CN103514434B - Method and device for identifying image - Google Patents
Method and device for identifying image Download PDFInfo
- Publication number
- CN103514434B CN103514434B CN201210227208.8A CN201210227208A CN103514434B CN 103514434 B CN103514434 B CN 103514434B CN 201210227208 A CN201210227208 A CN 201210227208A CN 103514434 B CN103514434 B CN 103514434B
- Authority
- CN
- China
- Prior art keywords
- characteristic
- image
- contrast
- classification
- contrast characteristic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method and device for identifying an image. The method and device solve the problems that in an existing image identification process, more resources are consumed, the efficiency is quite low, and the time cost is quite large. The method comprises the steps of establishing a first voting image for one image of each category, being capable of obtaining k comparison characteristics matched with practical characteristics, and obtaining the categories of the comparison characteristics and the relative central positions of the comparison characteristics, so that the various categories are obtained through one-time matching. Subsequently, the similarity can be calculated for the comparison characteristics of each category, the estimated central positions of the comparison characteristics in the first voting images are determined, then the similarities can be increased at the estimated central positions, and finally the first voting images of the various categories can be obtained. Then, the first voting image, corresponding to the estimated central position, with the maximum similarity is obtained, and the category of the corresponding first voting image is the category of the image to be detected.
Description
Technical field
The application is related to data processing technique, more particularly to a kind of image-recognizing method and device.
Background technology
Image recognition, is image to be processed using computer, analyzed and is understood, to recognize the mesh of various different modes
The technology of mark and object.Generally image is all processed into the vector of a N-dimensional in computer vision application, thus it is many with regard to
The method of object detection is also to set up on this basis.But the gray feature for depending only on image pixel is inevitable
There are many defects.
Therefore, Serge Belongie et al. are proposed a kind of based on SC using the edge and shape information of image（Shape
Context, boundary profile）The detection method of feature.Liming Wang et al. propose that one kind passes through on the basis of SC features again
The method that voting method estimates object center and confidence level.This method obtains first each sampling of altimetric image to be checked
The SC features of point, because the consistent point of relative position on similar fitgures generally has similar SC features, therefore can be special by SC
The SC features levied in the preset feature lexicon are matched, and determine corresponding similarity.Obtained according to the similarity
To the contrast images of matching.
But, this method is identified every time only for a class contrast images, i.e., only include in preset feature lexicon
Whether the SC features of such contrast images, can only also differentiate the altimetric image to be checked comprising such contrast images in identification process.
In actual process, contrast images may have n classes, then need to set up n feature lexicon, and be directed to each altimetric image to be checked,
To determine the classification belonging to the altimetric image to be checked, then n identification process is repeated.
Therefore, above-mentioned identification process can consume more resources, and efficiency is very low, and time overhead is very big.
The content of the invention
The application provides a kind of image-recognizing method and device, and with the identification process for solving conventional images identification ratio can be consumed
More resource, and efficiency is very low, the very big problem of time overhead.
In order to solve the above problems, this application discloses a kind of image-recognizing method, including：
One first ballot image is set up for the image of each classification；
Extract the actual characteristic of each sampled point of altimetric image to be checked and the physical location of the actual characteristic；
For each sampled point, the k contrast characteristic matched with the actual characteristic is obtained, and obtain the contrast characteristic
Classification and contrast characteristic relative center；
For the classification of each contrast characteristic, similarity is calculated according to the actual characteristic and the contrast characteristic；
For each contrast characteristic classification corresponding to the first ballot figure, according to the physical location of the actual characteristic and
The relative center of the contrast characteristic, determines estimation center of the contrast characteristic in the first ballot image, and
Increase the similarity in the corresponding estimation center position of the contrast characteristic；
The similarity of each estimation center position in each first ballot image is traveled through, the maximum estimation of similarity is obtained
The corresponding first ballot image in center, by the classification of the described first ballot image recognition altimetric image to be checked.
Preferably, described method also includes：
Contrast characteristic, the classification of the contrast characteristic and the contrast characteristic's for extracting each sampled point of contrast images
With respect to center, wherein, the relative center is the distance of sampled point object center in contrast images.
Preferably, described method also includes：
For the contrast characteristic of each sampled point, by carrying out n time clustering step by step to the contrast characteristic, n+2 levels are set up
Search tree, wherein, to search starting point, 2 to n+1 level nodes are the cluster centre of clusters at different levels to 1 grade of node of the search tree,
N+2 levels node be the contrast characteristic, n>1, n is positive integer.
Preferably, the k contrast characteristic matched with the actual characteristic is obtained, including：
Characteristic matching is carried out to the actual characteristic in the search tree, k matched with the actual characteristic is searched
Contrast characteristic.
Preferably, described method also includes：
Preset x zoom scale, for each zoom scale under same category a second ballot figure is set up respectively
Picture.
Preferably, it is described to be calculated after similarity according to the actual characteristic and the contrast characteristic, also include：
Distance in relative center of the zoom scale to the contrast characteristic is zoomed in and out, and obtains correspondence
Scaling center；
For contrast characteristic classification under the zoom scale second ballot image, by the reality of the actual characteristic
Position, according to the orientation angles in the relative center of the contrast characteristic and the scaling center line displacement is entered, and is thrown
Project estimation center of the contrast characteristic in the described second ballot image；
Estimation center position in the described second ballot image increases the similarity.
Preferably, described method also includes：
The similarity of each estimation center position in each second ballot image is traveled through, the maximum estimation of similarity is obtained
The corresponding second ballot image in center；
By the classification and yardstick of altimetric image to be checked described in the described second ballot image recognition.
Preferably, described method also includes：
For the classification of the altimetric image to be checked, whether identification has the classification using the user of the altimetric image to be checked
Access right.
Preferably, the contrast images are cartoon image and/or trademark image.
Accordingly, disclosed herein as well is a kind of pattern recognition device, including：
First ballot image sets up module, for setting up one first ballot image for the image of each classification；
Extraction module, for extracting the actual characteristic of each sampled point of altimetric image to be checked and the reality of the actual characteristic
Position；
Matching and acquisition module, for for each sampled point, obtaining k matched with the actual characteristic to bit
Levy, and obtain the classification of the contrast characteristic and the relative center of contrast characteristic；
Similarity calculation module, for for the classification of each contrast characteristic, according to the actual characteristic and the contrast
Feature calculation similarity；
It is determined that and add module, for for each contrast characteristic classification corresponding to first ballot figure, according to described
The physical location of actual characteristic and the relative center of the contrast characteristic, determine the contrast characteristic in the first ballot image
In estimation center, and estimate that center position adds the similarity the contrast characteristic is corresponding；
Obtain and identification module, for travel through each first ballot image in each estimate center position similarity,
The corresponding first ballot image in the maximum estimation center of similarity is obtained, it is to be detected by the described first ballot image recognition
The classification of image.
Compared with prior art, the application includes advantages below：
First, the application sets up one first ballot image for the image of each classification, and then actual characteristic is being entered
During row matching, the k contrast characteristic matched with the actual characteristic can be obtained, and obtain the classification of the contrast characteristic and right
Than the relative center of feature, therefore, the application just can get plurality of classes by once matching.Subsequently it is directed to each
The contrast characteristic of classification, can calculate similarity, and determine estimation centre bit of the contrast characteristic in the first ballot image
Put, then can increase the similarity in the estimation center position, the first throwing of multiple classifications may finally be got
Ticket image.Then the corresponding first ballot image in the maximum estimation center of similarity is obtained, the first ballot image
Classification is the classification of altimetric image to be checked.The application can once recognize plurality of classes, identification process save resources, and efficiency
Higher, time overhead is very low.
Secondly, prior art carries out characteristic matching by feature lexicon, and due to be matched one by one, therefore efficiency is very
It is low.The application, by carrying out n time clustering step by step to the contrast characteristic, sets up n+2 levels for the contrast characteristic of each sampled point
Search tree.To search starting point, 2 to n+1 level nodes are the cluster centre of clusters at different levels to 1 grade of node of the search tree, n+2 levels
Node is the contrast characteristic.Therefore, the application is when carrying out, by 1 grade of node of search tree, successively matched, can
Quickly to find the k contrast characteristic matched with the actual characteristic, the consumption of resource is further reduced, and matched
Efficiency is higher, reduces the time of matching process.
Again, prior art is typically only capable to identify the contrast images of consistent size, therefore in order to ensure in identification process
In do not affected by dimensional variation, therefore prior art can as far as possible cover all of size in feature lexicon, therefore can lead
Cause the feature in feature lexicon excessive, it is less efficient during characteristic matching.The application is prefixed x zoom scale, for same class
Each zoom scale under not sets up respectively one second ballot image.In identification process, it is only necessary to by the scaling
Yardstick is zoomed in and out to the relative center of the contrast characteristic, obtains corresponding scaling center, subsequently through
The physical location of the actual characteristic is added with the scaling center, to determine the contrast characteristic in the described second ballot
In estimation center.Thus the application can further reduce the consumption of resource, and matching efficiency is higher, reduces
Time of matching process.
Again, the application, can further to using described to be checked after the classification for having got the altimetric image to be checked
The user of altimetric image detected, detects whether it has the access right of the classification.Therefore the application can be used for for
The infringement detection application of image, application is very extensive.
Description of the drawings
Fig. 1 is a kind of image-recognizing method flow chart described in the embodiment of the present application；
Fig. 2 is contrast images schematic diagram in a kind of image-recognizing method described in the embodiment of the present application；
Fig. 3 is search tree schematic diagram in a kind of image-recognizing method described in the embodiment of the present application；
Fig. 4 is multi-class multiple dimensioned recognition methodss flow process in a kind of image-recognizing method described in the application preferred embodiment
Figure；
Fig. 5 is a kind of pattern recognition device structure chart described in the embodiment of the present application.
Specific embodiment
It is understandable to enable the above-mentioned purpose of the application, feature and advantage to become apparent from, it is below in conjunction with the accompanying drawings and concrete real
Apply mode to be described in further detail the application.
Prior art being identified only for a class contrast images every time, therefore, to determine and the mapping to be checked
As affiliated classification, then n identification process is repeated.Therefore resource consumption can be caused than larger, and efficiency is very low,
Time overhead is very big.
The application provides a kind of image-recognizing method, can once recognize plurality of classes, identification process save resources, and
Efficiency is higher, and time overhead is very low.
With reference to Fig. 1, a kind of image-recognizing method flow chart described in the embodiment of the present application is given.
Step 11, for the image of each classification one first ballot image is set up；
The application is not limited for the classification of image, and for example, the classification of cartoon image can include Donald Duck
（Donald duck）、Mickey（Micky Mouse）With Hello Kitty（Hello Kitty）Deng and for example, for trademark image, each trade mark can
To regard a classification as.
The application sets up one first ballot image for the image of each classification, and described first is voted in image
Pixel value a little be initialized as 0.In actual treatment, the object in the cartoon image of each classification, such as Donald Duck can
Can there are various different forms, but the classification of image is only considered when the first ballot image is set up, not consider thing in image
The form of body.I.e. no matter which kind of form objects in images is probably, and all sets up a ballot image just for a classification.And scheme
The processing method of the morphology issues of object can be as in：Under each classification, for the form of each object one is set up
Contrast images, so as to the form that can be directed to each object in contrast images has its corresponding contrast characteristic.
In actual treatment, an image can be regarded as a matrix, in image a pixel regards matrix as
In an element, therefore, when initial, the value of image each pixel of voting is 0, that is, in the matrix of image of voting each
The value of element is all 0.
Step 12, extracts the actual characteristic and the physical location of the actual characteristic of each sampled point of altimetric image to be checked；
For altimetric image to be checked, several sampled points are pre-set, therefore can extract each of altimetric image to be checked adopt
The actual characteristic of sampling point, at the same time it can also extract the physical location of the actual characteristic, i.e., described sampled point is in picture to be detected
In physical location.The physical location is coordinate position of the sampled point in altimetric image to be checked, it is assumed that altimetric image to be checked is left
The summit at upper angle is used as zero（0,0）, correspondence sampled point coordinate be（x,y）, now x is nonnegative number, and y is non-positive number.
Certainly, zero can also be defined as other positions, and the application is not limited this.
Wherein, sampled point is chosen in the picture according to a fixed step size, therefore, the size of altimetric image to be checked is different, adopts
The number of sampling point is just different.
Step 13, for each sampled point, obtains the k contrast characteristic matched with the actual characteristic, and obtains described
The classification of contrast characteristic and the relative center of contrast characteristic；
For each sampled point, the k contrast characteristic matched with the actual characteristic is obtained, while also to obtain described
The classification of contrast characteristic, and the relative center of the contrast characteristic.
Wherein, the contrast characteristic is the feature of the sampled point got from contrast images；The class of the contrast characteristic
Not Wei the affiliated contrast images of the contrast characteristic classification；The relative center of the contrast characteristic is the contrast characteristic couple
The distance and bearing angle at object center in the sampled point answered and contrast images.
With reference to Fig. 2, contrast images schematic diagram in a kind of image-recognizing method described in the embodiment of the present application is given.
It is the contrast images of Donald Duck classifications in Fig. 2 by taking cartoon image as an example, A is represented in contrast images in Fig. 2
Object, i.e. Donald Duck；A1 represents the center of the object in contrast images, the i.e. center of Donald Duck；A2 is represented
A sampled point in contrast images.
Then the feature of A2 is contrast characteristic, and the classification of contrast characteristic is Donald Duck classifications, the phase of contrast characteristic
It is the distance between A2 and A1 and orientation angles to center.Hypothesis A1 is origin（0,0）The coordinate of A2 is（x2,y2）, then
Distance in the relative center be sqrt [(x2)2+(y2)2], wherein sqrt represents extraction of square root, the relative center
Orientation angles in position are α=arttan (x2/y2).
It is of course also possible to use the summit in the contrast images upper left corner now obtains respectively A1's and A2 as zero
Coordinate, with above-mentioned calculating.
Wherein, pixel value not may be constructed object in contrast images for 0 point in contrast images.For example, cartoon image
In the contrast images of middle Donald Duck, the object in contrast images is Donald Duck.
Wherein, the application can adopt KNN（K-Nearest Neighbor algorithm, closest node）Algorithm,
If the great majority in the K in feature space most like (i.e. closest in feature space) sample of i.e. one sample belong to certain
One classification, then the sample fall within this classification.
Actual characteristic described herein and contrast characteristic can be SC（Shape Context, edge contour）Feature, institute
State SC and be characterized as that centered on certain point in image point sets up polar coordinate system, then polar coordinate are divided into into several different fan
Region, according to the distribution of central point surrounding pixel brightness value characteristic vector is calculated.
Step 14, it is similar with contrast characteristic calculating according to the actual characteristic for the classification of each contrast characteristic
Degree；
For the classification of each contrast characteristic, can be calculated according to the actual characteristic and the contrast characteristic similar
Degree, it is assumed that the value of the similarity is X.Wherein, the methods such as Euclidean distance, X 2 test can be adopted when calculating similarity, by
It is prior art in the computational methods of similarity, therefore the application is repeated no more.
By the value of above-mentioned calculated similarity, can be used to weigh the actual characteristic and the contrast characteristic
Similarity degree.
Step 15, for each contrast characteristic classification corresponding to first ballot figure, according to the reality of the actual characteristic
Border position and the relative center of the contrast characteristic, determine estimation center of the contrast characteristic in the first ballot image
Position, and increase the similarity in the corresponding estimation center position of the contrast characteristic；
Wherein, the estimation center is by thing in the first ballot image estimated by contrast characteristic and actual characteristic
The center of body, that is, the classification for assuming contrast characteristic is Donald Duck classifications, and the estimation center is in the first ballot figure
The center of the Donald Duck estimated as in.
The relative center of the contrast characteristic is the corresponding sampled point of the contrast characteristic and object in contrast images
The distance and bearing angle at center.If getting coordinate position of the corresponding sampled point of the contrast characteristic in contrast images,
Then the sampled point is into thing in contrast images according to the point that the orientation angles and distance of the relative center are projected
Body center.
The physical location of the actual characteristic is coordinate position of the corresponding sampled point of actual characteristic in altimetric image to be checked,
Now, if regarding the physical location of the actual characteristic the relative center of contrast characteristic as, by the actual characteristic
Physical location, i.e. the coordinate position of actual characteristic correspondence sampled point, according to the orientation angles in the relative center and
Distance is projected, and the point for being projected is the corresponding object center of contrast characteristic, i.e., described contrast characteristic is in the first ballot figure
Estimation center as in.
Due to the value of above-mentioned calculated similarity, can be used to weigh the actual characteristic and the contrast characteristic
Similarity degree, then may also used to weigh by the first ballot objects in images estimated by contrast characteristic and actual characteristic
The probability at center, therefore value X of the similarity can be added in the estimation center position.
Each sampled point of altimetric image to be checked extracts an actual characteristic in the application, each actual characteristic correspondence K
Contrast characteristic, therefore the corresponding classification of actual characteristic may be individual for [1, K].Even if likewise, belong to same category of two it is right
Than feature, the corresponding contrast images of lower different shape object of all categories are may be from, because the object center of different shape may
It is identical, it is also possible to different, therefore the estimation center calculated by above-mentioned two contrast characteristic may be identical, it is also possible to
Difference, depending on specific data.If the estimation center that above-mentioned two contrast characteristic calculates is identical, correspond in institute
The value for estimating center superposition correspondence similarity is stated, i.e., the described similarity for estimating center position, is multiple contrasts
The similarity that feature and actual characteristic are calculated, the estimation center position is added to successively to be carried out after cumulative summation
As a result.
I.e. in actual treatment, if calculating behind the corresponding estimation center of certain contrast characteristic, the estimation centre bit
Put that place is added to cross similarity, then directly carry out on the basis of the similarity cumulative.
Step 16, each estimates the similarity of center position to travel through each first ballot image, obtains similarity maximum
The corresponding first ballot image in estimation center, by the classification of the described first ballot image recognition altimetric image to be checked.
By above-mentioned step, in the first ballot figure corresponding to the classification of each contrast characteristic, all there are some and estimate
Meter center.
Therefore, it can travel through the similarity of each estimation center position in each first ballot image, for example, first
The similarity of each estimation center position is respectively 10,12,15 in ballot image 1, in the first ballot image 2 in each estimation
Similarity at heart position is respectively 20,13,31, then subsequently all first ballot images in obtain all similarity intermediate values most
The corresponding first ballot image in big estimation center, that is, go up all similarity intermediate values in example and be 31, corresponding first to the maximum
Ballot image is the first ballot image 2.Now can be using the classification of the described first ballot image as the altimetric image to be checked
Classification.As above in example, the classification of the first ballot image 2 is Hello Kitty, then the classification of the altimetric image to be checked is also
Hello Kitty。
In sum, the application sets up one first ballot image for the image of each classification, then to actual spy
Levy when being matched, the k contrast characteristic matched with the actual characteristic can be obtained, and obtain the classification of the contrast characteristic
With the relative center of contrast characteristic, therefore, the application just can get plurality of classes by once matching.Subsequently it is directed to
The contrast characteristic of each classification, can calculate similarity, and determine in estimation of the contrast characteristic in the first ballot image
Heart position, then can increase the similarity in the estimation center position, may finally get the of multiple classifications
One ballot image.Then the corresponding first ballot image in the maximum estimation center of similarity, the first ballot figure are obtained
The classification of picture is the classification of altimetric image to be checked.The application can once recognize plurality of classes, identification process save resources, and
Efficiency is higher, and time overhead is very low.
Preferably, the contrast characteristic of each sampled point that extracts contrast images, the classification of the contrast characteristic and described right
Than the relative center of feature, wherein, the relative center is object centre bit of the sampled point away from contrast images
The distance put.
Training sample set can be pre-build in the application, the sample in the training sample set is comparison diagram
Picture.
Then the contrast characteristic that each sampled point of contrast images can be extracted, extracts the classification of the contrast characteristic, i.e.,
The classification of the affiliated contrast images of the contrast characteristic, while extract the relative center of the contrast characteristic, it is described relatively in
Heart position is the distance and bearing angle of the sampled point and the object center of contrast images.
Wherein, the relative center is discussed referring to the correlation of Fig. 2, and here is omitted.
In actual treatment, the object in the image of each classification may have different forms, therefore under each classification, can
One contrast images is set up with the form for each object so that the comparison that the sample in training sample set is covered is complete
Face, and then the contrast characteristic of sampled point can be extracted from the corresponding contrast images of the object of each different shape so that correspondence
Contrast characteristic it is also relatively more comprehensive, be the more comprehensive foundation of offer of follow-up image recognition, improve accuracy.
Preferably, the contrast images are cartoon image and/or trademark image.
Contrast images described herein can be cartoon image, or trademark image, be likely to certainly as cartoon image and
Trademark image.
The classification of the cartoon image can be specific cartoon figure, and the trademark image can be each trade mark,
The application is not limited this.
With reference to Fig. 3, search tree schematic diagram in a kind of image-recognizing method described in the embodiment of the present application is given.
Preferably, for the contrast characteristic of each sampled point, by carrying out n time clustering step by step to the contrast characteristic, build
Vertical n+2 levels search tree, wherein, to search starting point, 2 to n+1 level nodes are the poly- of clusters at different levels to 1 grade of node of the search tree
Class center, n+2 levels node be the contrast characteristic, n>1, n is positive integer.
Prior art carries out characteristic matching by feature lexicon, i.e., include what is extracted from contrast images in feature lexicon
All contrast characteristics, are to travel through wherein all contrast characteristics in characteristic matching, are matched one by one, and efficiency is very low.
And the application can be clustered, for example, using K- for the contrast characteristic of said extracted to the contrast characteristic
Means clustering methods.What lower mask body discussed search tree sets up process.
1）1 grade of node of the search tree is set to search starting point；
2）1 grade of cluster is carried out using K-means methods to the contrast characteristic, K is got1Individual 1 grade of cluster centre, will be every
One 1 grade of cluster centre is used as 2 grades of nodes；
3）2 grades of clusters are carried out to the contrast characteristic under each 1 grade of cluster centre, K is got2Individual 2 grades of cluster centres, will be every
One 2 grades of cluster centre is used as 3 grades of nodes；
4）By that analogy, n level clusters are carried out to the contrast characteristic under each n-1 level cluster centre, gets Kn-1Individual n levels
Cluster centre, using each n levels cluster centre as a n+1 level node, until n levels cluster centre cannot carry out subordinate's cluster
Till.Then n+2 levels node is the contrast characteristic.
Wherein, n>1, n is positive integer.
It is, of course, also possible to the size of default n, then stops cluster after default size is reached.
After by traveling through above-mentioned search tree, the application can be opened when characteristic matching is carried out by the 1 of search tree grade of node
Begin, successively matched, can quickly find the k contrast characteristic matched with the actual characteristic, further reduce
The consumption of resource, and matching efficiency is higher, reduces the time of matching process.
With reference to Fig. 4, multi-class multiple dimensioned knowledge in a kind of image-recognizing method described in the application preferred embodiment is given
Other method flow diagram.
Step 401, preset x zoom scale sets up respectively one for each zoom scale under same category
Two ballot images；
Because the size of image is variable, and prior art is typically only capable to identify the contrast images of consistent size, because
This is in order to ensure in identification process not affected by dimensional variation, therefore prior art can as far as possible cover in feature lexicon
All of size, therefore the feature in feature lexicon can be caused excessive, it is less efficient during characteristic matching.
The application time wasted in order to further reduce matching, the consumption of resource is reduced, be prefixed x pantograph
Degree, then sets up respectively one second ballot image for each zoom scale under same category.
For example, if zoom scale has 3, respectively 0.5,1 and 2.3.Now to set up 3 second for each classification
Ballot image, wherein, the second ballot correspondence zoom scale of image 1 is 0.5；The second ballot correspondence zoom scale of image 2 is 1；The
The two ballot correspondence zoom scale of image 3 are 2.3.And, although each corresponding zoom scale of the second ballot image is different, but
The size of each the second ballot image is identical.
Step 402, extracts the actual characteristic and the physical location of the actual characteristic of each sampled point of altimetric image to be checked；
Step 403, for each sampled point, characteristic matching is carried out in the search tree to the actual characteristic, is searched
The k contrast characteristic matched with the actual characteristic, obtains the classification of the contrast characteristic and the relative centre bit of contrast characteristic
Put；
For each sampled point, a sampled point corresponds to an actual characteristic, therefore can be in the search tree to institute
Stating actual characteristic carries out characteristic matching, the classification according to the actual characteristic, searches k matched with the actual characteristic
Contrast characteristic, obtains the classification of the contrast characteristic and the relative center of contrast characteristic.
Step 404, it is similar with contrast characteristic calculating according to the actual characteristic for the classification of each contrast characteristic
Degree；
Step 405, the distance in relative center of the zoom scale to the contrast characteristic is zoomed in and out,
Obtain corresponding scaling center；
The change of graphical rule, can only change in image the distance between two points, without changing image in two points
Between orientation angles.Therefore, zoom scale is only zoomed in and out to the distance in the relative center of the contrast characteristic, and
The change of the angle in relative center will not be caused.
For example, the distance in the relative center of the contrast characteristic is 20, and the zoom scale is 0.5, then correspond to
Scaling center be 20*0.5=10.
Due to the size of the corresponding second ballot image of different zoom yardstick it is identical, therefore by zoom scale to described right
After zooming in and out than the distance in the relative center of feature, the corresponding scaling centre bit of each zoom scale can be got
Put.
The application is zoomed in and out to the distance in the relative center of contrast characteristic, only by simply arithmetic operator,
Identification process mesoscale can just be solved and change brought impact, the relative matching process with mechanization, it is possible to reduce a large amount of
Time, improve identification efficiency.
Step 406, for the second ballot image of the classification under the zoom scale of contrast characteristic, will be described actual special
The physical location levied, is carried out according to the orientation angles in the relative center of the contrast characteristic and the scaling center
Skew, projects estimation center of the contrast characteristic in the described second ballot image；
For contrast characteristic classification under the zoom scale second ballot image, it is above-mentioned to have got the contrast
Scaling center of the feature under correspondence zoom scale, then now under the zoom scale in the center of contrast characteristic
Orientation angles it is constant, but the distance in the center of contrast characteristic be changed to scale center.
If regarding the physical location of the actual characteristic the relative center of contrast characteristic as, will be described actual special
The coordinate position of the physical location levied, i.e. actual characteristic correspondence sampled point, according in the relative center of the contrast characteristic
Distance and bearing angle enter line displacement, will actual characteristic correspondence sampled point coordinate position, according under the zoom scale
Orientation angles and the scaling center in the center of contrast characteristic enter line displacement, and the point after skew projects described
In the second ballot image under zoom scale, estimation centre bit of the as described contrast characteristic in the described second ballot image
Put.
Step 407, the estimation center position in the described second ballot image increases the similarity.
Estimation center position in the described second ballot image, addition is calculated by contrast characteristic and actual characteristic
The value of the similarity for arriving.If the similarity is 10, before addition, if the value for estimating center position is 0, estimate after addition
The value of center position is 10；Before addition, if the value for estimating center position is 13, the value of center position is estimated after addition
For 23.
Step 408, travels through the similarity of each estimation center position in each second ballot image, obtains similarity most
The corresponding second ballot image in big estimation center；
The similarity of each estimation center position in each second ballot image is traveled through, in all second ballot images
It is middle to obtain the corresponding second ballot image in the maximum estimation center of all similarity intermediate values, herein with discuss at above-mentioned steps 16
State basically identical, therefore repeat no more.
Step 409, by the classification and size of altimetric image to be checked described in the described second ballot image recognition；
The corresponding second ballot image in the maximum estimation center of above-mentioned acquisition similarity, then the second ballot image
Classification and size, the classification and size of as described altimetric image to be checked.
Certainly, in actual treatment, after traveling through all second ballot images, the similarity for estimating center can also be obtained
Second ballot image of n positions before ranking（N is positive integer）, then according to the phase for estimating center of the described second ballot image
Whether collect moderate condition like the value and its distribution spent, the second ballot image being determined for compliance with, detect the second ballot image
Whether classification is the classification of the altimetric image to be checked.
Step 410, for the classification of the altimetric image to be checked, recognizes whether the altimetric image to be checked has the classification
Access right.
The classification of the altimetric image to be checked is got by above-mentioned calculating, therefore can further have been detected using to be checked
The user of altimetric image, if the access right with the classification.Detect whether the user has the behavior of infringement.
For example, in electronic transaction website, the classification for having got the altimetric image to be checked is Hello Kitty, can be with
Further detection uses the user of altimetric image to be checked（That is seller）It is whether the appointment of agent of Hello Kitty.
In sum, prior art carries out characteristic matching by feature lexicon, due to be matched one by one, therefore efficiency
It is very low.The application, by carrying out n time clustering step by step to the contrast characteristic, sets up n+2 for the contrast characteristic of each sampled point
Level search tree.To search starting point, 2 to n+1 level nodes are the cluster centre of clusters at different levels to 1 grade of node of the search tree, n+2
Level node is the contrast characteristic.Therefore, the application is when carrying out, by 1 grade of node of search tree, successively matched,
The k contrast characteristic matched with the actual characteristic can be quickly found, the consumption of resource is further reduced, and
It is higher with efficiency, reduce the time of matching process.
Secondly, prior art is typically only capable to identify the contrast images of consistent size, therefore in order to ensure in identification process
In do not affected by dimensional variation, therefore prior art can as far as possible cover all of size in feature lexicon, therefore can lead
Cause the feature in feature lexicon excessive, it is less efficient during characteristic matching.The application is prefixed x zoom scale, for same class
Each zoom scale under not sets up respectively one second ballot image.In identification process, it is only necessary to by the scaling
Yardstick is zoomed in and out to the relative center of the contrast characteristic, obtains corresponding scaling center, subsequently through
The physical location of the actual characteristic is added with the scaling center, to determine the contrast characteristic in the described second ballot
In estimation center.Thus the application can further reduce the consumption of resource, and matching efficiency is higher, reduces
Time of matching process.
Again, the application, can further to using described to be checked after the classification for having got the altimetric image to be checked
The user of altimetric image detected, detects whether it has the access right of the classification.Therefore the application can be used for for
The infringement detection application of image, application is very extensive.
With reference to Fig. 5, a kind of pattern recognition device structure chart described in the embodiment of the present application is given.
Accordingly, present invention also provides a kind of pattern recognition device, including：First ballot image is set up module 11, is carried
Delivery block 12, matching and acquisition module 13, similarity calculation module 14, determination and add module 15 and acquisition and identification module
16, wherein：
First ballot image sets up module 11, for setting up one first ballot image for the image of each classification；
Extraction module 12, for extracting the actual characteristic of each sampled point of altimetric image to be checked and the reality of the actual characteristic
Border position；
Matching and acquisition module 13, for for each sampled point, obtaining k matched with the actual characteristic to bit
Levy, and obtain the classification of the contrast characteristic and the relative center of contrast characteristic；
Similarity calculation module 14, for for the classification of each contrast characteristic, according to the actual characteristic and described right
Than feature calculation similarity；
It is determined that and add module 15, for for each contrast characteristic classification corresponding to first ballot figure, according to institute
The physical location of actual characteristic and the relative center of the contrast characteristic are stated, determines the contrast characteristic in the first ballot figure
Estimation center as in, and increase the similarity in the corresponding estimation center position of the contrast characteristic；
Obtain and identification module 16, for travel through each first ballot image in each estimate center position it is similar
Degree, obtains the corresponding first ballot image in the maximum estimation center of similarity, is treated by the described first ballot image recognition
The classification of detection image.
Preferably, described device also includes：
Extraction module, for extract the contrast characteristic of each sampled point of contrast images, the classification of the contrast characteristic and
The relative center of the contrast characteristic, wherein, the relative center is sampled point object in contrast images
The distance of center.
Search tree sets up module, for for the contrast characteristic of each sampled point, by carrying out n time to the contrast characteristic
Cluster step by step, set up n+2 level search trees, wherein, to search starting point, 2 to n+1 level nodes are 1 grade of node of the search tree
The cluster centre of clusters at different levels, n+2 levels node be the contrast characteristic, n>1, n is positive integer.
The matching and acquisition module 13, for carrying out characteristic matching to the actual characteristic in the search tree, look into
Look for the k contrast characteristic matched with the actual characteristic.
Preferably, described device also includes：
Second ballot image sets up module, for preset x zoom scale, for each pantograph under same category
Degree sets up respectively one second ballot image.
It is described to determine and add module 15, including：
Scaling submodule, enters for the distance in the relative center according to the zoom scale to the contrast characteristic
Row scaling, obtains corresponding scaling center；
Center determination sub-module, for the second ballot figure for the classification of contrast characteristic under the zoom scale
Picture, by the physical location of the actual characteristic, according to orientation angles in the relative center of the contrast characteristic and described
Line displacement is entered in scaling center, projects estimation center of the contrast characteristic in the described second ballot image；
Addition submodule, for the estimation center position in the described second ballot image the similarity is increased.
Preferably, obtain and identification module 16, be additionally operable to travel through each estimation center in each second ballot image
The similarity at place, obtains the corresponding second ballot image in the maximum estimation center of similarity；Schemed by the described second ballot
As the classification and yardstick of the identification altimetric image to be checked.
Preferably, described device also includes：
Authority recognition module, for for the classification of the altimetric image to be checked, recognizing whether the altimetric image to be checked has
The access right of the classification.
Preferably, the contrast images are cartoon image and/or trademark image.
For system embodiment, due to itself and embodiment of the method basic simlarity, so description is fairly simple, it is related
Part is illustrated referring to the part of embodiment of the method.
Each embodiment in this specification is described by the way of progressive, what each embodiment was stressed be with
The difference of other embodiment, between each embodiment identical similar part mutually referring to.
Those skilled in the art are it should be appreciated that embodiments herein can be provided as method, system or computer program
Product.Therefore, the application can be using complete hardware embodiment, complete software embodiment or with reference to the reality in terms of software and hardware
Apply the form of example.And, the application can be adopted and wherein include the computer of computer usable program code at one or more
Usable storage medium（Including but not limited to disk memory, CD-ROM, optical memory etc.）The computer program of upper enforcement is produced
The form of product.
Although having been described for the preferred embodiment of the application, those skilled in the art once know basic creation
Property concept, then can make other change and modification to these embodiments.So, claims are intended to be construed to include excellent
Select embodiment and fall into having altered and changing for the application scope.
The application is with reference to method, the equipment according to the embodiment of the present application（System）And the flow process of computer program
Figure and/or block diagram are describing.It should be understood that can be by computer program instructions flowchart and/or each stream in block diagram
The combination of journey and/or square frame and flow chart and/or the flow process in block diagram and/or square frame.These computer programs can be provided
The processor of general purpose computer, special-purpose computer, Embedded Processor or other programmable data processing devices is instructed to produce
A raw machine so that produced for reality by the instruction of computer or the computing device of other programmable data processing devices
The device of the function of specifying in present one flow process of flow chart or one square frame of multiple flow processs and/or block diagram or multiple square frames.
These computer program instructions may be alternatively stored in can guide computer or other programmable data processing devices with spy
In determining the computer-readable memory that mode works so that the instruction being stored in the computer-readable memory is produced to be included referring to
Make the manufacture of device, the command device realize in one flow process of flow chart or one square frame of multiple flow processs and/or block diagram or
The function of specifying in multiple square frames.
These computer program instructions also can be loaded in computer or other programmable data processing devices so that in meter
Series of operation steps is performed on calculation machine or other programmable devices to produce computer implemented process, so as in computer or
The instruction performed on other programmable devices is provided for realizing in one flow process of flow chart or multiple flow processs and/or block diagram one
The step of function of specifying in individual square frame or multiple square frames.
Finally, in addition it is also necessary to explanation, herein, such as first and second or the like relational terms be used merely to by
One entity or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or operation
Between there is any this actual relation or order.And, term " including ", "comprising" or its any other variant meaning
Covering including for nonexcludability, so that a series of process, method, commodity or equipment including key elements not only includes that
A little key elements, but also including other key elements being not expressly set out, or also include for this process, method, commodity or
The intrinsic key element of equipment.In the absence of more restrictions, the key element for being limited by sentence "including a ...", does not arrange
Except also there is other identical element in including the process of the key element, method, commodity or equipment.
Above to a kind of image-recognizing method and device provided herein, it is described in detail, it is used herein
Specific case is set forth to the principle and embodiment of the application, and the explanation of above example is only intended to help and understands
The present processes and its core concept；Simultaneously for one of ordinary skill in the art, according to the thought of the application, in tool
Will change in body embodiment and range of application, in sum, this specification content should not be construed as to the application
Restriction.
Claims (10)
1. a kind of image-recognizing method, it is characterised in that include：
One first ballot image is set up for the image of each classification；
Extract the actual characteristic of each sampled point of altimetric image to be checked and the physical location of the actual characteristic；
For each sampled point, the k contrast characteristic matched with the actual characteristic is obtained, and obtain the class of the contrast characteristic
The relative center of other and contrast characteristic；
For the classification of each contrast characteristic, similarity is calculated according to the actual characteristic and the contrast characteristic；
For each contrast characteristic classification corresponding to the first ballot figure, according to the physical location of the actual characteristic and described
The relative center of contrast characteristic, determines estimation center of the contrast characteristic in the first ballot image, and in institute
Stating the corresponding estimation center position of contrast characteristic increases the similarity；
The similarity of each estimation center position in each first ballot image is traveled through, the maximum estimation center of similarity is obtained
The corresponding first ballot image in position, by the classification of the described first ballot image recognition altimetric image to be checked.
2. method according to claim 1, it is characterised in that also include：
Contrast characteristic, the classification of the contrast characteristic and the contrast characteristic's for extracting each sampled point of contrast images is relative
Center, wherein, the relative center is the distance and bearing angle of the sampled point and object center in contrast images
Degree.
3. method according to claim 2, it is characterised in that also include：
For the contrast characteristic of each sampled point, by carrying out n time clustering step by step to the contrast characteristic, the lookup of n+2 levels is set up
Tree, wherein, to search starting point, 2 to n+1 level nodes are the cluster centre of clusters at different levels, n+2 for 1 grade of node of the search tree
Level node be the contrast characteristic, n>1, n is positive integer.
4. method according to claim 3, it is characterised in that the k contrast characteristic that acquisition is matched with the actual characteristic,
Including：
Characteristic matching is carried out to the actual characteristic in the search tree, the k contrast matched with the actual characteristic is searched
Feature.
5. method according to claim 1, it is characterised in that also include：
Preset x zoom scale, for each zoom scale under same category one second ballot image is set up respectively.
6. method according to claim 5, it is characterised in that described to be counted according to the actual characteristic and the contrast characteristic
After calculating similarity, also include：
Distance in relative center of the zoom scale to the contrast characteristic is zoomed in and out, and obtains corresponding contracting
Put center；
For contrast characteristic classification under the zoom scale second ballot image, by the actual bit of the actual characteristic
Put, according to the orientation angles in the relative center of the contrast characteristic and the scaling center line displacement is entered, project
Go out estimation center of the contrast characteristic in the described second ballot image；
Estimation center position in the described second ballot image increases the similarity.
7. method according to claim 6, it is characterised in that also include：
The similarity of each estimation center position in each second ballot image is traveled through, the maximum estimation center of similarity is obtained
The corresponding second ballot image in position；
By the classification and yardstick of altimetric image to be checked described in the described second ballot image recognition.
8. according to the arbitrary described method of claim 1 or 7, it is characterised in that also include：
For the classification of the altimetric image to be checked, recognize using whether the user of the altimetric image to be checked has making for the classification
Use authority.
9. according to the arbitrary described method of claim 2 to 4, it is characterised in that the contrast images are cartoon image and/or business
Logo image.
10. a kind of pattern recognition device, it is characterised in that include：
First ballot image sets up module, for setting up one first ballot image for the image of each classification；
Extraction module, for extracting the actual characteristic of each sampled point of altimetric image to be checked and the actual bit of the actual characteristic
Put；
Matching and acquisition module, for for each sampled point, obtaining the k contrast characteristic matched with the actual characteristic, and
Obtain the classification of the contrast characteristic and the relative center of contrast characteristic；
Similarity calculation module, for for the classification of each contrast characteristic, according to the actual characteristic and the contrast characteristic
Calculate similarity；
It is determined that and add module, for for each contrast characteristic classification corresponding to first ballot figure, according to the reality
The physical location of feature and the relative center of the contrast characteristic, determine the contrast characteristic in the first ballot image
Estimate center, and increase the similarity in the corresponding estimation center position of the contrast characteristic；
Obtain and identification module, for travel through each first ballot image in each estimate center position similarity, obtain
The corresponding first ballot image in the maximum estimation center of similarity, by the described first ballot image recognition altimetric image to be checked
Classification.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210227208.8A CN103514434B (en) | 2012-06-29 | 2012-06-29 | Method and device for identifying image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210227208.8A CN103514434B (en) | 2012-06-29 | 2012-06-29 | Method and device for identifying image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103514434A CN103514434A (en) | 2014-01-15 |
| CN103514434B true CN103514434B (en) | 2017-04-12 |
Family
ID=49897133
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210227208.8A Active CN103514434B (en) | 2012-06-29 | 2012-06-29 | Method and device for identifying image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103514434B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110019898A (en) * | 2017-08-08 | 2019-07-16 | 航天信息股份有限公司 | A kind of animation image processing system |
| CN111026641B (en) * | 2019-11-14 | 2023-06-20 | 北京云聚智慧科技有限公司 | Picture comparison method and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833672A (en) * | 2010-04-02 | 2010-09-15 | 清华大学 | Sparse representation face identification method based on constrained sampling and shape feature |
| CN102521565A (en) * | 2011-11-23 | 2012-06-27 | 浙江晨鹰科技有限公司 | Garment identification method and system for low-resolution video |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5984327B2 (en) * | 2010-07-24 | 2016-09-06 | キヤノン株式会社 | Information processing method and apparatus, and program |
-
2012
- 2012-06-29 CN CN201210227208.8A patent/CN103514434B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101833672A (en) * | 2010-04-02 | 2010-09-15 | 清华大学 | Sparse representation face identification method based on constrained sampling and shape feature |
| CN102521565A (en) * | 2011-11-23 | 2012-06-27 | 浙江晨鹰科技有限公司 | Garment identification method and system for low-resolution video |
Non-Patent Citations (1)
| Title |
|---|
| 加权K近邻和加权投票相结合的虹膜识别算法;孙彩堂,等.;《小型微型计算机***》;20100930;第31卷(第9期);1846-1849 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103514434A (en) | 2014-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Efficient global 2d-3d matching for camera localization in a large-scale 3d map | |
| CN111241989B (en) | Image recognition method and device and electronic equipment | |
| EP3074918B1 (en) | Method and system for face image recognition | |
| CN105956560A (en) | Vehicle model identification method based on pooling multi-scale depth convolution characteristics | |
| CN108549870A (en) | A kind of method and device that article display is differentiated | |
| CN104915673B (en) | A kind of objective classification method and system of view-based access control model bag of words | |
| JP2016062610A (en) | Feature model creation method and feature model creation device | |
| CN107610177B (en) | The method and apparatus of characteristic point is determined in a kind of synchronous superposition | |
| CN104036287A (en) | Human movement significant trajectory-based video classification method | |
| CN110070090A (en) | A kind of logistic label information detecting method and system based on handwriting identification | |
| CN111915015B (en) | Abnormal value detection method and device, terminal equipment and storage medium | |
| CN110751027B (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| CN103793926A (en) | Target tracking method based on sample reselecting | |
| CN113870254B (en) | Target object detection method and device, electronic equipment and storage medium | |
| CN107798691A (en) | A kind of unmanned plane independent landing terrestrial reference real-time detecting and tracking method of view-based access control model | |
| CN110942473A (en) | Moving target tracking detection method based on characteristic point gridding matching | |
| JP2016045884A (en) | Pattern recognition device and pattern recognition method | |
| Juang et al. | Stereo-camera-based object detection using fuzzy color histograms and a fuzzy classifier with depth and shape estimations | |
| CN111373393A (en) | Image retrieval method and device and image library generation method and device | |
| Abdullah et al. | Vehicle counting using deep learning models: a comparative study | |
| CN113496260A (en) | Grain depot worker non-standard operation detection method based on improved YOLOv3 algorithm | |
| CN103514434B (en) | Method and device for identifying image | |
| CN105190689A (en) | Image processing including adjoin feature based object detection, and/or bilateral symmetric object segmentation | |
| CN116704490B (en) | License plate recognition method, license plate recognition device and computer equipment | |
| Thu et al. | Pyramidal Part‐Based Model for Partial Occlusion Handling in Pedestrian Classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 1191718 Country of ref document: HK |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: GR Ref document number: 1191718 Country of ref document: HK |