CN105844290B - The method and device of multiple same objects in matching image - Google Patents
The method and device of multiple same objects in matching image Download PDFInfo
- Publication number
- CN105844290B CN105844290B CN201610153710.7A CN201610153710A CN105844290B CN 105844290 B CN105844290 B CN 105844290B CN 201610153710 A CN201610153710 A CN 201610153710A CN 105844290 B CN105844290 B CN 105844290B
- Authority
- CN
- China
- Prior art keywords
- point
- match
- image
- point set
- given object
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The application is the method and device about same objects multiple in a kind of matching image.The method of multiple same objects includes: to extract the feature point set of given object as fisrt feature point set in a kind of matching image, extracts the feature point set of image to be matched as second feature point set；The fisrt feature point set is matched with the characteristic point that the second feature point is concentrated, obtains the matching point set in the image to be matched；Classified to the matching point set according to distribution density using clustering algorithm, forms multiple match point subsets.The technical solution of the application can make classification more accurate.
Description
Technical field
This disclosure relates in field of image recognition more particularly to matching image multiple same objects method and device.
Background technique
It is very widely used that object in image etc. is searched in the picture.Such as when in graphic user interface have multiple poles
When for similar or even identical given object, all matching target queries for generally requiring to be included by graphic user interface go out
Come.
The relevant technologies provide a kind of multi-targets recognition algorithm, which forms SIFT feature with SUSAN angle point, use
Ladder image pyramid structure realizes Scale invariant, establishes unified overdetermined linear system for all match points and to the equation
Group coefficient matrix, which simplify, reduces its dimension by half, obtains augmented matrix.Rank transformation is carried out to augmented matrix, according to coordinate
The characteristic of conversion can therefrom extract the stabilization normal point of multiple target, realize the match point of quick separating multiple target.But it is above-mentioned
Method robustness is insufficient.In addition, being handled for the match point after separation without reference to relevant range acquisition.
Summary of the invention
The disclosure provides a kind of method and device of multiple same objects in matching image, can obtain with tag along sort
Match point, classify it is more acurrate.
According to the first aspect of the embodiments of the present disclosure, a kind of method of multiple same objects in matching image is provided, comprising:
The feature point set of given object is extracted as fisrt feature point set, extracts the feature point set of image to be matched as second feature point
Collection；
The fisrt feature point set is matched with the characteristic point that the second feature point is concentrated, is obtained described to be matched
Matching point set on image；
Classified to the matching point set according to distribution density using clustering algorithm, forms multiple match point subsets.
In an embodiment, this method further include: the multiple match point subset is traversed, respectively according to each matching idea
The size of collection and the given object obtains the model of the target object to match in the image to be matched with the given object
It encloses.
In an embodiment, the multiple match point subset is traversed, respectively according to each match point subset and described given
The range that the size of object obtains the target object to match in the image to be matched with the given object includes:
Two match points A and B are arbitrarily chosen from match point subset Q, calculate the distance between described match point A and B work
For first distance；
Match point C and D corresponding with the match point A and B in the given object is obtained respectively, calculates the match point
The distance between C and D are used as second distance；
The ratio between the first distance and the second distance is calculated as the first matching ratio；
According to institute corresponding with the match point A in first matching ratio, the match point A and the given object
The coordinate value of match point C and the size of the given object are stated, the target object to match with the given object is obtained
Range.
In an embodiment, the range of the target object is determined by following formula:
RectA=(xa-xc*p,ya-yc*p,w*p,h*p)；
The wherein corresponding rectangular extent of RectA, four elements are successively in RectA are as follows: the lower left corner abscissa of rectangular extent,
The lower left corner ordinate of rectangular extent, the width of rectangular extent and the height of rectangular extent；
xaFor coordinate of the match point A in x-axis；
xcFor coordinate of the match point C in x-axis；
yaFor the coordinate of the match point A on the y axis；
ycFor the coordinate of the match point C on the y axis；
P is the matching ratio；
W is the width pixel value of the given object；
H is the height pixel value of the given object.
In an embodiment, the given object includes icon and button.
In an embodiment, the clustering algorithm includes DBSCAN algorithm, OPTICS algorithm or DENCLUE algorithm.
In an embodiment, the feature point set for giving object is extracted as fisrt feature point set using SIFT algorithm, and
The feature point set of image to be matched is extracted as second feature point set.
In an embodiment, SIFT algorithm includes: to check the given image using Gaussian convolution to be handled, and is obtained more
Scale space images；Difference of Gaussian processing is carried out to the multiscale space image, constructs Gaussian difference scale space image；Inspection
The Local Extremum for surveying the Gaussian difference scale space image, it is using three-dimensional quadratic function is fitted that the Local Extremum is smart
Sub-pixel is really arrived, the Local Extremum is screened to obtain feature point set using threshold method and Hessian matrix method.
According to the second aspect of an embodiment of the present disclosure, a kind of device of multiple same objects in matching image is provided, comprising:
Feature point set extraction unit, for extracting the feature point set of given object as fisrt feature point set, extract to
Feature point set with image is as second feature point set；
Point set acquiring unit is matched, the feature for concentrating the fisrt feature point set and the second feature point clicks through
Row matching, obtains the matching point set in the image to be matched；
Taxon is formed multiple for being classified to the matching point set according to distribution density using clustering algorithm
Match point subset.
In an embodiment, the device further include: range acquiring unit, for traversing the multiple match point subset, point
Not according to the size of each match point subset and the given object obtain in the image to be matched with the given object phase
The range of matched target object.
In an embodiment, the range acquiring unit, which is configured that from match point subset Q, arbitrarily chooses two match points
A and B calculates the distance between described match point A and B as first distance；
Match point C and D corresponding with the match point A and B in the given object is obtained respectively, calculates the match point
The distance between C and D are used as second distance；
The ratio between the first distance and the second distance is calculated as the first matching ratio；
According to institute corresponding with the match point A in first matching ratio, the match point A and the given object
The coordinate value of match point C and the size of the given object are stated, the target object to match with the given object is obtained
Range.
In an embodiment, the clustering algorithm includes DBSCAN algorithm, OPTICS algorithm or DENCLUE algorithm.
In an embodiment, the feature point set extraction unit is configured to execute SIFT algorithm.
In an embodiment, the feature point set extraction unit, which is configured that, checks the given image using Gaussian convolution
It is handled, obtains multiscale space image；Difference of Gaussian processing is carried out to the multiscale space image, constructs difference of Gaussian
Scale space images；The Local Extremum for detecting the Gaussian difference scale space image, will using three-dimensional quadratic function is fitted
The Local Extremum is accurate to sub-pixel, is sieved using threshold method and Hessian matrix method to the Local Extremum
Choosing obtains feature point set.
According to the technical solution of the embodiment of the present disclosure, the interference of noise can be effectively reduced, classification can be made more accurate.Separately
Outside, according to the technical solution of other embodiments, the method for obtaining target object range according to isolated match point is provided, is more met
The purpose of Object identifying.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows and meets implementation of the invention
Example, and be used to explain the principle of the present invention together with specification.
Fig. 1 is the flow chart of the method for multiple same objects in matching image shown according to an exemplary embodiment；
Fig. 2 is the schematic diagram of multiple same objects in matching image shown according to an exemplary embodiment；
Fig. 3 is the flow chart of the method for multiple same objects in matching image shown according to an exemplary embodiment；
Fig. 4 is the flow chart of the method shown according to an exemplary embodiment for calculating target object range；
Fig. 5 is the block diagram of the device of multiple same objects in matching image shown according to an exemplary embodiment.
Specific embodiment
Example embodiments are described in detail here, and the example is illustrated in the accompanying drawings.Following description is related to
When attached drawing, unless otherwise indicated, the same numbers in different drawings indicate the same or similar elements.Following exemplary embodiment
Described in embodiment do not represent all embodiments consistented with the present invention.On the contrary, they be only with it is such as appended
The example of device and method being described in detail in claims, some aspects of the invention are consistent.
Fig. 1 is the flow chart of the method for multiple same objects in matching image shown according to an exemplary embodiment.This
Embodiment is applicable to match multiple feelings with the given given consistent object of object shapes content from image to be matched
Condition.As shown in Figure 1, the method for multiple same objects includes: in matching image described in the present embodiment
In step s 110, the feature point set of given object is extracted as fisrt feature point set, extracts image to be matched
Feature point set is as second feature point set.
It should be noted that the given object can be for arbitrarily with the pictorial element of clear profile, such as icon, button
Deng.The image to be matched is to contain the image of multiple given objects, can be graphic user interface etc., such as with
The button or icon of image or text.
Those of ordinary skill in the art it should be clear that, extract the feature point set of given object as fisrt feature point
A variety of methods execution can be used as second feature point set in collection, and the feature point set of extraction image to be matched.For example, can be based on
SIFT algorithm, SURF algorithm, the marginal point algorithm of wavelet transformation and Harris Corner Detection Algorithm etc., the disclosure to this simultaneously
With no restriction.
It is illustrated for using SIFT algorithm below.
The feature point set of given image is extracted using SIFT algorithm can include: check the given image using Gaussian convolution
It is handled, obtains multiscale space image；Difference of Gaussian processing is carried out to the multiscale space image, constructs difference of Gaussian
Scale space images；The Local Extremum for detecting the Gaussian difference scale space image, will using three-dimensional quadratic function is fitted
The Local Extremum is accurate to sub-pixel, is sieved using threshold method and Hessian matrix method to the Local Extremum
Choosing obtains feature point set.
In the step s 120, the fisrt feature point set is matched with the characteristic point that the second feature point is concentrated,
Obtain the matching point set in the image to be matched.
Any Feature Points Matching can be used in the present embodiment, and the disclosure is without limitation.For example, two characteristic points can be calculated
Feature vector similarity determination of the Euclidean distance as characteristic point in two images (such as given object and image to be matched)
Measurement.For example, drawing the characteristic point A for determining object, it is corresponding European to calculate all validity feature points by traversal image to be matched
Distance determines that the point is the match point of point A if distance is less than some threshold value.
For SIFT algorithm, since the translation of SIFT feature, scaling invariance can solve different resolution and show
The compatibility issue for showing content identical image for different resolution but shows the identical or very much like image to be matched of content
Corresponding match point can be obtained with given object, so as to realize the compatibility of different resolution image.
In step s 130, multiple match point subsets can be formed by classifying to the matching point set.There are many clusters to calculate
Method can be used for classifying to matching point set, such as partitioning, stratification, density algorithm etc..
Partition clustering method is largely and to need to give a number of partitions K based on distance, uncertain for the number of partitions
The case where then need to be iterated analysis, to obtain optimal solution.For given K, algorithm provides an initial point first
Group method changes grouping by the method to iterate later, so that the grouping scheme after improving each time is all primary earlier above
It is good.And so-called good standard is exactly: the closer record in same grouping the better, and the remoter record in different grouping the better.It draws
The related algorithm of point-score includes K-MEANS algorithm, K-MEDOIDS algorithm, CLARANS algorithm etc..
Hierarchy clustering method (hierarchical methods) can be based on distance or based on density or connectivity
's.This method decompose as level to given data set, until certain condition meets.Hierarchy clustering method
Some extensions have also contemplated subspace clustering.The defect of hierarchical method is, once a step (merge or divide) is completed, it
It cannot be revoked, therefore its decision that cannot right the wrong.The related algorithm of stratification include BIRCH algorithm, CURE algorithm,
CHAMELEON algorithm etc..
The guiding theory of density algorithm is exactly, as long as soon as the density of the point in region adds it greater than some threshold value
Into similar cluster therewith.Density algorithm includes DBSCAN algorithm, OPTICS algorithm, DENCLUE algorithm etc..Based on density
A fundamental difference of method and other methods be: it is not based on various distances, but based on density.It is close
The shortcomings that degree algorithm can overcome the algorithm based on distance that can only find the cluster of " similar round ", for some making an uproar far from cluster
Point has good excretion, and the cluster of arbitrary shape can be found in noisy spatial database.It is calculated below with DBSCAN
Illustrate assorting process according to an embodiment of the invention for method.
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is
One more representational density-based algorithms.Different from division and hierarchy clustering method, cluster is defined as close by it
The maximum set of the connected point of degree can be cluster having region division highdensity enough, and can be in the spatial data of noise
The cluster of arbitrary shape is found in library.
According to an embodiment, classified to obtain the process of multiple match point subsets such as to match point using DBSCAN algorithm
Under.
It concentrates to appoint from the match point in image to be matched and takes a untreated point.
Centered on the point of the selection, scanned within the scope of radius E.Point of the point set within the scope of radius E will be matched
Quantity be compared with a threshold value MinPts (MinPts > 1).If the quantity of the point searched is greater than the threshold value, the choosing
The point taken is core point, and all the points searched in the core point and radius E form a cluster.To in cluster in addition to the core point
Other points repeat the above steps, and carry out expansion cluster, and so on, until completing the expansion cluster of all the points.Then, selection is next does not locate
The point of reason, repeats the above steps.Radius E and threshold value MinPts can be chosen according to concrete application, such as can be empirical value.
If the quantity of the point searched is less than the threshold value, the point of the selection is marginal point, then chooses next untreated
Point, repeat the above steps.
After all the points are processed, match point is classified according to their distribution density on the image and marks off
Come, obtains multiple match point subsets.
According to an embodiment of the invention, feature extraction algorithm in conjunction with clustering algorithm, can effectively improve algorithm robustness, subtract
Few noise bring interference, classifies more acurrate.
For example, Fig. 2 is the schematic diagram of multiple same objects in matching image shown according to an exemplary embodiment, Fig. 2
Left side dashed box in part 210 be image to be matched, small " plunders " button 220 in the dashed box of the upper right corner is to give object, to
After matching image 210 is matched with " plunder " button 220, according to distribution density, its match point can be divided into three classes, be located at institute
State three " plunder " button positions in image to be matched.
According to an embodiment, disclosed method may also include step S140.In step 140, the multiple matching is traversed
Point subset, respectively according to the size of each match point subset and the given object obtain in the image to be matched with it is described to
Determine the range for the target object that object matches.The method of this step can have a various implementations, the disclosure and with no restriction.
For example, being illustrated by taking a match point subset Q in the multiple match point subset as an example below.
Two match points A and B are arbitrarily chosen from match point subset Q, calculate the distance between described match point A and B work
For first distance.
Match point C and D corresponding with the match point A and B in the given object is obtained respectively, calculates the match point
The distance between C and D are used as second distance.It can be readily appreciated that given object (for example, small " plunder " in right side shown in Fig. 2 is pressed
Button) when carrying out arithmetic operation, bottom left vertex of the given object is chosen as coordinate origin.It is right for image to be matched
It is not particularly limited in the selection of coordinate origin.
The ratio between the first distance and the second distance is calculated as the first matching ratio.
According to institute corresponding with the match point A in first matching ratio, the match point A and the given object
The coordinate value of match point C and the size of the given object are stated, the target object to match with the given object is obtained
Range.
In this way, the range of the target object to match in the available image to be matched with the given object.
Traverse the multiple match point subset, as above operation executed to each match point subset, can obtain respectively it is described to
The range of the target object to match in matching image with the given object.
Fig. 3 is the process of the method for multiple same objects in a kind of matching image shown according to an exemplary embodiment
Figure.
As shown in figure 3, the method for multiple same objects includes: in matching image described in the present embodiment
In step S301, image to be matched is inputted.
In step S303, SIFT feature is extracted.
The present embodiment, can be specifically using following operation by using for SIFT algorithm: using Gaussian convolution check described in
Determine image to be handled, obtains multiscale space image；Difference of Gaussian processing is carried out to the multiscale space image, building is high
This difference scale space image；The Local Extremum for detecting the Gaussian difference scale space image, it is three-dimensional secondary using fitting
The Local Extremum is accurate to sub-pixel by function, using threshold method and Hessian matrix method to the Local Extremum
It is screened to obtain feature point set.
In step S305, given object is inputted.
In step S307, SIFT feature is extracted.This step extracts characteristic point using SIFT algorithm can be with step S303
Identical, therefore not to repeat here.
It should be noted that in actual operation, may wait for just holding after step S303 is performed both by with step S307
Row step S309.That is, after needing to wait for step S307 execution, then carrying out step S309 after step S303 is executed.Together
Sample after step S307 is executed, after needing to wait for step S303 execution, then carries out step S309.
In addition, it should be noted that, step S301- step S303, and step S305- step S307, between two groups of steps,
Its sequencing is not limited, can also be executed parallel.
In step S309, the characteristic point of image to be matched and given object is matched.
In step S311, the match point in image to be matched is obtained.
In step S313, classify to match point, forms multiple match point subsets.
In step S315, the multiple match point subset is traversed.
In step S317, to the characteristic point of each match point subset, matching ratio is calculated.
In step S319, the range of target object corresponding with aforementioned match point is calculated.
Calculate target object range method may include it is a variety of, the disclosure to this with no restriction.
It is calculated for example, algorithm as shown in Figure 4 can be used.Fig. 4 is calculating shown according to an exemplary embodiment
The flow chart of target object range.As shown in figure 4, including: according to the method for the calculating target object range of the present embodiment
In step S410, two match points A and B are arbitrarily chosen from match point subset Q, calculate the match point A and B
The distance between be used as first distance；
In the step s 420, match point C and D corresponding with the match point A and B in the given object is obtained respectively,
The distance between described match point C and D is calculated as second distance；
In step S430, the ratio calculated between the first distance and the second distance matches ratio as first
Example.
Matching ratio calculating can be used under type such as and calculate, such as:
P=Euclidean (a (x, y), b (x, y))/Euclidean (c (x, y), d (x, y))
A (x, y) be image to be matched in from match point subset Q any one match point；
B (x, y) is any one match point removed from match point subset Q other than a (x, y) in image to be matched；
C (x, y) be in given object with the matched Corresponding matching point of a (x, y)；
D (x, y) be in given object with the matched Corresponding matching point of b (x, y)；
Euclidean (a (x, y), b (x, y)) is the Euclidean distance of a (x, y) and b (x, y)；
Euclidean (c (x, y), d (x, y)) is the Euclidean distance of c (x, y) and d (x, y)；
P is matching ratio.
In step S440, according in first matching ratio, the match point A and the given object with described
The size of coordinate value and the given object with the corresponding match point C of point A, obtain in the image to be matched with it is described
The range for the target object that given object matches.
Specifically, following method can be used in this step:
RectA=(xa-xc*p,ya-yc*p,w*p,h*p)；
The wherein corresponding rectangular extent of RectA, four elements are successively in RectA are as follows: the lower left corner abscissa of rectangular extent,
The lower left corner ordinate of rectangular extent, the width of rectangular extent and the height of rectangular extent；
xaFor coordinate of the match point A in x-axis；
xcFor coordinate of the match point C in x-axis；
yaFor the coordinate of the match point A on the y axis；
ycFor the coordinate of the match point C on the y axis；
P is the matching ratio；
W is the width pixel value of the given object；
H is the height pixel value of the given object.
In this way, the target to match in the image to be matched with the given object can be obtained according to the match point subset
The range of object.It can also be executed in this way similarly, for other match point subsets, i.e., it can also be to each match point subset
As above operation is executed, to obtain the range of the target object to match in the image to be matched with the given object respectively.
The noise in match point can be effectively removed according to the technical solution of the disclosure.In addition, the density pair based on match point
Different location match point is classified in image, can be positioned out position difference but be shown the model of the identical icon of content
It encloses.
Fig. 5 is the block diagram of the device of multiple same objects in matching image shown according to an exemplary embodiment.Such as Fig. 5
Shown, the device of multiple same objects includes feature point set extraction unit 510, match point in matching image described in the present embodiment
Collect acquiring unit 520, taxon 530.
Feature point set extraction unit 510 is used to extract the feature point set of given object as fisrt feature point set, extraction to
The feature point set of matching image is as second feature point set.
Match the characteristic point that point set acquiring unit 520 is used to concentrate the fisrt feature point set and the second feature point
It is matched, obtains the matching point set in the image to be matched.
Taxon 530 forms multiple match point subsets for classifying to the matching point set.
According to an embodiment of the disclosure, the device of multiple same objects may also include range acquiring unit in matching image
540.Range acquiring unit 540 according to each match point subset and described is given respectively for traversing the multiple match point subset
The size for determining object obtains the range of the target object to match in the image to be matched with the given object.
According to an embodiment, range acquiring unit 540, which is configured that from match point subset Q, arbitrarily chooses two match point A
And B, the distance between described match point A and B is calculated as first distance；Obtain respectively in the given object with the matching
The corresponding match point C and D of point A and B calculates the distance between described match point C and D as second distance；Calculate described first
Ratio between distance and the second distance is as the first matching ratio；According to first matching ratio, the match point A
And the ruler of the coordinate value of the match point C corresponding with the match point A and the given object in the given object
It is very little, obtain the range of the target object to match in the image to be matched with the given object.
According to an embodiment, the range of target object can be determined by following formula:
RectA=(xa-xc*p,ya-yc*p,w*p,h*p)；
The wherein corresponding rectangular extent of RectA, four elements are successively in RectA are as follows: the lower left corner abscissa of rectangular extent,
The lower left corner ordinate of rectangular extent, the width of rectangular extent and the height of rectangular extent；
xaFor coordinate of the match point A in x-axis；
xcFor coordinate of the match point C in x-axis；
yaFor the coordinate of the match point A on the y axis；
ycFor the coordinate of the match point C on the y axis；
P is the matching ratio；
W is the width pixel value of the given object；
H is the height pixel value of the given object.
As previously mentioned, the given object can be for arbitrarily with the pictorial element, such as icon, button etc. of clear profile.
The image to be matched is to contain the image of multiple given objects, can be graphic user interface etc..
According to example embodiment, taxon 530 is configured to close according to being distributed to the matching point set using clustering algorithm
Degree is classified, and forms multiple match point subsets, the clustering algorithm includes DBSCAN algorithm, OPTICS algorithm or DENCLUE
Algorithm.
According to example embodiment, the feature point set extraction unit 510 is configured to execute SIFT algorithm.
According to example embodiment, the feature point set extraction unit 510, which is configured that, checks described give using Gaussian convolution
Image is handled, and multiscale space image is obtained；Difference of Gaussian processing is carried out to the multiscale space image, constructs Gauss
Difference scale space image；The Local Extremum for detecting the Gaussian difference scale space image utilizes the three-dimensional secondary letter of fitting
The Local Extremum is accurate to sub-pixel by number, is clicked through using threshold method and Hessian matrix method to the local extremum
Row screening obtains feature point set.
About the device in above-described embodiment, wherein each unit executes the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
Each method embodiment institute of the present invention can be performed in the device of multiple same objects in matching image provided in this embodiment
The method of multiple same objects in the matching image of offer has the corresponding functional module of execution method and beneficial effect.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to of the invention its
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the invention, these modifications, purposes or
Person's adaptive change follows general principle of the invention and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and true scope and spirit of the invention are by following
Claim is pointed out.
It should be understood that the present invention is not limited to the precise structure already described above and shown in the accompanying drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present invention is limited only by the attached claims.
Claims (12)
1. a kind of method of multiple same objects in matching image characterized by comprising
The feature point set of given object is extracted as fisrt feature point set, extracts the feature point set of image to be matched as the second spy
Levy point set；Wherein, the image to be matched is to contain the image of multiple given objects；
The fisrt feature point set is matched with the characteristic point that the second feature point is concentrated, obtains the image to be matched
On matching point set；
Classified to the matching point set according to distribution density using clustering algorithm, forms multiple match point subsets；
Two match points A and B are arbitrarily chosen from match point subset Q, calculate the distance between described match point A and B as the
One distance；
Match point C and D corresponding with the match point A and B in the given object is obtained respectively, calculates the match point C and D
The distance between be used as second distance；
The ratio between the first distance and the second distance is calculated as the first matching ratio；
According to described corresponding with the match point A in first matching ratio, the match point A and the given object
The size of coordinate value and the given object with point C, obtains the model of the target object to match with the given object
It encloses.
2. the method according to claim 1, wherein the range of the target object is determined by following formula:
RectA=(xa-xc*p,ya-yc*p,w*p,h*p)；
The wherein corresponding rectangular extent of RectA, four elements are successively in RectA are as follows: lower left corner abscissa, the rectangle of rectangular extent
The lower left corner ordinate of range, the width of rectangular extent and the height of rectangular extent；
xaFor coordinate of the match point A in x-axis；
xcFor coordinate of the match point C in x-axis；
Ya is the coordinate of the match point A on the y axis；
ycFor the coordinate of the match point C on the y axis；
P is the matching ratio；
W is the width pixel value of the given object；
H is the height pixel value of the given object.
3. the method according to claim 1, wherein the given object includes icon and/or button.
4. the method according to claim 1, wherein the clustering algorithm includes DBSCAN algorithm, OPTICS calculation
Method or DENCLUE algorithm.
5. the method according to claim 1, wherein being made using the feature point set that SIFT algorithm extracts given object
It is fisrt feature point set and the feature point set for extracting image to be matched as second feature point set.
6. according to the method described in claim 5, it is characterized in that, the SIFT algorithm includes:
The given image is checked using Gaussian convolution to be handled, and multiscale space image is obtained；
Difference of Gaussian processing is carried out to the multiscale space image, constructs Gaussian difference scale space image；
The Local Extremum for detecting the Gaussian difference scale space image, using being fitted three-dimensional quadratic function for the local pole
Value point is accurate to sub-pixel, is screened to obtain feature to the Local Extremum using threshold method and Hessian matrix method
Point set.
7. the device of multiple same objects in a kind of matching image characterized by comprising
Feature point set extraction unit, the feature point set for extracting given object extract figure to be matched as fisrt feature point set
The feature point set of picture is as second feature point set；Wherein, the image to be matched is to contain the figure of multiple given objects
Picture；
Point set acquiring unit is matched, the characteristic point progress for concentrating the fisrt feature point set and the second feature point
Match, obtains the matching point set in the image to be matched；
Taxon forms multiple matchings for classifying to the matching point set according to distribution density using clustering algorithm
Point subset；
Range acquiring unit calculates the match point A and B for arbitrarily choosing two match points A and B from match point subset Q
The distance between be used as first distance；
Match point C and D corresponding with the match point A and B in the given object is obtained respectively, calculates the match point C and D
The distance between be used as second distance；
The ratio between the first distance and the second distance is calculated as the first matching ratio；
According to described corresponding with the match point A in first matching ratio, the match point A and the given object
The size of coordinate value and the given object with point C, obtains the model of the target object to match with the given object
It encloses.
8. device according to claim 7, which is characterized in that the range of the target object is determined by following formula:
RectA=(xa-xc*p,ya-yc*p,w*p,h*p)；
The wherein corresponding rectangular extent of RectA, four elements are successively in RectA are as follows: lower left corner abscissa, the rectangle of rectangular extent
The lower left corner ordinate of range, the width of rectangular extent and the height of rectangular extent；
xaFor coordinate of the match point A in x-axis；
xcFor coordinate of the match point C in x-axis；
yaFor the coordinate of the match point A on the y axis；
ycFor the coordinate of the match point C on the y axis；
P is the matching ratio；
W is the width pixel value of the given object；
H is the height pixel value of the given object.
9. device according to claim 7, which is characterized in that the given object includes icon and/or button.
10. device according to claim 7, which is characterized in that the clustering algorithm includes DBSCAN algorithm, OPTICS calculation
Method or DENCLUE algorithm.
11. device according to claim 7, which is characterized in that the feature point set extraction unit is configured to execute SIFT
Algorithm.
12. device according to claim 11, which is characterized in that the feature point set extraction unit is configured that
The given image is checked using Gaussian convolution to be handled, and multiscale space image is obtained；
Difference of Gaussian processing is carried out to the multiscale space image, constructs Gaussian difference scale space image；
The Local Extremum for detecting the Gaussian difference scale space image, using being fitted three-dimensional quadratic function for the local pole
Value point is accurate to sub-pixel, is screened to obtain feature to the Local Extremum using threshold method and Hessian matrix method
Point set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610153710.7A CN105844290B (en) | 2016-03-16 | 2016-03-16 | The method and device of multiple same objects in matching image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610153710.7A CN105844290B (en) | 2016-03-16 | 2016-03-16 | The method and device of multiple same objects in matching image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105844290A CN105844290A (en) | 2016-08-10 |
| CN105844290B true CN105844290B (en) | 2019-10-25 |
Family
ID=56587272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610153710.7A Active CN105844290B (en) | 2016-03-16 | 2016-03-16 | The method and device of multiple same objects in matching image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105844290B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778781B (en) * | 2016-12-14 | 2018-08-17 | 维沃移动通信有限公司 | A kind of image processing method and device |
| CN108268881A (en) * | 2016-12-30 | 2018-07-10 | 中兴通讯股份有限公司 | Electronic equipment, target image identification method and device |
| CN109543536B (en) * | 2018-10-23 | 2020-11-10 | 北京市商汤科技开发有限公司 | Image identification method and device, electronic equipment and storage medium |
| CN113743424B (en) * | 2021-09-15 | 2023-06-27 | 昆明理工大学 | Quantitative characterization method for dynamic flocculation sedimentation dispersion degree of flocs in dynamic flocculation sedimentation of full tailings |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
| CN102982528A (en) * | 2008-09-25 | 2013-03-20 | 佳能株式会社 | Image processing apparatus and image processing method |
| CN104766084A (en) * | 2015-04-10 | 2015-07-08 | 南京大学 | Nearly copied image detection method based on multi-target matching |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8295651B2 (en) * | 2008-09-23 | 2012-10-23 | Microsoft Corporation | Coherent phrase model for efficient image near-duplicate retrieval |
-
2016
- 2016-03-16 CN CN201610153710.7A patent/CN105844290B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102982528A (en) * | 2008-09-25 | 2013-03-20 | 佳能株式会社 | Image processing apparatus and image processing method |
| CN102855473A (en) * | 2012-08-21 | 2013-01-02 | 中国科学院信息工程研究所 | Image multi-target detecting method based on similarity measurement |
| CN104766084A (en) * | 2015-04-10 | 2015-07-08 | 南京大学 | Nearly copied image detection method based on multi-target matching |
Non-Patent Citations (2)
| Title |
|---|
| 《一种基于点匹配的图像重复区域检测方法》;甘玲等;《微电子学与计算机》;20150831;第32卷(第8期);第62-66页 * |
| 一种基于特征点集的图像匹配算法;杨小影等;《计算机与网络》;20091231(第17期);第63-68页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105844290A (en) | 2016-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105894047B (en) | A kind of face classification system based on three-dimensional data | |
| Gelfand et al. | Robust global registration | |
| US8830229B2 (en) | Recognition and pose determination of 3D objects in 3D scenes | |
| CN105844290B (en) | The method and device of multiple same objects in matching image | |
| US20140105506A1 (en) | Recognition and pose determination of 3d objects in multimodal scenes | |
| Yuan et al. | Learning to count buildings in diverse aerial scenes | |
| CN105718552A (en) | Clothing freehand sketch based clothing image retrieval method | |
| CN109035207B (en) | Density self-adaptive laser point cloud characteristic detection method | |
| CN105654122B (en) | Based on the matched spatial pyramid object identification method of kernel function | |
| CN110147841A (en) | The fine grit classification method for being detected and being divided based on Weakly supervised and unsupervised component | |
| Alidoost et al. | Knowledge based 3D building model recognition using convolutional neural networks from LiDAR and aerial imageries | |
| CN108182705A (en) | A kind of three-dimensional coordinate localization method based on machine vision | |
| CN109463003A (en) | Object identifying | |
| CN114358133B (en) | Method for detecting looped frames based on semantic-assisted binocular vision SLAM | |
| CN106203446B (en) | Three dimensional object recognition positioning method for augmented reality auxiliary maintaining system | |
| CN115588178B (en) | Automatic extraction method for high-precision map elements | |
| Liu et al. | An improved local descriptor based object recognition in cluttered 3D point clouds | |
| Krueger | Model based object classification and localisation in multiocular images | |
| Díaz et al. | Cluster-wise ratio tests for fast camera localization | |
| Bughin et al. | Planar patch detection for disparity maps | |
| Decrouez et al. | Extracting planar structures efficiently with revisited BetaSAC | |
| Rangel et al. | Object recognition in noisy rgb-d data | |
| Liu et al. | A novel algorithm for planar extracting of 3D point clouds | |
| Wang et al. | Enhanced histogram feature descriptor for automated point cloud registration | |
| Karuna et al. | An efficient representation of shape for object recognition and classification using circular shift method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |