CN103400384B  The wideangle image matching process of calmodulin binding domain CaM coupling and some coupling  Google Patents
The wideangle image matching process of calmodulin binding domain CaM coupling and some coupling Download PDFInfo
 Publication number
 CN103400384B CN103400384B CN201310325400.5A CN201310325400A CN103400384B CN 103400384 B CN103400384 B CN 103400384B CN 201310325400 A CN201310325400 A CN 201310325400A CN 103400384 B CN103400384 B CN 103400384B
 Authority
 CN
 China
 Prior art keywords
 region
 mser
 point
 characteristic point
 represents
 Prior art date
Links
Abstract
Description
Technical field
The invention belongs to technical field of image processing, further relate to the land in a kind of image registration techniques fieldThe wideangle image matching process of territory coupling and some coupling. The present invention can extract more accurately from the image of inputGo out conforming characteristic point, can be applicable to exist in image registration with great visual angle.
Background technology
Images match is a special dimension of image processing, by images match, the different images of Same Scene itBetween extract conforming characteristic point, determine between image corresponding geometrical relationship, obtain the image after a width coupling.The single image Description Image scene more accurately of image ratio after coupling, generally, can adopt based on officeThe method of portion's feature extracting and matching is carried out images match, and these local features extract and the method for coupling has mainly been consideredThe yardstick of input picture and rotational invariance, but often ignored because image aspects changes the distortion causing, thereby causeCan not obtain effectively, accurately corresponding matching characteristic point problem.
(application number: CN200710176039.9, application is open for the patent " image matching method " of TsingHua University's applicationNumber: a kind of method that discloses images match CN101140624). The concrete steps of the method are: first, utilizeMSER method detects the homogeneous region in image to be matched, with elliptic region equation to the each homogeneous region detectingCarry out matching; Then, utilize oval histogram of gradients and multiscale image to decompose, obtain elliptic region equation modelAfter the characteristic vector of homogeneous region place measured zone; Finally, adopt layering k means clustering method structure homogeneity districtThe binary tree of territory set, carries out images match. The method is mainly used for images match by MSER method for detecting area,But the deficiency that the method still exists is, because image is often by different sensors and at different visual anglesUnder obtain, if simply utilize MSER region to mate, the region obtaining does not have Shandong for dimensional variationRod, can not solve by changing with great visual angle the image fault problem causing, this can reduce coupling to a certain extentMethod is obtained the performance of robustness consistency point in modifiedimage with great visual angle.
The paper " Distinctiveimagefeaturesfromscaleinvariantkeypoints. " that LoweD deliversIn (InternationalJournalofComputerVision, vol.60, no.2, pp.91110), disclosing a kind of feature carriesGet and matching process. First the method adopts metric space extreme value to detect the picture position on all yardsticks of search; ThenOn each candidate's position, determine position and yardstick by a meticulous model of matching, based on image localGradient direction, distributes to the one or more directions in each key point position; Finally, the neighborhood around each key pointIn, utilize the gradient of image local to carry out key point description, utilize arest neighbors than inferior neighbour's plan to the characteristic point of describingSlightly mate, obtain the characteristic point of coupling. The deficiency that the method exists is, because the method has just been considered yardstickAnd rotational invariance, do not consider affine consistency. Therefore in the situation that there is geometric distortion in image, two width figureDescription region as corresponding points can not cover identical content, and the method often can not get satisfied result.
Paper " the ASIFT:Anewframeworkforfullyaffine that J.M.MorelandG.Yu deliversInvariantimagecomparison " (SIAMJournalonImagingSciences2,438,2009). in openThe constant image matching method of a kind of affine yardstick. First the method passes through the directioin parameter of two camera axis of conversion,Namely longitude angle and angle of latitude, carries out suitable sampled analog at two parameter spaces, makes two images to be matchedForm two groups of views; Then, utilize SIFT method to mate these two groups of views; Finally, extract this two picture groupAll match points in picture, and they are mapped on two width original images. The deficiency that the method exists is, due to the partyMethod is simulated on affine space image, forms the image at each visual angle, has consumed a large amount of internal memories; Exist simultaneouslyWhile extracting a large amount of correct match points, also introduce a large amount of Mismatching points, caused correct matching rate to decline.
Summary of the invention
The object of the invention is to overcome the shortcoming of abovementioned prior art, propose a kind of calmodulin binding domain CaM coupling and some couplingWideangle image matching process, to better extract the uniformity feature that more robustness is stronger from input picturePoint, and there is higher correct matching rate.
For achieving the above object, when the present invention carries out feature detection to input picture, calmodulin binding domain CaM coupling and some coupling,The spatial information of having introduced characteristic point in a coupling carrys out Expressive Features point simultaneously, and its technical scheme is first to two width figurePicture carries out the detection of MSER region and coupling, and on the MSER region of coupling, carry out yardstick invariant features point and detect,Then characteristic point is carried out describing and spatial description based on the 128 dimension SIFT of MSER; Finally adopt Binding distanceSimilitude strategy obtains comparatively accurate matching double points.
Concrete steps of the present invention are as follows:
(1) input picture: input exists the two width images that change with great visual angle respectively;
(2) MSER detects:
Two width images are carried out respectively to maximum stable extremal region MSER detection, obtain multiple affine consistency that haveIrregular extremal region;
(3) MSER matching:
3a) to each irregular extremal region, the point in digital simulation region according to the following formula:
(xμ)^{T}U^{1}(xμ)＝(xμ)^{T}M(xμ)＝1
Wherein, x represents the point of fitted area, and μ represents the average of irregular extremal region, and T represents transposition, U tableShow the variance of irregular extremal region, M represents the secondorder moments matrix of irregular extremal region;
3b) obtained all fitted area points are formed to ellipse fitting region;
(4) MSER normalization:
4a) calculate according to the following formula the point in normalization region:
z＝sH^{1}M^{1/2}(xμ)
Wherein, z represents the point in normalization region, and s represents zoom factor, 6≤s≤10, and H represents secondorder moments matrixThe symmetrical unitary matrice of reality that M singular value decomposition obtains, M represents the secondorder moments matrix of irregular extremal region, x representsThe point of fitted area, μ represents the average of irregular extremal region;
4b) obtained all normalization region point are formed to normalization region;
(5) MSER describes:
Normalization region is described with SIFT Feature Descriptor, forms 128 dimension SIFT descriptors;
(6) MSER coupling:
Adopt arest neighbors than inferior neighbour's strategy, in two width images, select initial matching area pair;
(7) yardstick invariant features point detects:
Internal at initial matching area, utilize the SIFT method to detect yardstick constant characteristic point, determine characteristic pointPrecise position information and characteristic dimension information;
(8) characteristic point is described:
8a) obtain the 128 dimension SIFT descriptors based on MSER of characteristic point;
8b) obtain the 2 dimension space descriptors based on MSER of characteristic point;
(9) the similitude strategy of employing Binding distance is selected meticulous Feature Points Matching pair in two width images.
The present invention has the following advantages compared with prior art:
The first, the present invention carries out MSER detection and coupling owing to adopting to two width input pictures, in the region of couplingOn carry out yardstick consistency feature detection, construct affine constant oval neighboring region, on normalization region, carry outSIFT describe processing method, overcome in prior art, be limited to circular neighboring region do not have affine consistency lackFall into, make matching process that the present invention designs compared with prior art, there is the main orientation of more stable characteristic point andAccurate affine consistency descriptor.
The second, the present invention is owing to having adopted characteristic point distinctive spatial relationship in MSER region, and it is imitative that formation has 2 dimensionsPenetrate the method for the space descriptor of consistency, overcome in the matching process of prior art and only considered optical characteristics, and do not hadThere is the defect of considering spatial relationship, make the present invention compared with prior art there is Geng Gaoke district to the description of characteristic pointDivide property.
Brief description of the drawings
Fig. 1 is flow chart of the present invention;
Fig. 2 is the wideangle image coupling simulated effect figure of structural type scene;
Fig. 3 is the wideangle image coupling simulated effect figure of texture type scene;
Detailed description of the invention
Below in conjunction with accompanying drawing, the present invention will be further described.
With reference to accompanying drawing 1, concrete steps of the present invention are as follows:
Step 1, input picture: input exists the two width images that change with great visual angle respectively.
Step 2, MSER detects:
Two width images are carried out respectively to maximum stable extremal region MSER detection, obtain multiple affine consistency that haveIrregular extremal region.
Step 3, MSER matching:
To each irregular extremal region, the point in digital simulation region according to the following formula:
(xμ)^{T}U^{1}(xμ)＝(xμ)^{T}M(xμ)＝1
Wherein, x represents the point of fitted area, and μ represents the average of irregular extremal region, and T represents transposition, U tableShow the variance of irregular extremal region, M represents the secondorder moments matrix of irregular extremal region.
Obtained all fitted area points are formed to ellipse fitting region.
Step 4, MSER normalization:
Calculate according to the following formula the point in normalization region:
z＝sH^{1}M^{1/2}(xμ)
Wherein, z represents the point in normalization region, and s represents zoom factor, and 6≤s≤10 adopt s=10 in experiment,H represents the symmetrical unitary matrice of reality that secondorder moments matrix M singular value decomposition obtains, and M represents two of irregular extremal regionRank square matrix, x represents the point of fitted area, μ represents the average of irregular extremal region.
Obtained all normalization region point are formed to normalization region.
Step 5, MSER describes:
Normalization region is described with SIFT Feature Descriptor, forms 128 dimension SIFT descriptors.
Step 6, MSER coupling:
Adopt arest neighbors than inferior neighbour's strategy, in two width images, select initial matching area pair.
Arest neighbors refers in two width images than inferior neighbour's strategy, and nearest neighbor distance in descriptor is removed to the following nearest neighbor distanceThe matching area pair that the value obtaining is less than 0.85, as a pair of initial matching area pair.
Step 7, yardstick invariant features point detects:
Internal at initial matching area, utilize the SIFT method to detect yardstick constant characteristic point, determine characteristic pointPrecise position information and characteristic dimension information.
Step 8, characteristic point is described:
The 128 dimension SIFT descriptors based on MSER that obtain characteristic point, its concrete implementation step is as follows:
Choose oval neighboring region: in the Gaussian image at characteristic point yardstick place, choose one centered by characteristic point,Meet the oval neighboring region of the ellipticity of oval neighboring region and two conditions of the area of oval neighboring region simultaneously, itsIn, the ellipticity of oval neighboring region equals the ellipticity of the affiliated MSER fitted area of this characteristic point, oval adjacent areaThe area in territory equals 4.5^{2}*σ^{2}, the characteristic dimension that wherein σ is characteristic point.
Oval neighboring region normalization: oval neighboring region is normalized to the round neighboring region that a radius is 4.5 σ,The characteristic dimension that wherein σ is characteristic point.
Circle neighboring region describe: with SIFT Feature Descriptor to circle neighboring region be described, form based on MSER128 dimension SIFT descriptors.
The 2 dimension space descriptors based on MSER that obtain characteristic point, its concrete implementation step is as follows:
The MSER region at characteristic point place is normalized to unit circle region.
In this unit circle region, 2 dimension space descriptor l and θ of feature points, l representation feature point is from circleThe distance descriptor of the heart, the principal direction of θ representation feature point deducts the center of circle and points to the poor angle of the direction of this direction and describeSymbol.
Step 9, the similitude strategy of employing Binding distance is selected meticulous Feature Points Matching pair in two width images,Its concrete implementation step is as follows:
The first step, calculate according to the following formula the Binding distance of any two characteristic points in two width input picture corresponding regions:
D(p，q)＝α_{1}E_{1}(p，q)+α_{2}E_{2}(p，q)+(1α_{1}α_{2})E_{3}(p，q)
Wherein, D (p, q) represents characteristic point p in two width input picture corresponding regions, the Binding distance of q; α_{1}And α_{2}PointDo not represent two weighted factors, 0≤α_{1}≤1，0≤α_{2}≤1，0≤α_{1}+α_{2}In≤1 experiment, adopt α_{1}＝0.3， α_{2}＝0.6；E_{1}(p, q) represents characteristic point p in two width input picture corresponding regions, between 2 SIFT descriptors of qEuclidean distance; E_{2}(p, q) represents characteristic point p in two width input picture corresponding regions, and q is apart from the distance between descriptorFrom; E_{3}(p, q) represents characteristic point p in two width input picture corresponding regions, the distance between q angle descriptor.
Second step, select according to the following formula meticulous Feature Points Matching pair:
D(k，m)/D(k，n)＜T
Wherein, D (k, m) represents characteristic point k and obtained mating in MSER region in a MSER regionCharacteristic point m between Binding distance, D (k, n) represent characteristic point k in a MSER region and obtainJoin the Binding distance between the characteristic point n in MSER region, T represents threshold value, and the span of T is 0.7 to 0.85Between, in experiment, adopt T=0.85.
Below in conjunction with Fig. 2 and 3 analogous diagram, effect of the present invention is further described.
For verifying validity of the present invention and correctness, adopt the wideangle image of two groups of different scenes to mate imitativeTrue experiment, and with existing three kinds of classical matching process SIFT methods, MSER method, ASIFT method and Ben FaBright method is carried out emulation comparison. All emulation experiments all adopt Visual under WindowsXP operating systemStudio2010 software is realized.
Emulation 1
What Fig. 2 (a) showed is, and different visual angles 0 is spent, 20 degree, 30 degree, 40 degree, 50 degree, 60 degree is lower obtainsThe gray level image of six width structural type scenes, the image size adopting is 800 × 600, wherein establishes piece image for ginsengExamine image, other images are respectively image to be matched. Fig. 2 (b) is the correct matching number of structural type image, horizontal seatWhat mark represented is visual angle change, and what ordinate represented is with correct match point number, and Fig. 2 (c) is structural type imageCorrect matching rate, what abscissa represented is visual angle change, what ordinate represented is correct matching rate; From Fig. 2 (b) andIn 2 (c), can find out, the correct coupling that the inventive method obtains in all visual angle change situations is counted and is all wanted obviousGround is higher than SIFT method and MSER method, and the inventive method obtained the correct matching rate close to 100%,Especially in enough large situations of visual angle change, namely, in the situations of 60 degree, the inventive method has obtained 162 pairsCorrect matching double points, SIFT method and MSER method have obtained respectively 1 pair and 39 pairs of correct matching double points; ThisBright method and ASIFT method compare, and can find out, although side of the present invention from Fig. 2 (b) and Fig. 2 (c)Method on correct matching double points number lower than ASIFT method, but the correct matching rate of the inventive method apparently higher thanASIFT method, therefore the inventive method not only can obtain higher correct matching double points, and has just higherReally matching rate.
Emulation 2
What emulation 3 (a) showed is that different visual angles 0 is spent, 20 degree, 30 are spent, 40 degree, 50 are spent, 60 lower of degree obtainThe gray level image of six width texture type scenes, the image size adopting is 800 × 600. Wherein establishing piece image isReference picture, other images are respectively image to be matched. Fig. 3 (b) is the correct matching number of texture type image, horizontal strokeWhat coordinate represented is visual angle change, and what ordinate represented is with correct match point number, and Fig. 3 (c) is texture type imageCorrect matching rate, what abscissa represented is visual angle change, what ordinate represented is correct matching rate; From Fig. 3 (b)With in 3 (c), can find out, visual angle change before 40 degree, the inventive method obtain correct coupling number and SIFTThe correct coupling number that method obtains is suitable, and the correct coupling number obtaining than MSER method is obviously higher, along with visual angle becomesThe increase of changing, the correct coupling number that the inventive method obtains, apparently higher than SIFT method and MSER method, especially existsVisual angle change is 60 while spending, and the inventive method has obtained 25 pairs of correct matching double points, and MSER method has obtained 7 pairsCorrect matching double points, SIFT has obtained 4 pairs of true matching double points; The present invention and ASIFT method compare, from figureIn 3 (b) and Fig. 3 (c), can find out, although the inventive method on correct matching double points number lower than ASIFTMethod, but the correct matching rate of the inventive method apparently higher than ASIFT method, therefore the inventive method can not onlyAccess higher correct matching double points, and there is higher correct matching rate.
Claims (5)
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

CN201310325400.5A CN103400384B (en)  20130722  20130722  The wideangle image matching process of calmodulin binding domain CaM coupling and some coupling 
Applications Claiming Priority (1)
Application Number  Priority Date  Filing Date  Title 

CN201310325400.5A CN103400384B (en)  20130722  20130722  The wideangle image matching process of calmodulin binding domain CaM coupling and some coupling 
Publications (2)
Publication Number  Publication Date 

CN103400384A CN103400384A (en)  20131120 
CN103400384B true CN103400384B (en)  20160525 
Family
ID=49563996
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

CN201310325400.5A CN103400384B (en)  20130722  20130722  The wideangle image matching process of calmodulin binding domain CaM coupling and some coupling 
Country Status (1)
Country  Link 

CN (1)  CN103400384B (en) 
Families Citing this family (6)
Publication number  Priority date  Publication date  Assignee  Title 

CN103617616A (en) *  20131203  20140305  康江科技(北京)有限责任公司  Affine invariant image matching method 
CN103679726B (en) *  20131216  20160817  成都安健发科技有限公司  A kind of method improving rock debris image image quality 
CN104134209B (en) *  20140718  20180410  北京国电富通科技发展有限责任公司  A kind of feature extracting and matching method and system in vision guided navigation 
CN104616280B (en) *  20141126  20170728  西安电子科技大学  Method for registering images based on maximum stable extremal region and phase equalization 
CN105160686B (en) *  20151021  20170825  武汉大学  A kind of low latitude various visual angles Remote Sensing Images Matching Method based on improvement SIFT operators 
CN105741297B (en) *  20160202  20180824  南京航空航天大学  A kind of repeat pattern image matching method with affineinvariant features 
Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN101140624A (en) *  20071018  20080312  清华大学  Image matching method 
CN101667293A (en) *  20090924  20100310  哈尔滨工业大学  Method for conducting highprecision and steady registration on diversified sensor remote sensing images 
CN102270303A (en) *  20110727  20111207  重庆大学  Joint detection method sensitive image 
Family Cites Families (3)
Publication number  Priority date  Publication date  Assignee  Title 

JP3166905B2 (en) *  19970702  20010514  インターナショナル・ビジネス・マシーンズ・コーポレ−ション  Image processing method and system by pattern matching 
US9070002B2 (en) *  20111018  20150630  King Fahd University Of Petroleum And Minerals  Method of performing fingerprint matching 
CN103136292B (en) *  20111205  20160907  中国移动通信集团公司  A kind of image search method based on characteristics of image and device 

2013
 20130722 CN CN201310325400.5A patent/CN103400384B/en active IP Right Grant
Patent Citations (3)
Publication number  Priority date  Publication date  Assignee  Title 

CN101140624A (en) *  20071018  20080312  清华大学  Image matching method 
CN101667293A (en) *  20090924  20100310  哈尔滨工业大学  Method for conducting highprecision and steady registration on diversified sensor remote sensing images 
CN102270303A (en) *  20110727  20111207  重庆大学  Joint detection method sensitive image 
Also Published As
Publication number  Publication date 

CN103400384A (en)  20131120 
Similar Documents
Publication  Publication Date  Title 

Wang et al.  MSLD: A robust descriptor for line matching  
Zhang et al.  An efficient and robust line segment matching approach based on LBD descriptor and pairwise geometric consistency  
Siagian et al.  Rapid biologicallyinspired scene classification using features shared with visual attention  
Gopalakrishnan et al.  Salient region detection by modeling distributions of color and orientation  
CN101976455B (en)  Color image threedimensional reconstruction method based on threedimensional matching  
Novatnack et al.  Scaledependent/invariant local 3D shape descriptors for fully automatic registration of multiple sets of range images  
Boltes et al.  Automatic extraction of pedestrian trajectories from video recordings  
Zhang et al.  A robust, realtime ellipse detector  
KR101121034B1 (en)  System and method for obtaining camera parameters from multiple images and computer program products thereof  
CN101950426B (en)  Vehicle relay tracking method in multicamera scene  
CN101673403A (en)  Target following method in complex interference scene  
CN102859535B (en)  Daisy descriptor is produced from precalculated metric space  
CN101894366A (en)  Method and device for acquiring calibration parameters and video monitoring system  
CN103503025A (en)  Determining model parameters based on transforming a model of an object  
CN103106688A (en)  Indoor threedimensional scene rebuilding method based on doublelayer rectification method  
Cheng et al.  Determining vision graphs for distributed camera networks using feature digests  
Jiménez et al.  Traffic sign shape classification and localization based on the normalized FFT of the signature of blobs and 2D homographies  
CN102044151B (en)  Night vehicle video detection method based on illumination visibility identification  
Yu et al.  Motion pattern interpretation and detection for tracking moving vehicles in airborne video  
CN101271582A (en)  Threedimensional reconstruction method based on multivision angle twodimension image combined with SIFT algorithm  
CN103164858B (en)  Adhesion crowd based on superpixel and graph model is split and tracking  
CN101667293A (en)  Method for conducting highprecision and steady registration on diversified sensor remote sensing images  
Drost et al.  3d object detection and localization using multimodal point pair features  
Werghi et al.  The meshlbp: a framework for extracting local binary patterns from discrete manifolds  
US20180268237A1 (en)  Method and system for determining at least one property related to at least part of a real environment 
Legal Events
Date  Code  Title  Description 

PB01  Publication  
C06  Publication  
SE01  Entry into force of request for substantive examination  
C10  Entry into substantive examination  
GR01  Patent grant  
C14  Grant of patent or utility model 