction, and L value represents the coordinate value of pixel for 6, x and y, and σ is filter scales parameter, and σ value is 1, θ
For the value in direction, scope is 0～2 π, is spaced apart π/L, and in this example, θ is taken as 0, π/6 ..., 11 π/6,2 π.
Step 3：The notable pixel in edge according to selected training image to former training coloured image respectively redness-R,
Green-G and three passages of blueness-B extract color-sift feature, obtain the red of the notable pixel in each edge of this image
Color-R, green-G and color-sift feature dec of three passages of blueness-Br(e)、decg(e) and decbE (), e represents this figure
E-th marginal point of picture, e=1,2 ..., E, E are the sum of the notable pixel in all edges of this training image.
Step 4：Each image is carried out to width training image execution step 2 step 3 each in training image database
The color-sift feature extraction of the notable pixel in all edges, all images in traversal training image database, its feature
It is followed successively by dec in three Color Channelsr,m(em)、decg,m(em) and decb,m(em), m=1,2 ..., M, M are training image
Database size, emRepresent that m opens the e of training imagemIndividual edge pixel point, em=1,2 ..., Em, EmOpen training figure for m
Sum as the notable pixel in all edges.
Step 5：Color-sift feature to the redness-R passage of the notable pixel in all edges of whole training images
decr,m(em), by K-means cluster calculation, take K cluster centre to obtain the w row of redness-R passage, the two dimensional character of K row
Dictionary codr, K is the number of K-means cluster centre, i.e. the size of dictionary, and same method is in green-G passage and blueness-B
In passage, respectively to decg,m(em) and decb,m(em) the execution same calculating of redness-R passage respectively obtains green-G passage w
Row, the two dimensional character dictionary cod of K rowgThe two dimensional character dictionary cod arranging with the w row of blueness-channel B, Kb, in this example, w is sift
Intrinsic dimensionality size is 128, K value is 500, and that is, dictionary size is 500.
Step 6：Input retrieval image, is in bus class image, for retrieval image execution step 2 step
Color-sift feature sec of 3 same extraction image border significant point redness-R, green-G and three passages of blueness-Br(e)、
secg(e) and secb(e), the characteristics dictionary cod being obtained by step 5r、codgAnd codbCarry out coding and be calculated retrieval image
The weights histogram feature X based on color-sift characteristics dictionary, weights histogram feature X contains redness-R, green-G
With blueness-B three-channel weights histogram feature；
6a）Color-sift feature sec of the notable pixel in retrieval image all edge on red channel Rr(o), meter
Calculate secrO () corresponds to red channel two dimensional character dictionary codrK cluster centre Euclidean distance, chosen distance value is minimum
When cluster centre as the affiliated center of this marginal point, single order is carried out to all edges significant point and is retrieved away from statistical computation
Image corresponds to two dimensional character dictionary cod in red channelrFrequency histogram hisr；
6b）Assume to fall in k-th cluster centre qkIndividual edge significant point, the q to cluster to k-th cluster centrekIndividual edge
Significant point calculates centrifugation weights li (k) for this cluster centre for this significant point, calculates all significant points of this cluster centre maximum
Centrifugation weights obtain weight vector α (k) of this cluster centre, calculated using equation below：
α (k)=max (li (u, k)),
Wherein, u represents to fall in u-th edge significant point of k-th cluster centre, and value is 1,2 ..., qk, k represents kth
Individual cluster centre, value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size, qkRepresent to fall in k-th cluster centre
Edge significant point sum.Frequency histogram his to retrieval imagerDo matrix point multiplication operation with corresponding weight vector α (k)
Obtain retrieving the weight vector hst of imager, i.e. corresponding element multiplication, hstrFor a K dimensional vector, k represents in k-th cluster
The heart, value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size；
6c) do equally in the calculating of red channel R in green channel G, blue channel B, finally give image green channel G
Weight vector hstgWeight vector hst with blue channel Bb, conformity calculation is carried out to the weight vector in three passages and obtains
The weights histogram feature X based on color-sift characteristics dictionary of retrieval image.
Step 7：Extract an image execution step to be retrieved from the image data base to be retrieved that total number of images size is S
2 steps 4 carry out the color-sift feature extraction of the notable pixel in all edges of each image, then execution step 6
To each image to be retrieved weights histogram feature X ' based on weights color-sift characteristics dictionaryi, travel through view data
All images in storehouse, i=1,2 ..., S, S are total number of images to be retrieved, and used in this example, database is Corel-1000,
Including 10 classes, each class includes 100 images, and the value of S is 1000..
Step 8：Retrieval image and image to be retrieved are carried out the similitude coupling based on weights histogram feature, calculates 2
Norm similarity distance obtains Di(X,X′i).
Step 9：For every image to be retrieved according to its Di(X,X′i) value carry out from small to large order arrangement, display
Wherein front n opens the result that image is retrieval, i=1, and 2 ..., S, S are total number of images to be retrieved, and n is to return retrieval picture number
Mesh, value is the positive integer of artificially autonomous determination.In this example, n value is 20.The present invention is successfully from Corel-1000 picture number
According in be retrieved the 20 width bus class image related with bus image exactly, but with regard to this, retrieval rate is
100%.
The content of image is quickly and accurately described and always is the emphasis of research and difficult point in image retrieval technologies.
Traditional image characteristic extracting method, substantially around the color of image, texture, shape and spatial relationship are launching.This
Invention randomly selects training image in training image database first and does greyscale transformation, at Steerable Filter
Reason, the notable pixel in edge according to whole training images to former training coloured image respectively redness-R, green-G and blueness-
Tri- passages of B extract color-sift features, obtain the notable pixel in each edge of training image redness-R, green-G and
The color-sift feature of three passages of blueness-B.Carry out coding to retrieval image and image feature based dictionary to be retrieved to calculate
Obtain the weights histogram based on color-sift characteristics dictionary, retrieval image and image to be retrieved are carried out based on weights Nogata
The similitude coupling of figure feature, obtains retrieval result, improves efficiency, speed and the readjustment rate of retrieving.
The image search method based on weights color-sift characteristics dictionary for the embodiment 2 is with embodiment 1
This example equally chooses Corel-1000 image data base, and image data base includes 10 class images, each class bag
Include 100 images, the same retrieving of embodiment 1 is executed to each image in database, calculate when return retrieval figure
As number n is the average retrieval accuracy rate of each class and the average retrieval of whole 10 1000 images of class in whole 10 classes when 20
Accuracy rate, to retrieval result statistics and list, and and state of the art in several known to search method such as Jhanwar,
Method that Hung is proposed and the method based on color-texture-shape, the method based on SIFT-BOF and be based on SIFT-
The method of SPM is contrasted, and comparing result is as shown in table 1.As seen from Table 1, the present invention's in return retrieval picture number n is
When 20 the average retrieval accuracy rate of whole 10 1000 images of class apparently higher than above-mentioned each be used for contrast search method, and
It is higher than the search method being mostly used in contrast in the average retrieval accuracy rate of each 100 images of class in whole 10 classes.Cause
This, the present invention when being applied to different classes of image and entering line retrieval, the higher average retrieval accuracy rate that all can take it is adaptable to
The image retrieval of the more large-scale view data of image species, and relatively stablize, preferably averagely examine for each class is all available
Rope accuracy rate.
Table 1
It is more than two examples of the present invention, do not constitute any limitation of the invention, emulation experiment shows, the present invention
Speed can not only be improve when application is with large-scale image data base, it is higher accurate also to enable to have for retrieval result
Rate and readjustment rate.
To sum up, the image search method based on weights color-sift characteristics dictionary of the present invention, is directed generally to existing
Technology is applied to the raising of speed during large-scale image data base, accuracy rate and readjustment rate.Its method and step is：In image to be retrieved
In randomly select training image, greyscale transformation is done to image；Training image is processed by Steerable Filter；Bonding position
The result of tunable filter extracts the edge of image；Extract the color-sift feature of all training images；By to all instructions
The color-sift feature practicing the edge pixel point of image carries out K-means cluster construction feature dictionary；Input figure to be retrieved
Picture simultaneously executes the step extraction color-sift feature being same as training image to retrieval image and image to be retrieved；To retrieval image
It is based on color-sift characteristics dictionary with image to be retrieved and extract weights histogram feature；To be checked in retrieval image database
Rope image carries out the similitude coupling based on weights histogram feature；According to similitude matching result display image retrieval result.
Particularly with the large-scale Image-Database Retrieval present invention, the present invention has that retrieval rate is fast, accuracy rate and higher excellent of readjustment rate
Gesture, can be applicable to real time human-machine interaction and the image retrieval of large-scale image data base.
Claims (8)
1. a kind of image search method based on weights color-sift characteristics dictionary it is characterised in that：It includes,
Randomly select training image in image to be retrieved, extract the edge of described training image, and extract all training images
The color-sift feature of marginal point, and with described color-sift feature construction characteristics dictionary；
Input image to be retrieved simultaneously extracts the color-sift feature retrieving image and image border to be retrieved point, based on described
Characteristics dictionary is to retrieval image and image zooming-out weights histogram feature to be retrieved；Figure to be retrieved in retrieval image database
As carrying out the similitude coupling based on weights histogram feature；
Detect whether to travel through the image whole to be retrieved in all databases, if so, then according to similitude matching result, display figure
As retrieval result, if it is not, then re-starting similitude coupling；
Retrieval Image Coding is calculated with the weights histogram feature based on color-sift characteristics dictionary of retrieval image, bag
Include following steps：
1) color-sift feature sec of the notable pixel in retrieval image all edge on red channel RrO (), calculates secr
O () corresponds to red channel two dimensional character dictionary codrK cluster centre Euclidean distance, poly- when chosen distance value is minimum
Class center as the affiliated center of this marginal point, all edges significant point is carried out single order away from statistical computation obtain retrieve image exist
Red channel corresponds to two dimensional character dictionary codrFrequency histogram hisr；
2) assume to fall in k-th cluster centre qkIndividual edge significant point, the q to cluster to k-th cluster centrekIndividual edge is notable
Point calculates centrifugation weights li (k) for this cluster centre for this significant point, calculate all significant points of this cluster centre maximum from
Heart weights obtain weight vector α (k) of this cluster centre, the frequency histogram his to retrieval imagerWith corresponding weight vector α
K () is done matrix point multiplication operation and is obtained retrieving the weight vector hst of imager, i.e. corresponding element multiplication, hstrFor a K dimensional vector, k
Represent k-th cluster centre, value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size；
3) do equally in the calculating of red channel R in green channel G, blue channel B, finally give the power of image green channel G
The vectorial hst of valuegWeight vector hst with blue channel Bb, conformity calculation is carried out to the weight vector in three passages and is retrieved
The weights histogram feature X based on color-sift characteristics dictionary of image.
2. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 1, its feature exists
In：Randomly select training image to include in image data base to be retrieved, every class image being randomly selected in image to be retrieved
L opens image composition training image database.
3. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 2, its feature exists
In the edge extracting described training image includes：Extract described training image edge include in training image database with
Machine is chosen a training image and is done greyscale transformation, and is processed by Steerable Filter, chooses two-dimensional Gaussian function and is
Filter kernel function, obtains the energy function W on 2L direction of each pixelσ(x, y, θ), and extract figure through threshold decision
The notable pixel in edge of picture, wherein L represents the number in direction, x and y represents the coordinate value of pixel, and θ is the value in direction, model
Enclosing is 0～2 π, is spaced apart π/L.
4. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 3, its feature exists
In the color-sift feature extracting all training image marginal points includes：The notable picture in edge according to selected training image
Vegetarian refreshments extracts color-sift feature in redness-R, green-G and three passages of blueness-B respectively to former training coloured image, obtains
To redness-R, the green-G of the notable pixel in each edge of this image and the color-sift feature of three passages of blueness-B
decr(e)、decg(e) and decbE (), e represents e-th marginal point of this image, e=1, and 2 ..., E, E are all sides of this image
The sum of the notable pixel of edge.
5. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 4, its feature exists
In the described color-sift feature extracting all training image marginal points includes：To width instruction each in training image database
Practice the color-sift feature extraction of the notable pixel in all edges of image, travel through all images in training image database,
Its feature is followed successively by dec in three Color Channelsr,m(em)、decg,m(em) and decb,m(em), m=1,2 ..., M, M are instruction
Practice image data base size, emRepresent that m opens the e of training imagemIndividual edge pixel point, em=1,2 ..., Em, EmOpen for m
The sum of the notable pixel in all edges of training image.
6. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 5, its feature exists
In the redness-R passage green-G that construction feature dictionary step includes the notable pixel in all edges to whole training images leads to
Color-sift feature dec of road and blueness-channel Br,m(em), decg,m(em) and decb,m(em), meter is clustered by K-means
Calculate, take K cluster centre to obtain the w row of redness-R passage, green-G passage and blueness-channel B, the two dimensional character dictionary of K row
codr、codgAnd codb, wherein K is the number of K-means cluster centre, i.e. the size of dictionary.
7. a kind of image search method based on weights color-sift characteristics dictionary as claimed in claim 1, its feature exists
In：Described weights histogram feature X contains redness-R, green-G and blueness-B three-channel weights histogram feature, inspection
Rope image and image to be retrieved carry out the similitude coupling based on weights histogram feature, calculate 2 norm similarity distances and obtain
Di(X,Xi').
8. a kind of image search method based on weights color-sift characteristics dictionary according to claim 1, its feature
It is：The described weights histogram based on color-sift characteristics dictionary that retrieval Image Coding is calculated with retrieval image
Feature, wherein step 2) described in cluster to k-th cluster centre qkIndividual edge significant point calculates this significant point for this
Centrifugation weights li (k) of cluster centre, calculates the maximum centrifugation weights of all significant points of this cluster centre and obtains this cluster centre
Weight vector α (k), using equation below calculate：
α (k)=max (li (u, k)),
Wherein, u represents to fall in u-th edge significant point of k-th cluster centre, and value is 1,2 ..., qk, k represents k-th cluster
Center, value is 1,2 ..., K, and K represents cluster centre number, i.e. dictionary size, qkRepresent to fall at the edge of k-th cluster centre
The sum of significant point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310294385.2A CN103336835B (en) | 2013-07-12 | 2013-07-12 | Image retrieval method based on weight color-sift characteristic dictionary |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310294385.2A CN103336835B (en) | 2013-07-12 | 2013-07-12 | Image retrieval method based on weight color-sift characteristic dictionary |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103336835A CN103336835A (en) | 2013-10-02 |
| CN103336835B true CN103336835B (en) | 2017-02-08 |
Family
ID=49245000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310294385.2A Expired - Fee Related CN103336835B (en) | 2013-07-12 | 2013-07-12 | Image retrieval method based on weight color-sift characteristic dictionary |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103336835B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103699436B (en) * | 2013-12-30 | 2017-02-22 | 西北工业大学 | Image coding method based on local linear constraint and global structural information |
| CN105138672B (en) * | 2015-09-07 | 2018-08-21 | 北京工业大学 | A kind of image search method of multiple features fusion |
| CN106503143B (en) * | 2016-10-21 | 2020-02-07 | 广东工业大学 | Image retrieval method and device |
| CN106570183B (en) * | 2016-11-14 | 2019-11-15 | 宜宾学院 | A kind of Color Image Retrieval and classification method |
| CN107103325B (en) * | 2017-04-20 | 2021-01-26 | 湘潭大学 | Tissue pathology image classification method |
| CN109711441B (en) * | 2018-12-13 | 2021-02-12 | 泰康保险集团股份有限公司 | Image classification method and device, storage medium and electronic equipment |
| CN111104936A (en) * | 2019-11-19 | 2020-05-05 | 泰康保险集团股份有限公司 | Text image recognition method, device, equipment and storage medium |
| CN111428122B (en) * | 2020-03-20 | 2023-09-01 | 南京中孚信息技术有限公司 | Picture retrieval method and device and electronic equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101546307A (en) * | 2008-09-17 | 2009-09-30 | 厦门市美亚柏科资讯科技有限公司 | Image retrieval system and image retrieval method based on edge gravity center characteristics |
| CN102236717A (en) * | 2011-07-13 | 2011-11-09 | 清华大学 | Image retrieval method based on sketch feature extraction |
| CN102629328A (en) * | 2012-03-12 | 2012-08-08 | 北京工业大学 | Probabilistic latent semantic model object image recognition method with fusion of significant characteristic of color |
| CN102693421A (en) * | 2012-05-31 | 2012-09-26 | 东南大学 | Bull eye iris image identifying method based on SIFT feature packs |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011257979A (en) * | 2010-06-09 | 2011-12-22 | Olympus Imaging Corp | Image retrieval device, image retrieval method, and camera |
-
2013
- 2013-07-12 CN CN201310294385.2A patent/CN103336835B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101546307A (en) * | 2008-09-17 | 2009-09-30 | 厦门市美亚柏科资讯科技有限公司 | Image retrieval system and image retrieval method based on edge gravity center characteristics |
| CN102236717A (en) * | 2011-07-13 | 2011-11-09 | 清华大学 | Image retrieval method based on sketch feature extraction |
| CN102629328A (en) * | 2012-03-12 | 2012-08-08 | 北京工业大学 | Probabilistic latent semantic model object image recognition method with fusion of significant characteristic of color |
| CN102693421A (en) * | 2012-05-31 | 2012-09-26 | 东南大学 | Bull eye iris image identifying method based on SIFT feature packs |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103336835A (en) | 2013-10-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103336835B (en) | Image retrieval method based on weight color-sift characteristic dictionary | |
| Feng et al. | Attention-driven salient edge (s) and region (s) extraction with application to CBIR | |
| Wang et al. | Joint learning of visual attributes, object classes and visual saliency | |
| Wang et al. | Salient object detection for searched web images via global saliency | |
| Li | Analysis of object detection performance based on Faster R-CNN | |
| Kim et al. | Accurate localization of 3D objects from RGB-D data using segmentation hypotheses | |
| CN104199931B (en) | A kind of consistent semantic extracting method of trademark image and trade-mark searching method | |
| CN106227851A (en) | Based on the image search method searched for by depth of seam division that degree of depth convolutional neural networks is end-to-end | |
| CN106021603A (en) | Garment image retrieval method based on segmentation and feature matching | |
| CN105844669A (en) | Video target real-time tracking method based on partial Hash features | |
| CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
| CN101706780A (en) | Image semantic retrieving method based on visual attention model | |
| CN104268514A (en) | Gesture detection method based on multi-feature fusion | |
| CN107292259A (en) | The integrated approach of depth characteristic and traditional characteristic based on AdaRank | |
| CN110413816A (en) | Colored sketches picture search | |
| CN104008375A (en) | Integrated human face recognition mehtod based on feature fusion | |
| CN109934852B (en) | Video description method based on object attribute relation graph | |
| Guo et al. | Image esthetic assessment using both hand-crafting and semantic features | |
| Wang et al. | S 3 D: Scalable pedestrian detection via score scale surface discrimination | |
| CN103413154A (en) | Human motion identification method based on normalized class Google measurement matrix | |
| CN107357834A (en) | Image retrieval method based on visual saliency fusion | |
| Jiang et al. | A unified tree-based framework for joint action localization, recognition and segmentation | |
| CN110287369B (en) | Semantic-based video retrieval method and system | |
| CN103399863A (en) | Image retrieval method based on edge direction difference characteristic bag | |
| CN105404682B (en) | A kind of book retrieval method based on digital image content |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170208 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |