CN102368237A - Image retrieval method, device and system - Google Patents

Abstract

The invention discloses an image retrieval method, an image retrieval device and an image retrieval system, wherein the image retrieval method comprises the following steps: extracting the local features of a query image, and quantizing the local features into visual words; querying a preset visual-word inverted list in an image database by using the visual words so as to obtain matched local-feature pairs and matched images; respectively carrying out space encoding on relative space positions between matched local features in the query image and the matched images so as to obtain a space code picture of the query image and space code pictures of the matched images; executing a space consistency check on the space code picture of the query image and the space code pictures of the matched images so as to obtain the number of the matched local-feature pair in conformity with the space consistency; and according to the numbers of the matched local-feature pairs (in conformity with the space consistency) of different matched images, returning to the matched images according to the similarity of the matched images. By using the method provided by the invention, the image retrieval accuracy and the retrieval efficiency can be improved, and the time consuming for retrieval can be reduced.

Description

Image search method, Apparatus and system

Technical field:

The present invention relates to the data retrieval field, relate in particular to a kind of image search method, Apparatus and system.

Background technology:

In recent years, along with the develop rapidly of technique of internet and popularizing fast of digital equipment, the image on the network has reached hundred billion scales, and is the growth of index scale always.In the face of the view data of magnanimity like this, how effectively to manage, can find interested image easily to make things convenient for the user, promptly image retrieval is that the while that is of practical significance very much also is very challenging work.In the image retrieval, the focus of a research is the retrieval of part duplicating image.The part duplicating image generally is because the user cuts out one and pastes then in the another one image from former figure, or to the modification that former figure adds some literal, perhaps former figure is carried out simple projective transformation.Based on the retrieval of part duplicating image, be with a wide range of applications in the network multimedia field.

Part duplicating image retrieval, promptly from the image data base of a magnanimity, finding with query image has closely similar image-region piece.The best method that addresses this problem at present just is based on the method for sight word code book; The sight word code book generally is a large amount of local feature that obtains through to sampling in the training image; Like SIFT (Scale Invariant Feature Transform; Yardstick invariant features converting characteristic), SURF (Speeded Up Robust Feature, fast robust property characteristic), carry out cluster and obtain.Training image can be an image subset in the image data base, also can be one group of incoherent image.Obtain after the sight word code book; Can extract some local features to a new image in the image data base; And local feature is quantified as sight word; Thereby with the graphical representation sight word vector that is a higher-dimension, then, the matching ratio between query image and the matching image more just be converted between higher-dimension sight word vector matching ratio.If two images respectively have a local feature to be quantified as same sight word, then these two local features constitute a pair of coupling local feature.If two images have no a pair of local feature to constitute coupling, then these two images are considered to incoherent fully.

In the prior art; A kind of basic thought of the images match retrieval based on complete geometric checking is following: quantize through local feature; Obtain the preliminary matching result between query image and the matching image; If two images are relevant really, must there be some total local features to constitute corresponding correct match in the image block that their part is duplicated so.Must satisfy certain affined transformation between these correct part couplings, the coupling of the mistake that those are remaining does not then satisfy this affined transformation.Based on this prerequisite hypothesis, geometric checking adopts the random sampling coherence method fully, and stochastic sampling is some to the coupling local feature, is used to estimate affined transformation; Check other the coupling local feature and the matching degree of said affined transformation then, and record meets the number of the coupling of this affined transformation.When number of samples is abundant immediately, that affined transformation that matching number is maximum then maybe be corresponding correct affined transformation.If a unique point f is arranged in the query image ₁On the coupling unique point f in the matching image ₂, f ₁And f ₂Coordinate in image be respectively (u, v) with (x, y).Then the affined transformation between them can be represented suc as formula (1):

$\left(\begin{array}{c}x\\ y\end{array}\right)=\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)\·\left(\begin{array}{c}u\\ v\end{array}\right)+\left(\begin{array}{c}{t}_1\\ t_2\end{array}\right)---\left(1\right)$

Because this affined transformation has six parameters, so need three pairs of coupling local feature points just possibly estimate to find the solution at least.If query image and all coupling local feature centerings of matching image, the shared ratio of correct coupling is p, three pairs of couplings of so each sampling all are that correct probability is p ³If in the correct coupling, can both correctly estimate affine transformation parameter for any three pairs, if sampling number is N, then having the local characteristic matching of three couples in the once sampling at least all is that correct probability is Np ³Obtain correct affine transformation parameter through such scheme,, obtain result for retrieval in order to judge the matching degree of query image and matching image.

Through the research to technique scheme, the inventor finds: when containing a large amount of erroneous matching in the Where topical characteristic matching, obtain wrong affined transformation possibly, the accuracy of influence retrieval.This scheme needs the sampling of more number of times simultaneously, just possibly on probability, realize sampling the coupling local feature point of three correct, thereby estimate correct affined transformation, and therefore above-mentioned image retrieval scheme based on complete geometric checking is very consuming time.

Summary of the invention

For solving the problems of the technologies described above, the object of the present invention is to provide a kind of image search method, device and system, to solve lower, the consuming time long and not high problem of recall precision of retrieval accuracy that the conventional images retrieval scheme exists.

Image search method provided by the invention comprises:

Extract the local feature of query image, and said local feature is quantified as sight word;

Use in the said sight word query image database preset sight word inverted list, obtain mating local feature to and matching image;

Relative tertiary location between the coupling local feature in the query image is carried out space encoding, obtain query image space code figure, the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure.In order to apply stronger geometrical constraint, can also be reference origin with each local feature respectively, just each quadrant of the plane of delineation evenly is divided into the plurality of sector zone, carries out space encoding then, obtains space code figure.Space code figure has described the relative position relation each other of the local feature in the image.

The Space Consistency check is carried out in the right volume coordinate position of coupling local feature among said query image space code figure and the matching image space code figure, obtain the right number of coupling local feature that meets Space Consistency;

The number of the coupling local feature that meets Space Consistency that comprises according to different matching images calculates the similarity between this matching image and query image, and returns matching image according to said similarity.

Corresponding to above-mentioned image search method, the present invention also provides a kind of image retrieving apparatus, comprising:

First characteristic extracting module is used to extract the local feature of image to be checked;

The first characteristic quantification module is used for said local feature is quantified as sight word;

Enquiry module is used for using the preset sight word inverted list of said sight word query image database, obtain mating local feature to and matching image;

The space encoding module; Be used for the relative tertiary location between the query image coupling local feature is carried out space encoding; Obtain query image space code figure, and the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure;

The Space Consistency inspection module is used for said query image space code figure and the right volume coordinate position of matching image space code figure coupling local feature are carried out the Space Consistency check, to obtain the right number of coupling local feature that meets Space Consistency;

Result for retrieval returns module, is used for the right number of coupling local feature that meets Space Consistency according to different matching images, calculates the similarity between this matching image and query image, returns matching image according to said similarity.

In addition, the present invention also provides a kind of image retrieving apparatus, comprising:

Above-mentioned image retrieving apparatus;

Image data base is used to store the image that supplies match retrieval.

Use the technical scheme that the embodiment of the invention provided; In the image search method that is provided, device and the system; Respectively the coding of the spatial relation between the local feature that is complementary in matching image and the retrieving images is formed space encoding figure; And whether consistent through the spatial relation of checking the local feature that matees, obtain meeting the correct coupling local feature of Space Consistency, thereby effectively got rid of a large amount of wrong local feature coupling that possibly exist; Make the similarity definition between the matching image in query image and the image data base more accurate, can improve the accuracy of retrieval.Simultaneously, the computation complexity of the consistent check algorithm in space is low, compares with the scheme of prior art, and what can reduce to retrieve be consuming time, improves recall precision.

Description of drawings

In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; To do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.

The image search method schematic flow sheet of Fig. 1 for providing in the embodiment of the invention one;

Fig. 2 is the spatial relation synoptic diagram of coupling local feature in the embodiment of the invention two;

Fig. 3 is a schematic flow sheet of confirming the locus of coupling local feature in the embodiment of the invention two;

Fig. 4 is the another kind of spatial relation synoptic diagram of coupling local feature in the embodiment of the invention two;

The image retrieving apparatus structural representation of Fig. 5 for providing in the embodiment of the invention four;

Fig. 6 is for setting up the modular structure synoptic diagram of inverted index table in the embodiment of the invention four;

Fig. 7 is a kind of structural representation of the Space Consistency inspection module that provides in the embodiment of the invention four;

Fig. 8 is a kind of working method synoptic diagram of the image indexing system that provides in the embodiment of the invention five.

Embodiment

Lower, the consuming time long and not high problem of recall precision of retrieval accuracy that the specific embodiment of the invention exists in order to solve the conventional images retrieval scheme provides a kind of image search method, device and system.

Said image search method comprises: extract the local feature of query image, and said local feature is quantified as sight word; Use in the said sight word query image database preset sight word inverted list, obtain mating local feature to and matching image; Relative tertiary location to coupling local feature in the query image carries out space encoding, obtains query image space code figure, and coupling local feature relative tertiary location in the matching image is carried out space encoding, obtains matching image space code figure; Said query image space code figure and matching image space code figure are carried out the Space Consistency check, obtain the right number of coupling local feature that meets Space Consistency; The right number of coupling local feature that meets Space Consistency according to different matching images returns matching image according to the similarity of matching image.

It more than is the application's core concept; To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention carried out clear, intactly description, obviously; Described embodiment only is the present invention's part embodiment, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.

Embodiment one:

Referring to shown in Figure 1, a kind of schematic flow sheet of the image search method that provides for present embodiment, this method may further comprise the steps:

Step S101, the local feature of extraction query image;

Step S102 is quantified as sight word with said local feature;

Wherein, said sight word can define in the following manner: from training image, extract local feature, said local feature is carried out cluster, the center of confirming each cluster is a sight word.Can obtain a plurality of sight word through a plurality of cluster centres, generate the sight word code book that comprises a plurality of local features, sight word and corresponding relation thereof.Because sight word does not have clear and definite semantic information, the size of sight word code book, the number of the sight word that promptly wherein comprises generally can be confirmed through the experience of test data of experiment.In addition, can also directly use the sight word code book that has defined in the identical or close technical field in the present embodiment.

After extracting the local feature of image to be checked; To each local feature wherein; Visit said sight word code book,, can this local feature be quantified as that minimum sight word of distance according to wherein local feature and the distance between each sight word in the vision code book.

Step S103 uses in the said sight word query image database preset sight word inverted list, obtain mating local feature to and matching image;

For fear of image to be checked unnecessary with image data base in the comparison of complete uncorrelated image; Used for reference the thought in the text retrieval in the present embodiment based on textual words inverted list index structure; Its basic thought comprises: to each sight word; Generate a tabulation, each in the tabulation is associated with the image that comprises this sight word, further can also comprise the frequency that comprises this sight word in the image.This tabulation is called the sight word inverted list; Through said inverted list; Can learn that each sight word is comprised by which image, further, a given image to be checked; Only need search the corresponding tabulation inverted list of sight word that said image to be checked comprises, just can know with said image to be checked to have the right matching image of coupling local feature.Wherein, there are one or more matching images in an image to be checked, exists one or more coupling local features right between an image to be checked and a matching image.The local feature of coupling described in the present embodiment the local feature of identical sight word to being meant the correspondence that exists in said image to be checked and the matching image.

Wherein, can adopt following mode to set up the sight word inverted list of image data base:

Extract the local feature of image in the image data base, said local feature is quantified as sight word;

Set up the related of image in said sight word and the said image data base, obtain the sight word inverted list.

Wherein, can extract the local feature of each image in the image data base, and all local features are quantified as sight word.

Image in the said image data base can be set up related with one or more vision dictionaries; Inverted list comprises one or more data table items in the said vision dictionary; Can related one or more images, further can also comprise the frequency that comprises this sight word in the image.

Step S104; Relative tertiary location between the coupling local feature in the query image is carried out space encoding; Obtain query image space code figure, the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure;

Wherein, Space encoding is in the query image; Describe by the relation of the relative tertiary location between the said coupling local feature that obtains among the step S103, and the relative tertiary location relation that obtains among the step S103 mating in the matching image between local feature is described.

Step S105 carries out the Space Consistency check to the space code figure that matees local feature in said query image and the matching image, obtains the right number of coupling local feature that meets Space Consistency;

If two images are that part is duplicated; Be that it owns certain image block areas together; Then in the piece zone that this duplicates; The local feature of these two images necessarily meets the consistance of relative space position, those incongruent this conforming couplings corresponding probably the coupling local feature of a mistake right, should be deleted.This Space Consistency detection can be carried out xor operation through the volume coordinate position that comprises in the space encoding of mating local feature and carry out.

Step S106, the right number of coupling local feature that meets Space Consistency according to different matching images calculates the similarity between this matching image and query image, returns matching image according to said similarity.

Suppose query image q and matching image p through behind the matching inquiry, a is arranged to the coupling local feature, wherein b has passed through the Space Consistency check to the coupling local feature, so, can define the similarity of query image q and matching image p, suc as formula (2):

$S(p,q)=b-\frac{a-b+1}{a}\·\frac{N\left(p\right)}{N_{\max }}---\left(2\right)$

Wherein N (p) representes local feature total number among the matching image p, N _MaxIn the presentation video database image maximum local feature numbers that possibly contain.

In this step; Can sort according to the similarity size of matching image in the database and query image; And return to the user, a threshold values that meets the coupling local feature number of Space Consistency can also be set, when the coupling local feature number that meets Space Consistency of a matching image reaches said threshold values; Confirm that this matching image is relevant with query image, returns to inquiring user with it; Perhaps can put the threshold values of similarity, when the similarity of a matching image reaches said threshold values, confirm that this matching image is relevant with query image, returns to inquiring user with it.

In the image search method that present embodiment provides; Respectively the coding of the spatial relation between the local feature that is complementary in matching image and the retrieving images is formed space encoding figure; And it is whether consistent through the spatial relation of checking the local feature that matees; Obtain meeting the correct coupling local feature of Space Consistency; Thereby effectively got rid of a large amount of wrong local feature coupling that possibly exist, made the similarity definition between the matching image in query image and the image data base more accurate, can improve the accuracy of retrieval.Simultaneously, the computation complexity of the consistent check algorithm in space is low, compares with the scheme of prior art, and what can reduce to retrieve be consuming time, improves recall precision.

Embodiment two:

Present embodiment provides a kind of relative tertiary location to the local feature in the image to concern the implementation of carrying out space encoding, and is specific as follows said:

Describe the relative position relation of the horizontal direction of two coupling local features, obtain horizontal direction space code figure;

Describe the relative position relation of the vertical direction of two coupling local features, obtain vertical direction space code figure.

For a query image or matching image, generate two space code figure, be designated as X-map and Y-map respectively.Wherein, X-map is in order to the relative tertiary location relation of the along continuous straight runs between the local feature of describing two couplings (x), and Y-map is in order to the relative tertiary location relation of the vertical direction between the local feature of describing two couplings (Y-axis).For example, there be K coupling local feature { v between the given query image I, itself and matching image _i, (i=1,2 ..., K) right, then the X-map of I and Y-map all can be the binary matrix of a K * K, define suc as formula (3) and formula (4):

$\mathrm{Xmap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& x_i<x_j\\ 1& \mathrm{if}& x_i\&GreaterEqual;x_j\end{array}\right.---\left(3\right)$

$\mathrm{Ymap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& y_i<y_j\\ 1& \mathrm{if}& y_i\&GreaterEqual;y_j\end{array}\right.---\left(4\right)$

Shown in Fig. 2 left side; The expression query image comprises 4 coupling local features; Fig. 2 has represented on the right side coupling local feature 1,3 and the 4 relative tertiary location relations with respect to coupling local feature 2, and wherein, coupling local feature 1 is on the upper left side of coupling local feature 2; Coupling local feature 3 is in the lower right of coupling local feature 2, and coupling local feature 4 is in the lower left of coupling local feature 2.By above-mentioned relative tertiary location relation, convolution (3) and formula (4) can obtain two space code figure of this query image, respectively suc as formula (5) and formula (6):

$\mathrm{Xmap}=\left(\begin{array}{cccc}1& 0& 0& 1\\ 1& 1& 0& 1\\ 1& 1& 1& 1\\ 0& 0& 0& 1\end{array}\right)---\left(5\right)$

$\mathrm{Ymap}=\left(\begin{array}{cccc}1& 1& 1& 1\\ 0& 1& 1& 1\\ 0& 0& 1& 1\\ 0& 0& 0& 1\end{array}\right)---\left(6\right)$

In addition; For this spatial division and coding are carried out the vague generalization popularization; Further the relative position relation between the refinement coupling local feature point is the center with the fixed reference feature point promptly, the plane of delineation is divided into four quadrants after; Again each quadrant on average is divided into a plurality of sector regions, which sector region definite then other each local feature points fall in.During concrete the realization, can adopt following method.In the image search method that schematic flow sheet as shown in Figure 3, present embodiment provide, before carrying out the space encoding step, can also may further comprise the steps:

Step S301 is a reference origin with each picture position, coupling local feature place respectively, and the image division that the coupling local feature is belonged to is four quadrants;

Step S302 evenly is divided into a plurality of sector regions with each quadrant, and confirms that other local feature point is with respect to quadrant and sector region that reference origin was positioned at;

Step S303, the volume coordinate position that each coupling local feature is belonged to is rotated about reference origin, obtains the new volume coordinate position of this coupling local feature;

Step S304 according to coupling local feature new volume coordinate position, obtains the relative tertiary location of itself and said reference origin.

After in the present embodiment said query image or matching image plane being divided into four quadrants, to can also each quadrant further evenly being divided into a plurality of sector regions.When implementation algorithm, this dividing mode can be decomposed into the r sub-divided.Each son can use formula (3) and formula (4) to carry out space encoding after dividing and being rotated counterclockwise the θ angle, and is final, can obtain the space code figure GX and the GY of two general three-dimensionals.

Concrete, can be to coupling local feature v in query image or the matching image _iVolume coordinate position (x _i, y _i) be rotated counterclockwise about image origin

(k=0,1 ..., r-1) degree obtains the new volume coordinate position (x of this coupling local feature _i ^k, y _i ^k) suc as formula shown in (7):

$\left(\begin{array}{c}{x_i}^k\\ {y_i}^k\end{array}\right)=\left(\begin{array}{cc}\cos \left(\mathrm{\&theta;}\right)& \sin \left(\mathrm{\&theta;}\right)\\ -\sin \left(\mathrm{\&theta;}\right)& \cos \left(\mathrm{\&theta;}\right)\end{array}\right)\&CenterDot;\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right)---\left(7\right)$

Then, according to the new volume coordinate position (x that obtains _i ^k, y _i ^k), can obtain the space code figure GX and the GY of two general three-dimensionals, shown in (8) and formula (9).

$\mathrm{GX}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {x_i}^k<{x_j}^k\\ 1& \mathrm{if}& {x_i}^k\&GreaterEqual;{x_j}^k\end{array}\right.---\left(8\right)$

$\mathrm{GY}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {y_i}^k<{y_j}^k\\ 1& \mathrm{if}& {y_i}^k\&GreaterEqual;{y_j}^k\end{array}\right.---\left(9\right)$

Wherein, x _i ^kAnd x _j ^kBe respectively k postrotational horizontal ordinate of i local feature point and j coupling local feature, y _i ^kAnd y _j ^kBe respectively k postrotational ordinate of i coupling local feature and j coupling local feature, shown in (7) formula.In fact; Three-dimensional space code figure GX and GY have carried out more careful description to the relative position relation between each coupling local feature; Pointed out that not only certain characteristic is positioned at which quadrant of the plane of delineation of dividing based on fixed reference feature point, points out further also it is positioned at which sheet sector region of this quadrant.

As shown in Figure 4, in order to show the division of sector region more intuitively, earlier image is divided into four quadrants, then each quadrant is divided into a plurality of uniform fan zone.The expression of Fig. 4 left side is divided into two sector regions uniformly with each quadrant of this image, thereby it is fan-shaped that entire image is divided into eight equal portions, and the division of this image can be resolved into two sub-divided; Respectively like the first half in the middle of Fig. 4 with shown in the middle the latter half of Fig. 4; Wherein, 4 middle the first half can adopt formula (3) and formula (4) to encode, and the latter half is after being rotated counterclockwise 45 in the middle of Fig. 4; Shown in Fig. 4 right side, can adopt formula (8) and (9) to encode.

The technical scheme that present embodiment provides is a kind of scheme that realizes the relative tertiary location relation of the local feature in the image is carried out space encoding.In practical application, those skilled in the art can select alternate manner to realize the relative tertiary location relation of different local features is carried out space encoding, and the technical scheme that provides among present embodiment and the embodiment one can cross-references, repeats no more at this.

Embodiment three:

A kind of implementation to the right volume coordinate position execution Space Consistency check of coupling local feature among said query image space code figure and the matching image space code figure that present embodiment provides specifically can be described below:

Quantize according to the coupling local feature, if query image I _qWith matching image I _mHave the local feature of N, then in the step S104 described in the embodiment one, respectively these are mated local features at I coupling _qAnd I _mIn relative tertiary location carry out space encoding, obtain query image space code figure (GX _q, GY _q) and matching image space code figure (GX _m, GY _m).For the consistance of the relative tertiary location between the coupling local feature between comparison query image and the matching image, can be respectively to GX _qAnd GX _m, GY _qAnd GY _mCarry out xor operation, shown in (10) and formula (11):

$V_x(i,j,k)={\mathrm{GX}}_q(i,j,k)\&CirclePlus;{\mathrm{GX}}_m(i,j,k)---\left(10\right)$

$V_y(i,j,k)={\mathrm{GY}}_q(i,j,k)\&CirclePlus;{\mathrm{GY}}_m(i,j,k)---\left(11\right)$

In addition, present embodiment also provides a kind of implementation of obtaining the right number of the coupling local feature that meets Space Consistency, specifically can be described below:

Under the ideal situation, if all N are correct to the coupling local feature, V so _xAnd V _yIn all elements all will be zero.But, if there are some wrong coupling local features, the coupling local feature of mistake corresponding at GX _qAnd GX _mIntermediate value will be inconsistent; In like manner, the coupling local feature of mistake corresponding at GY _qAnd GY _mIntermediate value also will be inconsistent.This value inconsistent will cause the V as a result of their XORs _xAnd V _yCorresponding value is 1.

Order $S_x\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_x(i,j,k),$ $S_y\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_y(i,j,k),$

If exist i to make S _x(i)＞0, we define so

I then ^*Individual local matching characteristic should be deleted it corresponding the coupling of a mistake probably.To S _yAlso do same operation.Iteration carry out above-mentioned deletion action, the matching characteristic that to the last remains is to pairing S _xAnd S _yIn value be complete zero.

Confirm that remaining coupling local feature is the right number of coupling local feature that meets Space Consistency in query image and the corresponding matching image to number.

The technical scheme that present embodiment provides; It is a kind of scheme of the right volume coordinate position of coupling local feature among said query image space code figure and the matching image space code figure being carried out the Space Consistency check; In practical application; Those skilled in the art can select alternate manner to realize the Space Consistency check that the coupling local feature is right, and the technical scheme that provides among present embodiment and the embodiment one can cross-references, repeats no more at this.

Embodiment four:

Correspond to above-mentioned image processing method, present embodiment also provides a kind of image retrieving apparatus, and is as shown in Figure 5, comprising:

First characteristic extracting module 501 is used to extract the local feature of image to be checked;

The first characteristic quantification module 502 is used for said local feature is quantified as sight word;

Enquiry module 503 is used for using the preset sight word inverted list of said sight word query image database, obtain mating local feature to and matching image;

Space encoding module 504; Be used for the relative tertiary location between the query image coupling local feature is carried out space encoding; Obtain query image space code figure, and the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure;

Space Consistency inspection module 505 is used for said query image space code figure and matching image space code figure are carried out the Space Consistency check, to obtain the right number of coupling local feature that meets Space Consistency;

Result for retrieval returns module 506, is used for the right number of coupling local feature that meets Space Consistency according to different matching images, calculates the similarity between this matching image and query image, returns matching image according to said similarity size.

In the present embodiment, the inverted index table of said sight word and said image data base can be by generating with lower module, and referring to shown in Figure 6, a kind of structural representation for this module specifically comprises:

Sight word generation module 601 is used for extracting local feature from training image, and said local feature is carried out cluster, and the center of confirming each cluster is a sight word.

Second characteristic extracting module 602 is used for extracting the local feature of image data base image;

The second characteristic quantification module 603 is used for the local feature that the image from image data base extracts is quantified as sight word;

Inverted list is set up module 604, is used for setting up image related of said sight word and said image data base, obtains the sight word inverted list.

Relative position relation between two local features can be decomposed into relative position relation and the relative position relation on the vertical direction on the horizontal direction, and therefore said space encoding module 504 specifically can comprise:

First coding unit is used to describe two relative position relations that mate the horizontal direction of local features, obtains horizontal direction space code figure;

Second coding unit is used to describe two relative position relations that mate the vertical direction of local features, obtains vertical direction space code figure.

Said first coding unit specifically can obtain horizontal direction space code figure through the following formula coding:

$\mathrm{Xmap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& x_i<x_j\\ 1& \mathrm{if}& x_i\&GreaterEqual;x_j\end{array}\right.;$

Said second coding unit specifically can obtain vertical direction space code figure through the following formula coding:

$\mathrm{Ymap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& y_i<y_j\\ 1& \mathrm{if}& y_i\&GreaterEqual;y_j\end{array}\right.;$

Wherein, x _iAnd x _jBe respectively the horizontal ordinate of two coupling local features, y _iAnd y _jBe respectively the ordinate of two coupling local features.

Via the operation of space encoding module, a pair of query image and matching image can obtain: query image horizontal direction space code figure, query image vertical direction space code figure, matching image horizontal direction space code figure and matching image vertical direction space code figure.

In addition, said image processing apparatus can also comprise:

The area dividing module, the coordinate position that is used for each coupling local feature is an initial point, with four quadrants of image division at coupling local feature place, and then evenly is divided into r sector region to each quadrant;

The coordinate determination module is used for the volume coordinate position at each coupling local feature place is rotated about image origin, obtains the new volume coordinate position of this coupling local feature;

The locus determination module is used for according to coupling local feature new volume coordinate position, obtains the relative tertiary location of itself and said reference origin.

Said coordinate determination module, specifically can confirm about the new volume coordinate position of the postrotational coupling local feature of reference origin through following mode:

Volume coordinate position (x to the coupling local feature that comprises in the image _i, y _i) be rotated counterclockwise about reference origin

(k=0,1 ..., r-1) degree obtains the new volume coordinate position (x of this coupling local feature _i ^k, y _i ^k) shown in the following formula:

$\left(\begin{array}{c}{x_i}^k\\ {y_i}^k\end{array}\right)=\left(\begin{array}{cc}\cos \left(\mathrm{\&theta;}\right)& \sin \left(\mathrm{\&theta;}\right)\\ -\sin \left(\mathrm{\&theta;}\right)& \cos \left(\mathrm{\&theta;}\right)\end{array}\right)\&CenterDot;\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right);$

Wherein, r is the sector region number that image comprises, i=0, and 1 ..., N-1, wherein N is the right total number of matching characteristic;

Said space encoding module, specifically can carry out space encoding to the relative tertiary location between the coupling local feature in the image through following mode:

By first coding unit, adopt the following formula coding to obtain horizontal direction space code figure:

$\mathrm{GX}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {x_i}^k<{x_j}^k\\ 1& \mathrm{if}& {x_i}^k\&GreaterEqual;{x_j}^k\end{array}\right.;$

By second coding unit, adopt the following formula coding to obtain vertical direction space code figure:

$\mathrm{GY}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {y_i}^k<{y_j}^k\\ 1& \mathrm{if}& {y_i}^k\&GreaterEqual;{y_j}^k\end{array}\right.;$

Wherein, x _i ^kAnd x _j ^kBe respectively k postrotational horizontal ordinate of i coupling local feature and j coupling local feature, y _i ^kAnd y _j ^kBe respectively k postrotational ordinate of i coupling local feature and j coupling local feature.Three-dimensional space code figure GX and GY have carried out more careful description to the relative position relation between each coupling local feature; Pointed out that not only certain characteristic is positioned at which quadrant of the plane of delineation of dividing based on fixed reference feature point, points out further also it is positioned at which sheet sector region of this quadrant.

Said Space Consistency inspection module 505 shown in the structural representation of Fig. 7, specifically can comprise:

XOR unit 505a is used for said query image space code figure and the right volume coordinate position of matching image space code figure coupling local feature are carried out xor operation;

Erroneous matching delete cells 505b; Xor operation result according to the right volume coordinate position of coupling local feature among query image space code figure and the matching image space code figure; The matching characteristic of confirming to meet Space Consistency is right, confirm and deletion not meet the matching characteristic of Space Consistency right; The space code figure xor operation of the volume coordinate position correspondence that the coupling local feature that meets Space Consistency that keeps at last is right sum as a result is 0.

Matching number is confirmed unit 505c, is used for confirming the remaining right number of correct match characteristic that meets Space Consistency.

Said XOR unit, specifically according to shown in the following formula, xor operation is carried out in the volume coordinate position that the coupling local feature is right:

$V_x(i,j,k)={\mathrm{GX}}_q(i,j,k)\&CirclePlus;{\mathrm{GX}}_m(i,j,k);$

$V_y(i,j,k)={\mathrm{GY}}_q(i,j,k)\&CirclePlus;{\mathrm{GY}}_m(i,j,k);$

Wherein, (GX _q, GY _q) be query image space code figure, (GX _m, GY _m) be matching image space code figure;

Said erroneous matching unit, it is right that the coupling local feature that does not meet Space Consistency is obtained and deleted to employing with following formula:

Order $S_x\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_x(i,j,k),$ $S_y\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_y(i,j,k),$

If exist i to make S _x(i)＞0, we define so

I then ^*Individual local matching characteristic is to corresponding the coupling of a mistake probably, and we delete it.To S _yAlso do same operation.Iteration carry out such deletion action, the matching characteristic that to the last remains is to pairing S _xAnd S _yIn value be complete zero.

Said matching number is confirmed the unit, obtain the right number of matching characteristic that is left to meet the demands, as the right number of coupling local feature that meets Space Consistency.

In addition, said result for retrieval returns module 506, specifically can comprise:

Similarity calculated is used for according to the right number of matching characteristic that meets Space Consistency, and the number of the local feature that is comprised separately in matching image and the query image, calculates the similarity between matching image and query image;

Said similarity calculated, specifically can calculate the similarity between matching image and query image in the following way:

If query image q and matching image p through inquiry after, a is arranged to the coupling local feature, wherein b has passed through the Space Consistency check to mating local feature, the similarity of query image q and matching image p then, S (p q) is shown below:

$S(p,q)=b-\frac{a-b+1}{a}\&CenterDot;\frac{N\left(p\right)}{N_{\max }};$

Wherein N (p) representes local feature total number among the matching image p, N _MaxIn the presentation video database image maximum local feature numbers that possibly contain.

Matching image returns the unit, is used for according to the said matching image of said similarity size ordering, and returns said matching image according to said ordering.

Present embodiment is the corresponding device embodiment of aforesaid way embodiment, and embodiment can repeat no more at this referring to the description of method embodiment.

In the image retrieving apparatus that present embodiment provides; Respectively the coding of the spatial relation between the local feature that is complementary in matching image and the retrieving images is formed space encoding figure, and pass through to check the spatial relation of the local feature that matees whether consistent, obtain meeting the correct coupling local feature of Space Consistency; Thereby effectively got rid of a large amount of wrong local feature coupling that possibly exist; Be that similarity definition between the matching image in query image and the image data base is more accurate, can improve the accuracy of retrieval, simultaneously; The computation complexity of the consistent check algorithm in space is low; Compare with the scheme of prior art, what can reduce to retrieve be consuming time, improves recall precision.

Embodiment five:

Corresponding to above-mentioned image processing method and image processing apparatus, present embodiment also provides a kind of image processing system, specifically comprises:

The image retrieving apparatus that is provided among the embodiment five;

And image data base, be used to store the image that supplies match retrieval.

Further, said image processing system can also comprise:

Image collection module is used for obtaining new image from network, and stores in the said image data base; Said image collection module can constantly expand the raw image data storehouse through web crawlers download pictures or the corresponding URL (Uniform/Universal Resource Locator, URL) of picture.

The inverted list update module is used for obtaining the corresponding inverted list of vision dictionary that said new image comprises, and upgrades image related in the said inverted list.

As shown in Figure 8, be a kind of working method synoptic diagram of this system.

Present embodiment is aforesaid way and the corresponding system embodiment of device embodiment, and embodiment can repeat no more at this referring to the description of method and apparatus embodiment.

In the image search method that the embodiment of the invention provides, device and the system; Respectively the coding of the spatial relation between the local feature that is complementary in matching image and the retrieving images is formed space encoding figure, and pass through to check the spatial relation of the local feature that matees whether consistent, obtain meeting the correct coupling local feature of Space Consistency; Thereby effectively got rid of a large amount of wrong local feature coupling that possibly exist; Be that similarity definition between the matching image in query image and the image data base is more accurate, can improve the accuracy of retrieval, simultaneously; The computation complexity of the consistent check algorithm in space is low; Compare with the scheme of prior art, what can reduce to retrieve be consuming time, improves recall precision.

For device of the present invention and system embodiment, because it is basically corresponding to method embodiment, so relevant part gets final product referring to the part explanation of method embodiment.Device embodiment described above only is schematic; Wherein said unit as the separating component explanation can or can not be physically to separate also; The parts that show as the unit can be or can not be physical locations also; Promptly can be positioned at a place, perhaps also can be distributed on a plurality of equipment.Can select wherein some or all of module to realize the purpose of present embodiment scheme according to the actual needs.Those of ordinary skills promptly can understand and implement under the situation of not paying creative work.

In several embodiment that the application provided, should be understood that the methods, devices and systems that disclosed not surpassing in the application's the spirit and scope, can be realized through other mode.Current embodiment is a kind of exemplary example, should be as restriction, and given particular content should in no way limit the application's purpose.For example, the division of said unit or subelement only is that a kind of logic function is divided, and during actual the realization other dividing mode can be arranged, and for example a plurality of unit or a plurality of subelement combine.In addition, a plurality of unit can or assembly can combine or can be integrated into another system, or some characteristics can ignore, or do not carry out.

In addition, the synoptic diagram of institute's describing method, device and system and different embodiment, in the scope that does not exceed the application, can with other system, module, technology or method combine or are integrated.Another point, the coupling each other that shows or discuss or directly coupling or communication to connect can be through some interfaces, the indirect coupling of device or unit or communication connect, and can be electrically, machinery or other form.

Each embodiment adopts the mode of going forward one by one to describe in this instructions, and what each embodiment stressed all is and the difference of other embodiment that identical similar part is mutually referring to getting final product between each embodiment.To the above-mentioned explanation of the disclosed embodiments, make this area professional and technical personnel can realize or use the present invention.Multiple modification to these embodiment will be conspicuous concerning those skilled in the art, and defined General Principle can realize under the situation that does not break away from the spirit or scope of the present invention in other embodiments among this paper.Therefore, the present invention will can not be restricted to these embodiment shown in this paper, but will meet and principle disclosed herein and features of novelty the wideest corresponding to scope.

Claims (24)

1. an image search method is characterized in that, comprising:

Extract the local feature of query image, and said local feature is quantified as sight word;

Use in the said sight word query image database preset sight word inverted list, obtain mating local feature to and matching image;

Relative tertiary location between the coupling local feature in the query image is carried out space encoding, obtain query image space code figure, the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure;

The Space Consistency check is carried out in the right volume coordinate position of coupling local feature among said query image space code figure and the matching image space code figure, obtain the right number of coupling local feature that meets Space Consistency;

The number of the coupling local feature that meets Space Consistency that comprises according to different matching images calculates the similarity between this matching image and query image, and returns matching image according to said similarity size.

2. method according to claim 1 is characterized in that, also comprises:

From training image, extract local feature, said local feature is carried out cluster, the center of confirming each cluster is a sight word.

3. method according to claim 1 is characterized in that, also comprises:

Extract the local feature of image in the image data base, said local feature is quantified as sight word;

Set up the related of image in said sight word and the said image data base, obtain the sight word inverted list.

4. method according to claim 1 is characterized in that, the sight word inverted list of presetting in the said sight word query image of the said use database comprises:

Use sight word to search sight word inverted list preset in the image data base, all associated pictures of confirming to comprise in the image data base of this sight word are the matching image of said query image;

Set up the matching relationship between the local feature between query image and matching image.

5. method according to claim 1 is characterized in that, the relative tertiary location between the coupling local feature is carried out space encoding, specifically comprises:

Describe the relative position relation of the horizontal direction of two coupling local features, obtain horizontal direction space code figure;

Describe the relative position relation of the vertical direction of two coupling local features, obtain vertical direction space code figure.

6. method according to claim 5 is characterized in that:

With image level director space sign indicating number seal is X-map, then:

$\mathrm{Xmap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& x_i<x_j\\ 1& \mathrm{if}& x_i\&GreaterEqual;x_j\end{array}\right.;$

With image vertical direction space code seal is Y-map, then:

$\mathrm{Ymap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& y_i<y_j\\ 1& \mathrm{if}& y_i\&GreaterEqual;y_j\end{array}\right.;$

Wherein, x _iAnd x _jBe respectively the horizontal ordinate of two coupling local features, y _iAnd y _jBe respectively the ordinate of two coupling local features.

7. method according to claim 1 is characterized in that, the relative tertiary location of coupling local feature is carried out also comprising before the space encoding:

Be reference with each coupling local feature respectively, its position in image is regarded as reference origin, the image division that the coupling local feature is belonged to is four quadrants;

Each quadrant evenly is divided into a plurality of sector regions, and confirms that other coupling local feature is with respect to quadrant and sector region that reference origin was positioned at;

Volume coordinate position to each coupling local feature is rotated about reference origin, obtains the new volume coordinate position of this coupling local feature;

According to coupling local feature new volume coordinate position, obtain the relative tertiary location of itself and said reference origin.

8. method according to claim 7 is characterized in that:

Volume coordinate position to each coupling local feature is rotated about reference origin, obtains the new volume coordinate position of this coupling local feature, specifically comprises:

Volume coordinate position (x to the coupling local feature that comprises in the image _i, y _i) be rotated counterclockwise about reference origin (k=0,1 ..., r-1) degree obtains the new volume coordinate position (x of this coupling local feature _i ^k, y _i ^k) shown in the following formula:

$\left(\begin{array}{c}{x_i}^k\\ {y_i}^k\end{array}\right)=\left(\begin{array}{cc}\cos \left(\mathrm{\&theta;}\right)& \sin \left(\mathrm{\&theta;}\right)\\ -\sin \left(\mathrm{\&theta;}\right)& \cos \left(\mathrm{\&theta;}\right)\end{array}\right)\&CenterDot;\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right);$

Wherein, the sector region number that r comprises for each quadrant, i=0,1 ..., N-1, wherein N is the right total number of matching characteristic;

The space code figure that relative tertiary location between the coupling local feature in the image is carried out obtaining after the space encoding comprises:

Horizontal direction space code figure: $\mathrm{GX}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{If}& {x_i}^k<{x_j}^k\\ 1& \mathrm{If}& {x_i}^k\&GreaterEqual;{x_j}^k\end{array}\right.;$

Vertical direction space code figure: $\mathrm{GY}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{If}& {y_i}^k<{y_j}^k\\ 1& \mathrm{If}& {y_i}^k\&GreaterEqual;{y_j}^k\end{array}\right.;$

Wherein, x _i ^kAnd x _j ^kBe respectively k postrotational horizontal ordinate of i coupling local feature and j coupling local feature, y _i ^kAnd y _j ^kBe respectively k postrotational ordinate of i coupling local feature and j coupling local feature; Confirm each coupling local feature after image is divided with respect to reference origin through horizontal direction space code figure GX and vertical direction space code figure GY, quadrant that is positioned at and sector region.

9. method according to claim 1 is characterized in that:

Coupling local feature among said query image space code figure and the matching image space code figure is carried out the Space Consistency check, specifically comprises:

Xor operation is carried out in the right volume coordinate position of coupling local feature among said query image space code figure and the matching image space code figure;

The said right number of coupling local feature that meets Space Consistency that obtains specifically comprises:

Confirm and to delete all coupling local features that do not meet Space Consistency right;

Confirm that remaining coupling local feature is the right number of coupling local feature that meets Space Consistency to number.

10. method according to claim 9 is characterized in that:

Specifically according to down shown in the mode, to mating the right volume coordinate position execution xor operation of local feature:

$V_x(i,j,k)={\mathrm{GX}}_q(i,j,k)\&CirclePlus;{\mathrm{GX}}_m(i,j,k);$

$V_y(i,j,k)={\mathrm{GY}}_q(i,j,k)\&CirclePlus;{\mathrm{GY}}_m(i,j,k);$

Wherein, (GX _q, GY _q) be query image space code figure, (GX _m, GY _m) be matching image space code figure;

Adopt following mode to confirm and to delete all coupling local features that do not meet Space Consistency right:

Order $S_x\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_x(i,j,k),$ $S_y\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_y(i,j,k),$

If exist i to make S _xOr S (i)＞0 _y(i)＞0, definition

Or

And delete i ^*Right to local matching characteristic;

Iteration carry out above-mentioned deletion action, the matching characteristic that to the last remains is to pairing S _xAnd S _yIn value be complete zero.

11. method according to claim 1 is characterized in that, specifically according to shown in the following formula, calculates the similarity between matching image and query image:

If query image q and matching image p through inquiry after, a is arranged to the coupling local feature, wherein b has passed through the Space Consistency check to mating local feature, the similarity of query image q and matching image p then, S (p q) is shown below:

$S(p,q)=b-\frac{a-b+1}{a}\&CenterDot;\frac{N\left(p\right)}{N_{\max }};$

Wherein N (p) representes local feature total number among the matching image p, N _MaxIn the presentation video database image maximum local feature numbers that possibly contain.

12. an image retrieving apparatus is characterized in that, comprising:

First characteristic extracting module is used to extract the local feature of image to be checked;

The first characteristic quantification module is used for said local feature is quantified as sight word;

Enquiry module is used for using the preset sight word inverted list of said sight word query image database, obtain mating local feature to and matching image;

The space encoding module; Be used for the relative tertiary location between the query image coupling local feature is carried out space encoding; Obtain query image space code figure, and the relative tertiary location between the coupling local feature in the matching image is carried out space encoding, obtain matching image space code figure;

The Space Consistency inspection module is used for said query image space code figure and the right volume coordinate position of matching image space code figure coupling local feature are carried out the Space Consistency check, to obtain the right number of coupling local feature that meets Space Consistency;

Result for retrieval returns module, is used for the right number of coupling local feature that meets Space Consistency according to different matching images, calculates the similarity between this matching image and query image, according to said similarity size, returns matching image.

13. device according to claim 12 is characterized in that, also comprises:

The sight word generation module is used for extracting local feature from training image, and said local feature is carried out cluster, and the center of confirming each cluster is a sight word.

14. device according to claim 12 is characterized in that, also comprises:

Second characteristic extracting module is used for extracting the local feature of image data base image;

The second characteristic quantification module is used for the local feature that the image from image data base extracts is quantified as sight word;

Inverted list is set up module, is used for setting up image related of said sight word and said image data base, obtains the sight word inverted list.

15. device according to claim 12 is characterized in that, said space encoding module specifically comprises:

First coding unit is used to describe two relative position relations that mate the horizontal direction of local features, obtains horizontal direction space code figure;

Second coding unit is used to describe two relative position relations that mate the vertical direction of local features, obtains vertical direction space code figure.

16. device according to claim 15 is characterized in that:

Said first coding unit specifically obtains horizontal direction space code figure through the following formula coding:

$\mathrm{Xmap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& x_i<x_j\\ 1& \mathrm{if}& x_i\&GreaterEqual;x_j\end{array}\right.;$

Said second coding unit specifically obtains vertical direction space code figure through the following formula coding:

$\mathrm{Ymap}(i,j)=\left\{\begin{array}{ccc}0& \mathrm{if}& y_i<y_j\\ 1& \mathrm{if}& y_i\&GreaterEqual;y_j\end{array}\right.;$

Wherein, x _iAnd x _jBe respectively the horizontal ordinate of two coupling local features, y _iAnd y _jBe respectively the ordinate of two coupling local features.

17. device according to claim 12 is characterized in that, also comprises:

The area dividing module, the image that is used for coupling local feature place is that reference origin is divided into four quadrants with certain matching characteristic position, and each quadrant is divided into a plurality of sector regions;

The coordinate determination module is used for the volume coordinate position at each coupling local feature place is rotated about reference origin, obtains the new volume coordinate position of this coupling local feature;

The locus determination module is used for according to coupling local feature new volume coordinate position, obtains the relative tertiary location of itself and said reference origin.

18. device according to claim 17 is characterized in that:

Said coordinate determination module; Be used for confirming each coupling local feature at the quadrant and the sector region that are positioned at based on the reference origin divided image, specifically definite about the new volume coordinate position of the postrotational coupling local feature of reference origin through following mode:

Volume coordinate position (x to the coupling local feature that comprises in the image _i, y _i) be rotated counterclockwise about reference origin

(k=0,1 ..., r-1) degree obtains the new volume coordinate position (x of this coupling local feature _i ^k, y _i ^k) shown in the following formula:

$\left(\begin{array}{c}{x_i}^k\\ {y_i}^k\end{array}\right)=\left(\begin{array}{cc}\cos \left(\mathrm{\&theta;}\right)& \sin \left(\mathrm{\&theta;}\right)\\ -\sin \left(\mathrm{\&theta;}\right)& \cos \left(\mathrm{\&theta;}\right)\end{array}\right)\&CenterDot;\left(\begin{array}{c}{x}_i\\ y_i\end{array}\right);$

Wherein, r is the included sector region number of each quadrant, i=0, and 1 ..., N-1, wherein N is the right total number of matching characteristic;

Said space encoding module, specifically the relative tertiary location between the coupling local feature in the image is carried out space encoding through following mode:

Adopt the following formula coding to obtain horizontal direction space code figure:

$\mathrm{GX}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {x_i}^k<{x_j}^k\\ 1& \mathrm{if}& {x_i}^k\&GreaterEqual;{x_j}^k\end{array}\right.;$

Adopt the following formula coding to obtain vertical direction space code figure:

$\mathrm{GY}(i,j,k)=\left\{\begin{array}{ccc}0& \mathrm{if}& {y_i}^k<{y_j}^k\\ 1& \mathrm{if}& {y_i}^k\&GreaterEqual;{y_j}^k\end{array}\right.;$

Wherein, x _i ^kAnd x _j ^kBe respectively k postrotational horizontal ordinate of i coupling local feature and j coupling local feature, y _i ^kAnd y _j ^kBe respectively k postrotational ordinate of i coupling local feature and j coupling local feature; Confirm each coupling local feature after image is divided with respect to reference origin through horizontal direction space code figure GX and vertical direction space code figure GY, quadrant that is positioned at and sector region.

19. device according to claim 12 is characterized in that, said Space Consistency inspection module specifically comprises:

The XOR unit is used for said query image space code figure and matching image space code figure are carried out xor operation;

The erroneous matching delete cells, be used for confirming and deletion wrong matching characteristic right;

Matching number is confirmed the unit, is used for confirming that remaining coupling local feature is the right number of coupling local feature that meets Space Consistency to number.

20. device according to claim 19 is characterized in that:

Said XOR unit, specifically according to shown in the following formula, right space code figure carries out xor operation to the coupling local feature:

$V_x(i,j,k)={\mathrm{GX}}_q(i,j,k)\&CirclePlus;{\mathrm{GX}}_m(i,j,k);$

$V_y(i,j,k)={\mathrm{GY}}_q(i,j,k)\&CirclePlus;{\mathrm{GY}}_m(i,j,k);$

Wherein, (GX _q, GY _q) be query image space code figure, (GX _m, GY _m) be matching image space code figure;

Said erroneous matching unit, it is right that the coupling local feature that does not meet Space Consistency is obtained and deleted to employing with following formula:

Order $S_x\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_x(i,j,k),$ $S_y\left(i\right)=\underset{j=0}{\overset{N-1}{\mathrm{\&Sigma;}}}\underset{k=0}{\overset{r-1}{\mathrm{\&cup;}}}{V}_y(i,j,k),$

If exist i to make S _xOr S (i)＞0 _y(i)＞0, definition

Or

And delete i ^*Right to local matching characteristic;

Iteration carry out above-mentioned deletion action, the matching characteristic that to the last remains is to pairing S _xAnd S _yIn value be complete zero;

Said matching number is confirmed the unit, obtain the right number of matching characteristic that is left to meet the demands, as the right number of coupling local feature that meets Space Consistency.

21. device according to claim 12 is characterized in that, said result for retrieval returns module, specifically comprises:

Similarity calculated is used for according to the right number of matching characteristic that meets Space Consistency, and the number of the local feature that is comprised separately in matching image and the query image, calculates the similarity between matching image and query image;

Matching image returns the unit, is used for according to the said matching image of said similarity size ordering, and returns said matching image according to said ordering.

22. device according to claim 21 is characterized in that:

Said similarity calculated, specifically calculate the similarity between matching image and query image in the following way:

If query image q and matching image p through inquiry after, a is arranged to the coupling local feature, wherein b has passed through the Space Consistency check to mating local feature, the similarity of query image q and matching image p then, S (p q) is shown below:

$S(p,q)=b-\frac{a-b+1}{a}\&CenterDot;\frac{N\left(p\right)}{N_{\max }};$

Wherein N (p) representes local feature total number among the matching image p, N _MaxIn the presentation video database image maximum local feature numbers that possibly contain.

23. an image indexing system is characterized in that, comprising:

Any described image retrieving apparatus of claim 12 to 22;

Image data base is used to store the image that supplies match retrieval.

24. image indexing system according to claim 23, its characteristic with, also comprise:

Image collection module is used for obtaining new image from network, and stores in the said image data base;

The inverted list update module is used for obtaining the corresponding inverted list of vision dictionary that said new image comprises, and upgrades image related in the said inverted list.

Images