CN103903249B - Image matching system and method - Google Patents
Image matching system and method Download PDFInfo
- Publication number
- CN103903249B CN103903249B CN201210581470.2A CN201210581470A CN103903249B CN 103903249 B CN103903249 B CN 103903249B CN 201210581470 A CN201210581470 A CN 201210581470A CN 103903249 B CN103903249 B CN 103903249B
- Authority
- CN
- China
- Prior art keywords
- image
- triangle
- same place
- target
- area
- 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
- Image Analysis (AREA)
Abstract
The invention discloses image matching system and method, method includes step：S1, respectively from the first image with extracting N number of characteristic point in the second image and matching as same place；S2, respectively in the first image and the second image build at least one triangle using same place；S3, for every a pair of triangles by corresponding same place structure, by the larger triangle resampling of area to after identical with the area of the less triangle of area, statistics grey level histogram；S4, formula is run to every a pair of triangles S5, every a pair of triangles to calculating similarity average to calculate the similarity of first image and second image.The comparison for the similarity that the present invention passes through commodity image, to analyze whether two commodity belong to same type, can accurately represent result with numerical value, with the similarity degree of two commodity images of numerical quantization so that result of the comparison is more directly perceived credible.
Description
Technical field
The present invention relates to a kind of image matching system and method, more particularly to one kind can be by extracting characteristic point, structure
Build triangle and calculate the image matching system for carry out the similarity of commodity image quantitative analysis calculating and utilize and be somebody's turn to do
The image matching method that image matching system is realized.
Background technology
With the gradually popularization of the Internet, applications, occur in that increasing on-line shop now, and in on-line shop commodity price
The problem of undoubtedly most being paid close attention into user and shop business.And the rate of exchange are the primary hands for contrasting opponent's commodity with itself commodity price difference
Section, the key problem of the rate of exchange is the matching of commodity, i.e., found in the commodity of opponent belong to same type with itself commodity, it is same
The commodity of money, then carry out the contrast of price again.
And the matching of commodity typically can be by the detailed letters of the commodity such as trade name, marque, Brand at present
Cease to match commodity, but these merchandise newss can not be got under many circumstances, or the commodity letter obtained
Breath is not very comprehensive, so, is difficult to carry out goods matching again in the prior art.
The content of the invention
The technical problem to be solved in the present invention is can not to obtain merchandise news or acquisition in the prior art to overcome
Merchandise news it is incomplete in the case of be difficult carry out again the defect of goods matching there is provided one kind can by extract characteristic point,
Build triangle and calculate image matching system and the utilization for carry out the similarity of commodity image quantitative analysis calculating
The image matching method that the image matching system is realized.
The present invention is to solve above-mentioned technical problem by following technical proposals：
The invention provides a kind of image matching method, its feature is that it comprises the following steps：
S1, N number of characteristic point is extracted using feature point extraction algorithm from one first image and one second image respectively and will
N number of Feature Points Matching in N number of characteristic point and second image in first image is one-to-one N to same place,
Wherein N is positive integer；
S2, respectively in first image and second image build at least one triangle, first image with this
Every a pair of triangles in two images are built by corresponding same place, wherein each same place is at least one triangle
Same place is not included without public domain between summit, and triangle, and each in the region of triangle covering；
S3, for first image and every a pair of triangles for being built respectively by corresponding same place in second image,
Keep the sample rate of the larger triangle of area and by the larger triangle resampling of the area to area less three
After angular area is identical, the grey level histogram of area two triangles of identical is counted；
S4, to step S2In every a pair of triangles for constructing run formulaIts
Middle S is step S2In the similarity of a pair of triangles that is built by corresponding same place, M is by the gray scale in grey level histogram
The hop count that the interval of value is marked off, LiFor by i-th section of value after the larger triangle resampling of the area in gray value
The number of pixel, R in intervaliFor of the less triangle of area pixel in i-th section of interval of gray value
Number；
S5, to step S4In the similarity of every a pair of triangles that calculates average with calculate first image with
The similarity of second image.
Wherein, in step S1Middle utilization feature point extraction algorithm is extracted after characteristic point, respectively will be every in first image
One characteristic point is compared with all characteristic points in second image, and is found out in second image with maximum
The point of measuring similarity value as same place, by this method just can will first image with it is all in second image
Characteristic point all match into one-to-one same place, and its concrete implementation principle already belongs to techniques known,
Just repeat no more herein.And the characteristic point in image is typically all the point acquired a special sense on image, characteristic point is usually
Larger point of bright spot, the curvature of marginal point, dark areas on image etc., and the spy that different feature point extraction algorithms is extracted
It is a little different for levying.
And in step S2In can be specifically respectively in first image with choosing in correspondence with each other of the same name in second image
Point, and triangle in correspondence with each other is built using same place in correspondence with each other, and there is no public area between each two triangle
Domain ensures that the region covered between triangle is not repeated so that the calculating of follow-up similarity is more accurate.
Step S3In keep resolution ratio on the premise of by the larger triangle resampling of area to area less three
Angular area is identical, and this falls within techniques known, just repeats no more herein.
Step S4In counted grey level histogram and image pixel is normalized to [0,255] it is interval in after, can be with
Merge gray value according to actual conditions, gray value is divided into M sections, therefore M value can be determined voluntarily as needed, and it is public
I in formula is then positive integer and 1≤i≤M.
Pass through the image matching method, it becomes possible to incomplete in the merchandise news that can not obtain the details of commodity or acquisition
, still can be only by the comparison of the similarity of the images of commodity, to analyze whether two commodity belong to same in the case of face
Type or with a.And result of the comparison can accurately be showed with numerical value so that analysis, result of the comparison are more
It is directly perceived credible.
It is preferred that each triangle is respectively provided with a minimum outsourcing rectangle.
It is preferred that step S1In feature point extraction algorithm for SIFT algorithms, Harris algorithms, Moravec algorithms or
Robert algorithms（The SIFT algorithms, the Harris algorithms, the Moravec algorithms and the Robert algorithms are that characteristic point is carried
Take the specific species of algorithm）.
Present invention also offers a kind of image matching system, its feature is that it includes：
One feature point extraction module, for utilizing feature point extraction algorithm respectively from one first image and one second image
Extract N number of characteristic point and by N number of Feature Points Matching in N number of characteristic point in first image and second image for one by one
Corresponding N is to same place, and wherein N is positive integer；
One triangular models block, for building at least one triangle in first image and second image respectively
Shape, first image is built with every a pair of triangles in second image by corresponding same place, wherein each same place
It is not have public domain between at least one vertex of a triangle, and triangle, and the region of each triangle covering
In do not include same place；
One computing module, for firstly for first image with being built respectively by corresponding same place in second image
Every a pair of triangles, keeping the sample rate of the larger triangle of area and arriving the larger triangle resampling of the area
After identical with the area of the less triangle of area, the grey level histogram of area two triangles of identical is counted；
Formula is run to every a pair of triangles that triangular modeling block is constructed
Wherein S is the similarity of a pair of the triangles built by corresponding same place, and M is taking the gray value in grey level histogram
The hop count that value interval division goes out, LiFor by after the larger triangle resampling of the area in i-th section of interval of gray value
The number of pixel, RiFor the number of the less triangle of area pixel in i-th section of interval of gray value；
Finally the similarity of every a pair of triangles to calculating average with calculate first image with this second
The similarity of image.
It is preferred that each triangle is respectively provided with a minimum outsourcing rectangle.
It is preferred that this feature point extraction algorithm is SIFT algorithms, Harris algorithms, Moravec algorithms or Robert algorithms.
It is an object of the invention to additionally provide a kind of image matching method, its feature is that it comprises the following steps：
S1, N number of characteristic point is extracted using feature point extraction algorithm from one first image and one second image respectively and will
N number of Feature Points Matching in N number of characteristic point and second image in first image is one-to-one N to same place,
Wherein N is positive integer；
S2, a starting triangle, this pair starting triangle are built in first image and second image respectively
Summit is respectively the first image not triangle same place corresponding with second image and this pair starting triangle
Same place is not included in the region of shape covering；
S3, judge in first image and second image to whether there is in not triangle same place respectively with
The distance of at least two triangle same places is less than the target same place of a first threshold；
If so, then judging all compositions respectively in first image and second image for each target same place
With the presence or absence of at least two basic same places in the same place of triangle so that at least two basic same place and the target are same
The distance of famous cake is respectively less than the first threshold and in first image and second image at least two basic same place
Two bases of ratio and this of line segment in correspondence with each other for being constituted respectively with the target same place of two basic same places it is of the same name
The difference of ratio for the line segment in correspondence with each other that point is constituted is respectively less than a Second Threshold, if so, then first image and this
Target same place is utilized respectively in two images and builds target triangles with this two basic same places, first image and this
This two basic same places are corresponding in two images, without public area between the target triangle and remaining triangle
Same place is not included in domain, and the region of target triangle covering, if it is not, then rejecting target same place；
If it is not, then return to step S2；
S4, for first image and every a pair of triangles for being built respectively by corresponding same place in second image,
Keep the sample rate of the larger triangle of area and by the larger triangle resampling of the area to area less three
After angular area is identical, the grey level histogram of area two triangles of identical is counted；
S5, formula is run to every a pair of the triangles constructed by corresponding same place
Wherein S is step S3In the similarity of a pair of triangles that is built by corresponding same place, M is by the ash in grey level histogram
The hop count that the interval of angle value is marked off, LiWill to be taken after the larger triangle resampling of the area at i-th section of gray value
The number of pixel, R in value is intervaliFor the less triangle of the area in i-th section of interval of gray value pixel
Number；
S6, to step S5In the similarity of every a pair of triangles that calculates average with calculate first image with
The similarity of second image.
It can be filtered out in above-mentioned method using the principle of triangle of the same name come the same place to error hiding, of the same name three
Angular is that the same place of three matchings corresponding with second image using first image is constituted as summit.Work as triangle
After fixation, the relative coordinate and orientation on three summits therein be all to determine it is constant, and if two triangles are same
Name triangle, then it is corresponding and match to illustrate two vertexs of a triangle, that is, one of triangle can be with
By another triangle is obtained by the deformation of the non-qualitative changes such as rotation, scaling or movement, therefore three tops of a triangle
Relative coordinate, relative bearing etc. of point and relative coordinate, the relative bearing on three corresponding summits of another triangle etc.
It is proportion relation.It therefore, it can show that two triangles of the same name are similar triangles.
During actual images match, when being same place by Feature Points Matching, always in the presence of some error hidings
Point, these points can carry out error to follow-up calculating analytic band, cause the result for calculating analysis deviation occur.Therefore, it is only maximum
The point of the error hiding filtered out in same place of limit, just can ensure that the accuracy and reliability of the calculating of the similarity of image.
And due in first image and second image, if there is summit to be the same of error hiding in corresponding triangle
Famous cake, may result in two corresponding triangles dissmilarities, then the ratio difference on the corresponding side of triangle also can be very big, also
The ratio on the side being made up of the same places of two matchings in triangle with by a same place matched and an error hiding
The ratio difference on the side that same place is constituted can be very big.Therefore, according to the relative theory of triangle of the same name, it becomes possible to reject error hiding
Same place.
Wherein, with above-mentioned step S2Based on the starting triangle of middle structure, above-mentioned step S is recycled3It can just pick
Except the same place of error hiding, pass through step S3In first time judge, it becomes possible to ensure first all target same places with extremely
The distance of few two triangle same places is less than the first threshold, so, and new three are being built using target same place
When angular, the region for allowing for triangle covering will not be very big, so as to be to ensure that the accurate of the calculating of the similarity of image
Property and reliability.
And if being unsatisfactory for the requirement judged for the first time, just illustrate not triangle same place with it is triangle
The distance between same place is far, if now recycling not triangle same place and triangle same place structure
Build new triangle, will because of covering region it is excessive and have influence on the accuracy of the calculating of the similarity of follow-up image and
Reliability.Therefore, now will return to step S2, new starting triangle is built using not triangle same place,
Certainly, it also must assure that no public domain between new starting triangle and original starting triangle.
Then, for each target same place, step S can all be passed through3In second judgement, if target same place energy
Enough distances with least two basic same places are less than the first threshold, and target same place and two basic same places are distinguished
The ratio of the line segment in correspondence with each other constituted in first image and second image and two basic same places are respectively at this
The ratio difference of line segment in correspondence with each other of first image with being constituted in second image is less than the Second Threshold, then illustrates the mesh
Mark same place meets the requirements, and just builds new triangle with this two basic same places using the target same place.Certainly, this two
Individual basic same place can be chosen by user from least two basic same place according to actual needs.
And for the target same place for the requirement for being unsatisfactory for above-mentioned second of judgement, then explanation is the same place of error hiding,
Therefore it must be rejected, to ensure the accuracy of the Similarity Measure to the first image and the second image.So, will by mistake
The same place matched somebody with somebody carries out step S again after rejecting4-S6Calculating, it becomes possible to be more precisely calculated first image and this second
The similarity of image.
It is preferred that step S2The step of middle this pair of structure starting triangle, also includes：
S21, respectively in first image and second image build multiple triangles, first image and second figure
Every a pair of triangles as in are built by corresponding same place, do not have public domain between each two triangle wherein and each
Same place is not included in the region of triangle covering；
S22, calculate first image and every a pair are built by corresponding same place in second image triangle respectively
The average deviation of the ratio on corresponding side, chooses a pair of minimum triangles of average deviation as starting triangle.
In step S2In, multiple triangles are built first, then therefrom choose a pair of minimum triangles of deviation by calculating
It is used as starting triangle, it becomes possible to further ensure the accuracy of result of calculation.
It is preferred that each triangle is respectively provided with a minimum outsourcing rectangle.
It is preferred that step S1In feature point extraction algorithm for SIFT algorithms, Harris algorithms, Moravec algorithms or
Robert algorithms.
Present invention also offers a kind of image matching system, its feature is that it includes：
One feature point extraction module, for utilizing feature point extraction algorithm respectively from one first image and one second image
Extract N number of characteristic point and by N number of Feature Points Matching in N number of characteristic point in first image and second image for one by one
Corresponding N is to same place, and wherein N is positive integer；
One triangular models block, for building a starting triangle in first image and second image respectively
Shape, this pair starting vertex of a triangle is respectively not triangle same corresponding with second image of first image
Same place is not included in the region of famous cake and this pair starting triangle covering；
One judge module, for judging not triangle same place respectively in first image and second image
In whether there is the target same place for being less than a first threshold with the distances of at least two triangle same places；
If so, then judging all compositions respectively in first image and second image for each target same place
With the presence or absence of at least two basic same places in the same place of triangle so that at least two basic same place and the target are same
The distance of famous cake is respectively less than the first threshold and in first image and second image at least two basic same place
Two bases of ratio and this of line segment in correspondence with each other for being constituted respectively with the target same place of two basic same places it is of the same name
The difference of the ratio for the line segment in correspondence with each other that point is constituted is respectively less than a Second Threshold, if so, then calling the triangular to model
Block is utilized respectively target same place in first image and second image and builds target triangle with this two basic same places
Shape, first image and in second image this two basic same places be it is corresponding, the target triangle and remaining
Without same place is not included in public domain, and the region of target triangle covering between triangle, if it is not, then rejecting
Target same place；
If it is not, then call the triangular to model block builds a starting in first image and second image respectively
Triangle, this pair starting vertex of a triangle is respectively not triangle corresponding with second image of first image
Same place and this pair starting triangle covering region in do not include same place；
One computing module, for firstly for first image with being built respectively by corresponding same place in second image
Every a pair of triangles, keeping the sample rate of the larger triangle of area and arriving the larger triangle resampling of the area
After identical with the area of the less triangle of area, the grey level histogram of area two triangles of identical is counted；
Formula is run to every a pair of triangles that triangular modeling block is constructed
Wherein S is the similarity of a pair of the triangles built by corresponding same place, and M is taking the gray value in grey level histogram
The hop count that value interval division goes out, LiFor by after the larger triangle resampling of the area in i-th section of interval of gray value
The number of pixel, RiFor the number of the less triangle of area pixel in i-th section of interval of gray value；
Finally the similarity of every a pair of triangles to calculating average with calculate first image with this second
The similarity of image.
It is preferred that characterized in that, triangular modeling block is used for first respectively in first image and second figure
Multiple triangles are built as in, first image is with every a pair of triangles in second image by corresponding same place structure
Build, do not include same place in the region for not having public domain and the covering of each triangle between each two triangle wherein；
Then first image is calculated respectively and every a pair are built by corresponding same place in second image triangle
Corresponding side ratio average deviation, choose a pair of minimum triangles of average deviation and be used as starting triangle.
It is preferred that each triangle is respectively provided with a minimum outsourcing rectangle.
It is preferred that this feature point extraction algorithm is SIFT algorithms, Harris algorithms, Moravec algorithms or Robert algorithms.
The positive effect of the present invention is：The present invention can be in the details that can not obtain commodity or the business of acquisition
In the case of product information is incomplete, by the comparison of the similarity of the image of commodity, to analyze whether two commodity belong to same
Type or with a.And result of the comparison can accurately be showed with numerical value, with two commodity images of numerical quantization
Similarity degree so that analysis, result of the comparison are more directly perceived credible.
Brief description of the drawings
Fig. 1 is the structure chart of the goods matching system of embodiments of the invention 1.
Fig. 2 is the flow chart of the goods matching method of embodiments of the invention 1.
Fig. 3 is the structure chart of the goods matching system of embodiments of the invention 2.
Fig. 4 is the flow chart of the goods matching method of embodiments of the invention 2.
Embodiment
Present pre-ferred embodiments are provided below in conjunction with the accompanying drawings, to describe technical scheme in detail.
Embodiment 1：
Modeled as shown in figure 1, the goods matching system of the present embodiment includes a feature point extraction module 1, a triangular
The computing module 3 of block 2 and one.
The goods matching system only can just be calculated by the image of two commodity and judge the similarity of image, and then just
It may determine that whether two commodity belong to same type or with a.This feature point extraction module 1 is then first with characteristic point
Extraction algorithm extracts N number of characteristic point and by N number of feature in first image from one first image and one second image respectively
N number of Feature Points Matching in point and second image is one-to-one N to same place, and wherein N is positive integer.
And feature point extraction algorithm therein can be calculated for SIFT algorithms, Harris algorithms, Moravec algorithms or Robert
Method, these belong to algorithm well known in the art, just repeated no more herein.
Using feature point extraction algorithm extract characteristic point after, respectively by each characteristic point in first image with
All characteristic points in second image are compared, and are found out in second image with maximum similarity metric
Point by this method just can all match first image with all characteristic points in second image as same place
Into one-to-one same place, and its concrete implementation principle already belongs to techniques known, just repeats no more herein.
And the characteristic point in image is typically all the point acquired a special sense on image, characteristic point be usually marginal point on image,
Larger point of bright spot, the curvature of dark areas etc., and the characteristic point that different feature point extraction algorithms is extracted is different.
Then triangular modeling block 2 just builds at least one triangle in first image and second image respectively
Shape, first image is built with every a pair of triangles in second image by corresponding same place, wherein each same place
It is not have public domain between at least one vertex of a triangle, and triangle, and the region of each triangle covering
In do not include same place.
Same place in correspondence with each other can be specifically chosen in first image and second image respectively, and using mutually
Corresponding same place builds triangle in correspondence with each other, and when the quantity of triangle is more than two, it is follow-up in order to ensure
The accuracy of calculating, must be requested that all without public domain between any two triangle.In order that the result that must be calculated is more
Optimization, the outsourcing rectangular area of the triangle that can be constructed with optimal selection minimum same place builds triangle.
And the computing module 3 then can firstly for first image with second image respectively by corresponding same place structure
Every a pair of the triangles built, are keeping the sample rate of the larger triangle of area and by the larger triangle resampling of the area
To after identical with the area of the less triangle of area, the grey level histogram of area two triangles of identical is counted；
Formula is run to every a pair of triangles that triangular modeling block is constructed
Wherein S is the similarity of a pair of the triangles built by corresponding same place, and M is taking the gray value in grey level histogram
The hop count that value interval division goes out, LiFor by after the larger triangle resampling of the area in i-th section of interval of gray value
The number of pixel, RiFor the number of the less triangle of area pixel in i-th section of interval of gray value；
Finally the similarity of every a pair of triangles to calculating average with calculate first image with this second
The similarity of image.
Wherein keep resolution ratio on the premise of by the larger triangle resampling of area to the less triangle of area
Area it is identical, this falls within techniques known, just repeats no more herein.
, can be according to reality after having counted grey level histogram and having normalized to image pixel in [0,255] interval
Situation merges gray value, and gray value is divided into M sections, therefore M value can be determined voluntarily as needed, and the i in formula
It is then positive integer and 1≤i≤M.
For example, following form gives three pairs by the life in the first image and the second image respectively of same place in correspondence with each other
Into triangle pair A and a, B and b and C and c grey level histogram distribution situation, taken wherein having marked off five gray values
Value is interval, therefore the M values of above-mentioned formula are 5, and can just calculate every a pair of triangles respectively using above-mentioned formula
Similarity, result of calculation is as follows：
S (a, A)=1/5 × ((1- | 10-15 |/15)+(1- | 30-26 |/30)+(1
-|25-20|/25)+(1-|8-4|/8)+(1-|0-5|/5))
=1/5 × (0.667+0.867+0.8+0.5+0)=0.5668
S (b, B)=1/5 × ((1- | 10-15 |/15)+(1- | 5-4 |/5)+(1- | 20
-18|/20)+(1-|30-25|/30)+(1-|8-10|/10))
=1/5 × (0.667+0.8+0.9+0.834+0.8)=0.802
S (c, C)=1/5 × ((1- | 30-25 |/30)+(1- | 20-28 |/28)+(1
-|8-10|/10)+(1-|0-5|/5)+(1-|10-6|/10))
=1/5 × (0.834+0.714+0.8+0+0.6)=0.5896
And then to S（A, A）、S（B, B）And S（C, C）First image and second image can just be calculated by averaging
Similarity, i.e.,
Utilize above-mentioned calculation procedure, it becomes possible to calculate the similarity of two images.
As shown in Fig. 2 the goods matching method that the present invention is realized using the goods matching system of the present embodiment is including following
Step：
Step 100, using feature point extraction algorithm extract N number of characteristic point from one first image and one second image respectively
And by N number of Feature Points Matching in N number of characteristic point in first image and second image be one-to-one N to of the same name
Point.
Step 101, in first image and second image build at least one triangle respectively, first image with
Every a pair of triangles in second image are built by corresponding same place, wherein each same place is at least one triangle
The summit of shape, and do not include same place without public domain between triangle, and each in the region of triangle covering.
Step 102, for first image and built respectively by corresponding same place in second image every a pair three
It is angular, keeping the sample rate of the larger triangle of area and by the larger triangle resampling of the area to smaller with area
Triangle area it is identical after, count area two triangles of identical grey level histogram.
Step 103, formula is run to every a pair of the triangles constructed in step 101
Step 104, the similarity of every a pair of triangles to being calculated in step 103 average with calculate this first
The similarity of image and second image, then terminates flow.
Embodiment 2：
As shown in figure 3, the image matching system of the present embodiment is except including a feature point extraction module 1, a triangular
Model outside the computing module 3 of block 2 and one, in addition to a judge module 4.
The present embodiment and the difference of embodiment 1 are：In the present embodiment, triangular modeling block 2 can be respectively at this
First image is with building a starting triangle in second image, this pair starting vertex of a triangle is respectively first image
Do not wrapped in the region of the not triangle same place corresponding with second image and this pair starting triangle covering
Containing same place.
In specific build, triangular modeling block 2 is built in first image and second image respectively first
Multiple triangles, first image is built with every a pair of triangles in second image by corresponding same place, wherein often
Same place is not included in the region for not having public domain and the covering of each triangle between two triangles.
Then first image is calculated respectively and every a pair are built by corresponding same place in second image triangle
Corresponding side ratio average deviation, choose a pair of minimum triangles of average deviation and be used as starting triangle.
Also, the judge module 4 is in first image with judging not triangle of the same name in second image respectively
The distance that whether there is in point with least two triangle same places is less than the target same place of a first threshold；
If so, then judging all compositions respectively in first image and second image for each target same place
With the presence or absence of at least two basic same places in the same place of triangle so that at least two basic same place and the target are same
The distance of famous cake is respectively less than the first threshold and in first image and second image at least two basic same place
Two bases of ratio and this of line segment in correspondence with each other for being constituted respectively with the target same place of two basic same places it is of the same name
The difference of the ratio for the line segment in correspondence with each other that point is constituted is respectively less than a Second Threshold, if so, then calling the triangular to model
Block 2 is utilized respectively target same place in first image and second image and builds target triangle with this two basic same places
Shape, first image and in second image this two basic same places be it is corresponding, the target triangle and remaining
Without same place is not included in public domain, and the region of target triangle covering between triangle, if it is not, then rejecting
Target same place；
If it is not, then call the triangular to model block 2 builds one in first image and second image respectively
Beginning triangle, this pair starting vertex of a triangle be respectively first image it is corresponding with second image do not constitute triangle
Same place is not included in the region of the same place of shape and this pair starting triangle covering.
During actual images match, when being same place by Feature Points Matching, always in the presence of some error hidings
Point, these points can carry out error to follow-up calculating analytic band, cause the result for calculating analysis deviation occur.Therefore, it is only maximum
The point of the error hiding filtered out in same place of limit, just can ensure that the accuracy and reliability of the calculating of the similarity of image.
Wherein, based on the triangular models the starting triangle that block 2 is built, the judge module 4 is recycled with regard to energy
The same place of error hiding is enough rejected, is judged by the first time of the judge module 4, it becomes possible to ensure that all targets are of the same name first
The distance of point and at least two triangle same places is less than the first threshold, so, is built using target same place
During new triangle, the region for allowing for triangle covering will not be very big, so as to be to ensure that the calculating of the similarity of image
Accuracy and reliability.
And if being unsatisfactory for the requirement judged for the first time, just illustrate not triangle same place with it is triangle
The distance between same place is far, if now recycling not triangle same place and triangle same place structure
Build new triangle, will because of covering region it is excessive and have influence on the accuracy of the calculating of the similarity of follow-up image and
Reliability.Therefore, now just call the triangular to model block 2 and rebuild starting triangle, using not triangle
Same place builds new starting triangle, certainly, must also be protected between new starting triangle and original starting triangle
Demonstrate,prove no public domain.
Then, for each target same place, it can all be judged by second of the judge module 4, if target is of the same name
Point can be less than the first threshold, and target same place and two basic same places with the distance of at least two basic same places
The ratio of the line segment in correspondence with each other constituted respectively in first image and second image and two basic same places are distinguished
The ratio difference of the line segment in correspondence with each other constituted in first image and second image is less than the Second Threshold, then illustrates
The target same place meets the requirements, and just builds new triangle with this two basic same places using the target same place.Certainly,
This two basic same places can be chosen by user from least two basic same place according to actual needs.
And for the target same place for the requirement for being unsatisfactory for above-mentioned second of judgement, then explanation is the same place of error hiding,
Therefore it must be rejected, to ensure the accuracy of the Similarity Measure to the first image and the second image.So, will by mistake
The same place matched somebody with somebody is calculated by the computing module 3 again after rejecting, it becomes possible to be more precisely calculated first image with
The similarity of second image.
As shown in figure 4, the image matching method that the present invention is realized using the image matching system of the present embodiment is including following
Step：
Step 200, using feature point extraction algorithm extract N number of characteristic point from one first image and one second image respectively
And by N number of Feature Points Matching in N number of characteristic point in first image and second image be one-to-one N to of the same name
Point, wherein N are positive integer.
Step 201, respectively one starting triangle of structure in first image and second image, this pair starting triangle
The summit of shape is respectively the first image not triangle same place corresponding with second image and this pair is originated
Same place is not included in the region of triangle covering.
Step 202, judge respectively in first image and second image in not triangle same place whether
In the presence of the target same place for being less than a first threshold with the distance of at least two triangle same places, if so, then performing
Step 203, if it is not, then return to step 201.
Step 203, for each target same place, judge all compositions respectively in first image and second image
With the presence or absence of at least two basic same places in the same place of triangle so that at least two basic same place and the target are same
The distance of famous cake is respectively less than the first threshold and in first image and second image at least two basic same place
Two bases of ratio and this of line segment in correspondence with each other for being constituted respectively with the target same place of two basic same places it is of the same name
The difference of ratio for the line segment in correspondence with each other that point is constituted is respectively less than a Second Threshold, if so, then first image and this
Target same place is utilized respectively in two images and builds target triangles with this two basic same places, first image and this
This two basic same places are corresponding in two images, without public area between the target triangle and remaining triangle
Same place is not included in domain, and the region of target triangle covering, if it is not, then rejecting target same place.
Step 204, for first image and built respectively by corresponding same place in second image every a pair three
It is angular, keeping the sample rate of the larger triangle of area and by the larger triangle resampling of the area to smaller with area
Triangle area it is identical after, count area two triangles of identical grey level histogram.
Step 205, formula is run to every a pair of the triangles constructed by corresponding same place
Step 206, the similarity of every a pair of triangles to being calculated in step 205 average with calculate this first
The similarity of image and second image, then terminates flow.
Although the foregoing describing the embodiment of the present invention, it will be appreciated by those of skill in the art that these
It is merely illustrative of, protection scope of the present invention is defined by the appended claims.Those skilled in the art is not carrying on the back
On the premise of principle and essence from the present invention, various changes or modifications can be made to these embodiments, but these are changed
Protection scope of the present invention is each fallen within modification.
Claims (14)
1. a kind of image matching method, it is characterised in that it comprises the following steps：
S1, extracted respectively from one first image and one second image using feature point extraction algorithm N number of characteristic point and by this first
N number of Feature Points Matching in N number of characteristic point and second image in image is one-to-one N to same place, and wherein N is
Positive integer；
S2, respectively in first image and second image build at least one triangle, first image and second image
In every a pair of triangles built by corresponding same place, wherein each same place is at least one vertex of a triangle,
And do not include same place without public domain between triangle, and each in the region of triangle covering；
S3, for first image and every a pair of triangles for being built respectively by corresponding same place in second image, protecting
On the premise of the sample rate for holding the larger triangle of area, by the larger triangle resampling of the area to area less three
After angular area is identical, the grey level histogram of area two triangles of identical is counted；
S4, to step S2In every a pair of triangles for constructing run formulaWherein S is
Step S2In the similarity of a pair of triangles that is built by corresponding same place, M is by the gray value in grey level histogram
The hop count that interval is marked off, LiFor by i-th section of interval after the larger triangle resampling of the area in gray value
The number of interior pixel, RiFor the number of the less triangle of area pixel in i-th section of interval of gray value；
S5, to step S4In the similarity of every a pair of triangles that calculates average with calculate first image with this
The similarity of two images.
2. image matching method as claimed in claim 1, it is characterised in that each triangle is respectively provided with a minimum outsourcing square
Shape.
3. image matching method as claimed in claim 1 or 2, it is characterised in that step S1In feature point extraction algorithm be
SIFT algorithms, Harris algorithms, Moravec algorithms or Robert algorithms.
4. a kind of image matching system, it is characterised in that it includes：
One feature point extraction module, for being extracted respectively from one first image and one second image using feature point extraction algorithm
N number of characteristic point and by N number of Feature Points Matching in N number of characteristic point in first image and second image for correspond
N to same place, wherein N is positive integer；
One triangular models block, should for building at least one triangle in first image and second image respectively
First image is built with every a pair of triangles in second image by corresponding same place, wherein each same place be to
Few one no public domain between vertex of a triangle, and triangle, and in the region that each triangle is covered not
Include same place；
One computing module, for firstly for first image with second image respectively by corresponding same place build it is every
A pair of triangles, on the premise of the sample rate of the larger triangle of area is kept, the larger triangle of the area is adopted again
Sample counts the grey level histogram of area two triangles of identical to after identical with the area of the less triangle of area；
Formula is run to every a pair of triangles that triangular modeling block is constructed
Wherein S is the similarity that triangular models a pair of the triangles built in block by corresponding same place, and M is by intensity histogram
The hop count that the interval of gray value in figure is marked off, LiFor by after the larger triangle resampling of the area in gray value
The number of pixel, R in i-th section of intervaliFor the less triangle of the area in i-th section of interval of gray value picture
The number of vegetarian refreshments；
Finally the similarity of every a pair of triangles to being calculated in computing module average with calculate first image with
The similarity of second image.
5. image matching system as claimed in claim 4, it is characterised in that each triangle is respectively provided with a minimum outsourcing square
Shape.
6. the image matching system as described in claim 4 or 5, it is characterised in that this feature point extraction algorithm be SIFT algorithms,
Harris algorithms, Moravec algorithms or Robert algorithms.
7. a kind of image matching method, it is characterised in that it comprises the following steps：
S1, extracted respectively from one first image and one second image using feature point extraction algorithm N number of characteristic point and by this first
N number of Feature Points Matching in N number of characteristic point and second image in image is one-to-one N to same place, and wherein N is
Positive integer；
S2, a starting triangle, this pair starting vertex of a triangle point are built in first image and second image respectively
Wei not the first image not triangle same place corresponding with second image and this pair starting triangle covering
Region in do not include same place；
S3, judge in first image and second image to whether there is in not triangle same place respectively with least
The distance of two triangle same places is less than the target same place of a first threshold；
If so, then judging all composition triangles respectively in first image and second image for each target same place
With the presence or absence of at least two basic same places in the same place of shape so that at least two basic same place and the target same place
Distance be respectively less than the first threshold and in first image and second image at least two basic same place two
The ratio for the line segment in correspondence with each other that individual basic same place is constituted with the target same place respectively and this two basic same place structures
Into the difference of ratio of line segment in correspondence with each other be respectively less than a Second Threshold, if so, then in first image and second figure
Target same place is utilized respectively as in and builds target triangle with this two basic same places, in first image and second figure
This two basic same places are corresponding as in, without public domain between the target triangle and remaining triangle,
And do not include same place in the region of target triangle covering, if it is not, then rejecting target same place；
If it is not, then return to step S2；
S4, for first image and every twin target triangle for being built respectively by corresponding same place in second image,
On the premise of the sample rate of the larger target triangle of area is kept, by the larger target triangle resampling of the area to
After the area of the less target triangle of area is identical, the grey level histogram of area two target triangles of identical is counted；
S5, formula is run to the every twin target triangle constructed by corresponding same place
Wherein S is step S3In the similarity of twin target triangle that is built by corresponding same place, M is by grey level histogram
Gray value the hop count that marks off of interval, LiFor by after the larger target triangle resampling of the area in gray value
The number of pixel, R in i-th section of intervaliFor the less target triangle of the area gray value i-th section of interval
The number of interior pixel；
S6, to step S5In the similarity of every twin target triangle that calculates average with calculate first image with
The similarity of second image.
8. image matching method as claimed in claim 7, it is characterised in that step S2It is middle to build the step that this pair originates triangle
Suddenly also include：
S21, respectively in first image and second image build multiple triangles, first image with second image
Every a pair of triangles built by corresponding same place, there is no public domain and each triangle between each two triangle wherein
Same place is not included in the region of shape covering；
S22, to calculate first image relative with every a pair are built by corresponding same place in second image triangle respectively
The average deviation of the ratio on the side answered, chooses a pair of minimum triangles of average deviation as starting triangle.
9. image matching method as claimed in claim 8, it is characterised in that each triangle is respectively provided with a minimum outsourcing square
Shape.
10. the image matching method as described in any one in claim 7-9, it is characterised in that step S1In characteristic point carry
It is SIFT algorithms, Harris algorithms, Moravec algorithms or Robert algorithms to take algorithm.
11. a kind of image matching system, it is characterised in that it includes：
One feature point extraction module, for being extracted respectively from one first image and one second image using feature point extraction algorithm
N number of characteristic point and by N number of Feature Points Matching in N number of characteristic point in first image and second image for correspond
N to same place, wherein N is positive integer；
One triangular models block, should for building a starting triangle in first image and second image respectively
It is respectively the first image not triangle same place corresponding with second image to starting vertex of a triangle
And do not include same place in the region of this pair starting triangle covering；
One judge module, for first image with judge respectively in second image be in not triangle same place
No target same place of the distance less than a first threshold existed with least two triangle same places；
If so, then judging all composition triangles respectively in first image and second image for each target same place
With the presence or absence of at least two basic same places in the same place of shape so that at least two basic same place and the target same place
Distance be respectively less than the first threshold and in first image and second image at least two basic same place two
The ratio for the line segment in correspondence with each other that individual basic same place is constituted with the target same place respectively and this two basic same place structures
Into the difference of ratio of line segment in correspondence with each other be respectively less than a Second Threshold, exist if so, then calling the triangular to model block
First image builds target triangles with being utilized respectively target same place and this two basic same places in second image,
First image is corresponding, the target triangle and remaining triangle with this two in second image basic same places
Between without public domain, and do not include same place in the region of target triangle covering, if it is not, it is same then to reject target
Famous cake；
If it is not, then call the triangular to model block builds a starting triangle in first image and second image respectively
Shape, this pair starting vertex of a triangle is respectively not triangle same corresponding with second image of first image
Same place is not included in the region of famous cake and this pair starting triangle covering；
One computing module, for firstly for first image with second image respectively by corresponding same place build it is every
Twin target triangle, on the premise of the sample rate of the larger target triangle of area is kept, by the mesh that the area is larger
Triangle resampling is marked to after identical with the area of the less target triangle of area, statistics area two target triangles of identical
The grey level histogram of shape；
Formula is run to every twin target triangle that target triangular modeling block is constructed
Wherein S is the similarity of the twin target triangle built in judge module by corresponding same place, and M is by grey level histogram
In gray value the hop count that marks off of interval, LiFor by after the larger target triangle resampling of the area in gray value
I-th section of interval in pixel number, RiIt is the less target triangle of the area in i-th section of value area of gray value
The number of interior pixel；
Finally the similarity of every twin target triangle to being calculated in computing module averages to calculate first figure
As the similarity with second image.
12. image matching system as claimed in claim 11, it is characterised in that triangular modeling block is used to distinguish first
Multiple triangles are built in first image and second image, first image and every a pair of triangles in second image
Shape is built by corresponding same place, the region for not having public domain and the covering of each triangle between each two triangle wherein
In do not include same place；
Then the phase of first image and the triangle that every a pair are built by corresponding same place in second image is calculated respectively
The average deviation of the ratio on corresponding side, chooses a pair of minimum triangles of average deviation as starting triangle.
13. image matching system as claimed in claim 12, it is characterised in that each triangle is respectively provided with a minimum outsourcing square
Shape.
14. the image matching system as described in any one in claim 11-13, it is characterised in that this feature point, which is extracted, to be calculated
Method is SIFT algorithms, Harris algorithms, Moravec algorithms or Robert algorithms.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210581470.2A CN103903249B (en) | 2012-12-27 | 2012-12-27 | Image matching system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210581470.2A CN103903249B (en) | 2012-12-27 | 2012-12-27 | Image matching system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103903249A CN103903249A (en) | 2014-07-02 |
| CN103903249B true CN103903249B (en) | 2017-10-13 |
Family
ID=50994556
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210581470.2A Active CN103903249B (en) | 2012-12-27 | 2012-12-27 | Image matching system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103903249B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104766323B (en) * | 2015-04-07 | 2018-03-06 | 北京航空航天大学 | A kind of Point matching method of remote sensing images |
| CN106484118B (en) * | 2016-10-24 | 2020-01-14 | 福建北极光虚拟视觉展示科技有限公司 | Augmented reality interaction method and system based on fixed trigger source |
| CN107084712B (en) * | 2017-04-21 | 2019-11-29 | 北京京东尚科信息技术有限公司 | Data processing method and device and method for calibrating compass and device |
| CN107256509B (en) * | 2017-05-27 | 2021-05-14 | 北京小米移动软件有限公司 | Price comparison method and device, terminal, server and storage medium |
| CN108921020A (en) * | 2018-05-29 | 2018-11-30 | 北大青鸟环宇消防设备股份有限公司 | A kind of fire image-recognizing method and its device |
| CN109190655B (en) * | 2018-07-12 | 2021-12-14 | 成都信息工程大学 | NCC image matching method based on enzyme numerical value membrane system |
| CN109447154B (en) * | 2018-10-29 | 2021-06-04 | 网易（杭州）网络有限公司 | Picture similarity detection method, device, medium and electronic equipment |
| CA3154216A1 (en) | 2019-10-11 | 2021-04-15 | Beyeonics Surgical Ltd. | System and method for improved electronic assisted medical procedures |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002069249A1 (en) * | 2001-02-21 | 2002-09-06 | Campbell Shannon R | A method for matching spatial patterns |
| CN1426021A (en) * | 2002-12-19 | 2003-06-25 | 上海交通大学 | Non-linear registration method for remote sensing image |
| JP2004334337A (en) * | 2003-04-30 | 2004-11-25 | Canon Inc | Image processor |
| CN1912889A (en) * | 2005-08-11 | 2007-02-14 | 中国科学院自动化研究所 | Deformed fingerprint identification method based on local triangle structure characteristic collection |
| CN101320432A (en) * | 2008-07-04 | 2008-12-10 | 浙江大学 | Image matching propagation method for error control based on triangle restriction |
| CN101556695A (en) * | 2009-05-15 | 2009-10-14 | 广东工业大学 | Image matching method |
| CN102332137A (en) * | 2011-09-23 | 2012-01-25 | 纽海信息技术（上海）有限公司 | Goods matching method and system |
| CN102693542A (en) * | 2012-05-18 | 2012-09-26 | 中国人民解放军信息工程大学 | Image characteristic matching method |
| JP2012203823A (en) * | 2011-03-28 | 2012-10-22 | Kddi Corp | Image recognition device |
-
2012
- 2012-12-27 CN CN201210581470.2A patent/CN103903249B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002069249A1 (en) * | 2001-02-21 | 2002-09-06 | Campbell Shannon R | A method for matching spatial patterns |
| CN1426021A (en) * | 2002-12-19 | 2003-06-25 | 上海交通大学 | Non-linear registration method for remote sensing image |
| JP2004334337A (en) * | 2003-04-30 | 2004-11-25 | Canon Inc | Image processor |
| CN1912889A (en) * | 2005-08-11 | 2007-02-14 | 中国科学院自动化研究所 | Deformed fingerprint identification method based on local triangle structure characteristic collection |
| CN101320432A (en) * | 2008-07-04 | 2008-12-10 | 浙江大学 | Image matching propagation method for error control based on triangle restriction |
| CN101556695A (en) * | 2009-05-15 | 2009-10-14 | 广东工业大学 | Image matching method |
| JP2012203823A (en) * | 2011-03-28 | 2012-10-22 | Kddi Corp | Image recognition device |
| CN102332137A (en) * | 2011-09-23 | 2012-01-25 | 纽海信息技术（上海）有限公司 | Goods matching method and system |
| CN102693542A (en) * | 2012-05-18 | 2012-09-26 | 中国人民解放军信息工程大学 | Image characteristic matching method |
Non-Patent Citations (4)
| Title |
|---|
| "A New Algorithm of Global Feature Matching Based on Triangle Regions for Image Registration";Zhaoxia Liu et al.;《ICSP2010 Proceedings》;20101231;1248-1251 * |
| "An Algorithm for Distorted Fingerprint Matching Based on Local Triangle Feature Set";Xinjian Chen et al.;《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》;20060630;第1卷(第2期);169-177 * |
| "三角形约束下的图像特征点匹配方法";吴飞 等;《计算机辅助设计与图形学学报》;20100331;第22卷(第3期);503-510 * |
| "基于特征点相似度的匹配定位算法";甄巍松等;《武汉工程大学学报》;20110430;第33卷(第4期);85-88 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103903249A (en) | 2014-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103903249B (en) | Image matching system and method | |
| Mesev | The use of census data in urban image classification | |
| WO2020143377A1 (en) | Industry recognition model determination method and apparatus | |
| CN109800698A (en) | Icon detection method based on depth network | |
| Im et al. | An automated binary change detection model using a calibration approach | |
| Myint | An exploration of spatial dispersion, pattern, and association of socio-economic functional units in an urban system | |
| CN103530590A (en) | DPM (direct part mark) two-dimensional code recognition system | |
| CN108875727B (en) | The detection method and device of graph-text identification, storage medium, processor | |
| CN109784290A (en) | A kind of object detection method, device, equipment and readable storage medium storing program for executing | |
| CN201673505U (en) | Artwork identification system based on texture analysis | |
| CN110909738A (en) | Automatic reading method of pointer instrument based on key point detection | |
| CN109117717A (en) | A kind of city pedestrian detection method | |
| CN110232592A (en) | County domain electric business developing state appraisal procedure and system based on web crawlers technology | |
| Liang et al. | Planning control over rural land transformation in Hong Kong: A remote sensing analysis of spatio-temporal land use change patterns | |
| CN109102324B (en) | Model training method, and red packet material laying prediction method and device based on model | |
| CN113869331A (en) | Water body extraction method, device and equipment based on remote sensing image and storage medium | |
| Zhang et al. | Novel shape indices for vector landscape pattern analysis | |
| CN110597990B (en) | Financial analysis method and system based on intelligent classification | |
| CN104331747B (en) | Malice escapes single detection method | |
| CN113469127B (en) | River sensibility grading method, device, equipment and storage medium | |
| CN114418970A (en) | Haze distribution and aerosol optical thickness detection method and device based on satellite remote sensing | |
| Zhang et al. | Spatio-temporal subpixel mapping with cloudy images | |
| CN104616265B (en) | A kind of time domain method for reconstructing of remote sensing sequence data | |
| CN115082546B (en) | Method and device for determining pollutant discharge amount, electronic equipment and medium | |
| CN108897781A (en) | Paper figure duplicate checking system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20160914 Address after: East Building 11, 100195 Beijing city Haidian District xingshikou Road No. 65 west Shan creative garden district 1-4 four layer of 1-4 layer Applicant after: Beijing Jingdong Shangke Information Technology Co., Ltd. Address before: 201203 Shanghai city Pudong New Area Zu Road No. 295 Room 102 Applicant before: Niuhai Information Technology (Shanghai) Co., Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201112 Address after: No.8-6, Putou South Road, Haicang District, Xiamen City, Fujian Province Patentee after: Xiamen xinjianfu e-commerce Co., Ltd Address before: East Building 11, 100195 Beijing city Haidian District xingshikou Road No. 65 west Shan creative garden district 1-4 four layer of 1-4 layer Patentee before: BEIJING JINGDONG SHANGKE INFORMATION TECHNOLOGY Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210517 Address after: 361000 No.8, Putou South Road, Haicang District, Xiamen City, Fujian Province Patentee after: Xiamen Jianfu Chain Management Co.,Ltd. Address before: No.8-6, Putou South Road, Haicang District, Xiamen City, Fujian Province 361022 Patentee before: Xiamen xinjianfu e-commerce Co., Ltd |