CN107170001A - Method and apparatus for carrying out registration to image - Google Patents
Method and apparatus for carrying out registration to image Download PDFInfo
- Publication number
- CN107170001A CN107170001A CN201710278075.XA CN201710278075A CN107170001A CN 107170001 A CN107170001 A CN 107170001A CN 201710278075 A CN201710278075 A CN 201710278075A CN 107170001 A CN107170001 A CN 107170001A
- Authority
- CN
- China
- Prior art keywords
- registration
- image
- double points
- matching double
- point
- 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.)
- Withdrawn
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
It is an object of the invention to provide a kind of method and apparatus for being used to carry out image registration.The method according to the invention comprises the following steps：Multiple characteristic points pair in thick matching double points based on image subject to registration and standard picture, calculate the transformation model matrix for acting on image subject to registration, to obtain corresponding preliminary registration image；Moving distance information based on image subject to registration and the marginal point of preliminary registration image is screened to thick matching double points, so as to obtain accurate matching double points；Based on the accurate matching double points, repeat the operation of above-mentioned two step, new transformation model matrix is calculated to act on the preliminary registration image, to obtain new preliminary registration image, and the moving distance information of the marginal point based on new preliminary registration image and former preliminary registration image is screened to the accurate matching double points, to obtain more accurate matching double points.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of method and apparatus for being used to carry out image registration.
Background technology
Scale invariant feature conversion (Scale-invariant feature transform, SIFT) algorithm is a kind of normal
The algorithm for being used to carry out image registration seen, the calculation obtains feature by the characteristic point and its description tried to achieve in image and carried out
Image Feature Point Matching.
Method for registering images based on SIFT feature generally comprises feature generation phase
Wherein, feature generation phase main flow is：
Step one：Metric space is built, extreme point is detected
Characteristic point is obtained by the extreme point of difference of Gaussian (Difference of Gaussian, DOG) image.Each
Characteristic point includes three partial informations, i.e. position, residing yardstick and direction.Due to realizing that the unique linear core of change of scale is Gauss
Convolution kernel, the metric space of image can be expressed as：
D (x, y, δ)=(G (x, y, k δ)-G (x, y, δ)) * I (x, y)
=L (x, y, k δ)-L (x, y, δ) (1)
Wherein, L (x, y, δ)=G (x, y, δ) * I (x, y),
Wherein, D (x, y, δ) is Gaussian difference scale space, is generated by the Gaussian difference pyrene and image convolution of different scale,
I (x, y) represents original image, and G (x, y, δ) represents changeable scale Gaussian function, and L (x, y, δ) represents Gaussian convolution image.
Image pyramid is set up, extreme point is obtained by detecting Gaussian difference scale space.
Step 2：Characteristic point is filtered and is accurately positioned
It is the characteristic point stablized, it is necessary to filter out unstable low contrast features point and skirt response point.By step
The Taylor expansion in rapid mesoscale space can obtain：
Derivation is carried out to above formula, and makes its result be 0, characteristic point position is can obtain
With reference to above-mentioned two formula, obtain:
Pass throughSpan filter out the characteristic point of low contrast.
Then, made a return journey except edge response point, be expressed as using a Hessian matrix H：
Made a return journey by following constraints except edge response point：
Wherein γ is the ratio of larger characteristic value and smaller characteristic value, DxxRepresent a certain yardstick in difference of Gaussian pyramid
Image x direction derivations are twice.
Step 3：It is characterized a distribution direction value
The gradient magnitude at (x, y) place is expressed as：
The direction at (x, y) place is expressed as：
Wherein, L represents the yardstick residing for each characteristic point.It is crucial using statistics with histogram to determine the direction of characteristic point
The gradient direction of vertex neighborhood pixel, assign histogrammic peak value as the direction of the key point.
Step 4：Generate Feature Descriptor
Specifically, the 16*16 neighborhood territory pixels of each characteristic point are selected, 16 4*4 neighborhood window size are divided into, often
Individual window has the information in 8 directions, therefore description that can be tieed up to each characteristic point formation 4*4*8=128.
After the completion of generation phase, choose between image subject to registration and the Feature Descriptor of standard video individual features point
The minimum point of Euclidean distance elects match point as, is screened, obtained according to the Euclidean distance ratio relation of arest neighbors and time neighbour
First group of thick matching double points.
However, the characteristic matching accuracy of the image registration scheme based on SIFT algorithms still has much room for improvement, particularly exist
Remote sensing image is carried out with punctual, can be to of characteristic point because the geometric position of characteristics of remote sensing image point may be not accurate enough
Impacted with accuracy.
The content of the invention
It is an object of the invention to provide a kind of method and apparatus for being used to carry out image registration.
According to an aspect of the invention, there is provided a kind of method for being used to carry out image registration, wherein, methods described
The characteristic point of image subject to registration and standard picture is obtained using SIFT algorithms and thick matching double points, methods described are obtained by screening
Comprise the following steps：
Multiple characteristic points pair in thick matching double points of a based on image subject to registration and standard picture, calculating, which is acted on, to be waited to match somebody with somebody
The transformation model matrix of quasi- image, to obtain corresponding preliminary registration image；
Moving distance informations of the b based on image subject to registration and the marginal point of preliminary registration image is sieved to thick matching double points
Choosing, so as to obtain accurate matching double points；
Wherein, it the described method comprises the following steps：
C is based on the accurate matching double points, and repeat step a and b operation calculate new transformation model matrix to act on
The preliminary registration image, to obtain new preliminary registration image, and based on new preliminary registration image and former preliminary registration figure
The moving distance information of the marginal point of picture is screened to the accurate matching double points, to obtain more accurate matching double points.
According to an aspect of the present invention, a kind of registration apparatus for being used to carry out image registration is additionally provided, wherein, institute
Registration apparatus is stated to obtain the characteristic point of image subject to registration and standard picture using SIFT algorithms and obtain thick match point by screening
Right, the registration apparatus includes：
Computing device, for multiple characteristic points pair in the thick matching double points based on image subject to registration and standard picture, meter
It can be regarded as the transformation model matrix for image subject to registration, to obtain corresponding preliminary registration image；
Screening plant, for the moving distance information based on image subject to registration and the marginal point of preliminary registration image to thick
With point to screening, so as to obtain accurate matching double points；
Wherein, it the described method comprises the following steps：
Coalignment, for based on the accurate matching double points, repeat step a and b operation to calculate new transformation model
Matrix acts on the preliminary registration image, to obtain new preliminary registration image, and based on new preliminary registration image and
The moving distance information of the marginal point of former preliminary registration image is screened to the accurate matching double points, more accurate to obtain
Matching double points.
Compared with prior art, the present invention has advantages below：Passed through based on image subject to registration before and after registration several times
The distance between corresponding marginal point information is screened to the matching double points obtained by SIFT algorithms, so as to obtain more
Accurate matching double points, compared with the general scheme based on SIFT algorithms, the method according to the invention improves figure subject to registration
The accuracy of the characteristic matching of picture and standard picture, realizes matching somebody with somebody for not accurate enough the high score image in geometric position and standard video
Standard, so as to reach preferably registration effect on time to remote sensing image match somebody with somebody.
Brief description of the drawings
By reading the detailed description made to non-limiting example made with reference to the following drawings, of the invention is other
Feature, objects and advantages will become more apparent upon：
Fig. 1 illustrates a kind of method flow diagram for being used to carry out image registration according to the present invention；
Fig. 2 illustrates a kind of structural representation for being used to carry out image the registration apparatus of registration according to the present invention；
Fig. 3 a illustrate the schematic diagram of an exemplary translation transformation relation according to the present invention；
Fig. 3 b illustrate the schematic diagram of an exemplary rotation transformation relation according to the present invention；
Fig. 4 illustrates the schematic diagram of an exemplary satellite image according to the present invention.
Same or analogous reference represents same or analogous part in accompanying drawing.
Embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings.
Fig. 1 illustrates a kind of method flow diagram for being used to carry out image registration according to the present invention.According to the present invention
Method include step S1, step S2 and step S3.
Wherein, the method according to the invention is realized by the registration apparatus being contained in computer equipment.It is described to calculate
Machine equipment according to the instruction for being previously set or storing, can carry out the electricity of numerical computations and/or information processing automatically including a kind of
Sub- equipment, its hardware includes but is not limited to microprocessor, application specific integrated circuit (ASIC), programmable gate array (FPGA), numeral
Processor (DSP), embedded device etc..The computer equipment includes the network equipment and/or user equipment.Wherein, the net
Network equipment includes but is not limited to single network server, the server group of multiple webservers composition or based on cloud computing
The cloud being made up of a large amount of main frames or the webserver of (Cloud Computing), wherein, cloud computing is the one of Distributed Calculation
Kind, a super virtual computer being made up of the computer collection of a group loose couplings.The user equipment includes but is not limited to
Any one can carry out the electricity of man-machine interaction with user by modes such as keyboard, mouse, remote control, touch pad or voice-operated devices
Sub- product, for example, personal computer, tablet personal computer, smart mobile phone etc..Wherein, residing for the user equipment and the network equipment
Network includes but is not limited to internet, wide area network, Metropolitan Area Network (MAN), LAN, VPN etc..
It should be noted that the user equipment, the network equipment and network are only for example, other are existing or from now on may be used
Can occur user equipment, the network equipment and network be such as applicable to the present invention, also should be included in the scope of the present invention with
It is interior, and be incorporated herein by reference.
Preferably, it is used to carry out registration to remote sensing image according to the registration apparatus of the present invention.
Preferably, the registration apparatus obtains the characteristic point of image subject to registration and standard picture using SIFT algorithms and passed through
Screening obtains thick matching double points, before step S1, the described method comprises the following steps：Registration apparatus is extracted using SIFT algorithms
Multiple characteristic points and feature point description of image and standard picture subject to registration；Then, by being screened to each characteristic point
Obtain the thick matching double points of image subject to registration and standard picture.The process is illustrated in background parts, and here is omitted.
Preferably, registration apparatus is based on consistent (Random Sample Consensus, the RANSAC) algorithm pair of random sampling
Each characteristic point is screened, to obtain the thick matching double points of image subject to registration and standard picture.
Reference picture 1 is in step sl, more in thick matching double points of the registration apparatus based on image subject to registration and standard picture
Individual characteristic point pair, calculates the transformation model matrix for acting on image subject to registration, to obtain corresponding preliminary registration image.
Wherein, methods described represents the position of characteristic point using three-dimensional homogeneous coordinates, and the transformation model matrix is answered to be single
Property matrix.
Preferably, registration apparatus is based on RANSAC algorithms, is selected from the thick matching double points of image subject to registration and standard picture
Matching preferable characteristic point is selected to calculating transformation model matrix.
For example, being carried out to remote sensing image with punctual, for image I subject to registration1With standard video I2, M represents transformation model
Matrix, and M is homography matrix, is obtained：
Wherein, parameter m0, m1, m3, m4Play yardstick and rotation amount；Parameter m2Play horizontal direction displacement；m5
Play vertical direction displacement；m6, m7Play the deflection of level and vertical direction.
Above-mentioned matrix M is acted on into image I subject to registration1(x, y), then can obtain image I subject to registration1Preliminary registration image
I′1, I1With the point I in standard video2The transformation relation existed between (x ', y ') is expressed as：
When calculating obtains 8 transformation parameter m0, m1, m2, m3, m4, m5, m6, m7When, matrix M is to determine.In homogeneous coordinates,
If point P (x on original imagei,yi, 1) and it is changed into P ' (x ' by Metzler matrixi,y′i, 1) and it is p'=M*p.There is following relation in it：
Assuming that there are 8 frees degree for Metzler matrix, 4 pairs of characteristic points are so at least needed to solution.4 pairs of characteristic points set up 8
Equation.When there is n to characteristic point, overdetermined equation is constructed, approximate solution is obtained by least square fitting：
Solution p'=M*p equation groups can be converted into the solution to homogeneous equation group Ax=0, that is, be converted into min | | Ax | |2's
Nonlinear optimal problem.By asking for characteristic value to coefficient matrices A and characteristic vector is obtained, the minimal eigenvalue of coefficient matrices A
Corresponding characteristic vector is exactly over-determined systems Ax=0 least square solution.
Preferably, translation transformation relation and rotation transformation relation of the registration apparatus based on image subject to registration and standard picture,
Calculate the transformation model matrix for acting on image subject to registration.
Illustrate that the translation transformation relation between general image is shown with reference to the schematic diagram shown by Fig. 3 a and Fig. 3 b
Intention and rotation transformation relation.The point coordinates (x, y) of two-dimensional space is represented using three-dimensional homogeneous coordinates (x, y, 1), is united
One is a 3*3 matrix.In order to represent that affine projection is converted with matrix, three-dimensional (x, y, 1) is represented into two dimension using homogeneous coordinates
(x,y)。
1) translation transformation relation
Reference picture 3a, the translation transformation of image is generally left and right, translated up and down, so in translation transformation, waiting to match somebody with somebody
X '=x+t is met between quasi- imaging point (x, y) and the corresponding point of standard video (x ', y ')x, y '=y+ty.Solid line shown in Fig. 3 a
Image is expressed as by translation transformation to dashed line view picture, its translation transformation relation：
Wherein, txFor left and right translational movement (can be positive and negative), ty be upper and lower translation amount (can be positive and negative),For translation
Transformation matrix.
2) rotation transformation relation
Rotation transformation is that point all in artwork all is rotated into same angle to same direction around a fixed point, from
And artwork deformation is changed to another figure.In rotary course, rotation three elements are kept mutually to unify, thirdly element is 1. to revolve
Turn center；2. direction of rotation；3. the anglec of rotation.
Real diagram picture in Fig. 3 b obtains dashed line view picture after rotated counterclockwise by angle θ.Corresponding transformation matrix is expanded
A line and a row are opened up to obtain, and the matrix lower right corner is set to 1, remaining is filled with 0, then the rotation transformation relation is represented
For：.
When turning clockwise angle, θ around origin, between image (x, y) subject to registration and standard video (x ', y ')
When there is rotation transformation, condition x '=xcos θ+ysin θ, y '=- xsin θ+ycos θ are met, then the rotation transformation relation is represented
For：
Illustrated with continued reference to Fig. 1, in step s 2, registration apparatus is based on image subject to registration and preliminary registration image
The moving distance information of marginal point is screened to thick matching double points, so as to obtain accurate matching double points.
Preferably, the moving distance information includes image subject to registration maximum corresponding with the marginal point of preliminary registration image
Displacement and Minimum sliding distance, registration apparatus calculate image subject to registration maximum corresponding with the marginal point of preliminary registration image
Displacement and Minimum sliding distance；Then, registration apparatus is screened to thick matching double points so that the characteristic point pair after screening
Displacement between the maximum moving distance and Minimum sliding distance.
Obtained for example, referring to Fig. 4, the quadrangle ABCD as image subject to registration shown in Fig. 4 by rotation, translation transformation
It is used as the quadrangle A ' B ' C ' D ' of preliminary registration image.As can be seen that point on two image borders is the distance between corresponding always
More than the distance of the point between two image inside, and the distance between 4 summits in respective edges and the point on remaining edge
The distance between be equal, i.e. maximum moving distance dmaxWith Minimum sliding distance dminIt is present on edge.Therefore, it can choose
Try to achieve maximum moving distance d in four summits on edgemaxWith Minimum sliding distance dmin.That is maximum dmaxWith minimum range dminI.e.
Refer to the maximal and minmal value in distance between four boundary points A and A ', point B and B ', point C and C ', point D and D '.Registration apparatus
Based on constraints dmin<d<dmaxThick matching double points are screened, further to eliminate the mispairing point pair in thick matching double points.
Illustrated with continued reference to Fig. 1, in step s3, registration apparatus is based on the accurate matching double points, repeat step
S1 and step S2 operation, calculate new transformation model matrix to act on the preliminary registration image, new preliminary to obtain
Registering image, and the moving distance information of the marginal point based on new preliminary registration image and former preliminary registration image is accurate to this
Matching double points are screened, to obtain more accurate matching double points.
Preferably, registration apparatus is based on the accurate matching double points, by way of iteration, continuous repeat step S1 and step
Rapid S2 operation comes, and calculates new transformation model matrix to act on the preliminary registration image that previous process conversion is obtained, with
To new preliminary registration image, and the edge based on the new preliminary registration image and the preceding former preliminary registration image once obtained
Moving distance information between point is constantly screened to the accurate match point, until meeting predetermined stopping criterion for iteration.
For example, when carrying out nth iteration, registration apparatus calculates new transformation model matrix to act on (n-1)th time repeatedly
In generation, obtains new preliminary registration image, obtains new preliminary registration image, and new preliminary is matched somebody with somebody based on what the nth iteration was obtained
Quasi- image obtains moving distance information between the marginal point of new preliminary registration image with (n-1)th iteration and this accurate is matched
Point is constantly screened, until meeting predetermined stopping criterion for iteration.
Wherein, the stopping criterion for iteration includes but is not limited to any one of following：
1) iterations exceedes predetermined threshold；For example, iterations is more than 200 times；
2) displacement of the characteristic point pair after screening is less than zero.
Preferably, registration apparatus is based on the accurate matching double points, and new transformation model is calculated using RANSAC algorithms
Matrix is simultaneously screened to the accurate matching double points, to obtain more accurate matching double points.
It is highly preferred that registration apparatus iterates to calculate new transformation model matrix using least square method, so as to further carry
Registering precision between high image and standard picture subject to registration.
For example, building sample set P for the first group of thick matching double points obtained by SIFT algorithms, therefrom random selection has
There is the subset S (S ∈ P) of n matching double points, corresponding matching double points calculate initial transformation matrix M from subset S；To remainder
Subset carry out corresponding matrixing, selection and initial transformation matrix M error are less than the sample set of a certain threshold value, with subset
Point set S in S compositions*.Utilize interior point set S*Recalculate new transformation model matrix M '；Again it is sampled at random, resets S, repeatedly
For above procedure.Finally give relatively good 3*3 homography matrix H.It is special that homography matrix H determines two images
Relation is uniquely determined between levying a little, if (x, y) is target image characteristics point position, (x ', y ') is image characteristic point subject to registration
Position, then corresponding scale parameter be expressed as：
Wherein, S represents scale parameter,Represent homography matrix H.In homography matrix H
Parameter h31、h32、h33It is so as to represent affine transformation with H, i.e., final only to need three points to try to achieve remaining 6 for 0,0,1
Parameter.
After homography matrix is drawn, remaining data set is tested using the model, and finds out and meets the model
Matching double points number and cost function so that following cost function is minimum, so as to finally filter out, to meet the matrix relative
The matching pair answered：
According to a preferred embodiment of the present invention, registration apparatus is from the matching double points set after step S2 screenings
4 preferable matching double points are selected at random, transformation matrix H1 are obtained, and be designated as model M 1.Then, registration apparatus is in step
Compare the error between the remaining matching double points of calculating and above-mentioned model M 1 in S3, should if the error is less than certain threshold value
Matching double points add interior point set C；If optimal interior point set C ' is less than number in current interior point set C, assignment C '=C, simultaneously
Update iterations K；Repeat the above steps, until iterations is more than 200.
The method according to the invention, based on image subject to registration by between the corresponding marginal point before and after registration several times
Range information the matching double points obtained by SIFT algorithms are screened, so as to obtain more accurate matching double points, with
The general scheme based on SIFT algorithms is compared, and the method according to the invention improves the feature of image subject to registration and standard picture
The accuracy of matching, realizes the registration of not accurate enough the high score image in geometric position and standard video, so as to remote sensing shadow
Preferably registration effect is reached as match somebody with somebody on time.
Fig. 2 illustrates a kind of structural representation for being used to carry out image the registration apparatus of registration according to the present invention.
Computing device 1, screening plant 2 and coalignment 3 are included according to the registration apparatus of the present invention.
Multiple characteristic points in reference picture 2, thick matching double points of the computing device 1 based on image subject to registration and standard picture
It is right, the transformation model matrix for acting on image subject to registration is calculated, to obtain corresponding preliminary registration image.
Wherein, the computing device 1 represents the position of characteristic point using three-dimensional homogeneous coordinates, and the transformation model matrix is
Homography matrix.
Preferably, computing device 1 is based on Ransac algorithms, is selected from the thick matching double points of image subject to registration and standard picture
Matching preferable characteristic point is selected to calculating transformation model matrix.
Preferably, translation transformation relation and rotation transformation relation of the computing device 1 based on image subject to registration and standard picture,
Calculate the transformation model matrix for acting on image subject to registration.
Illustrated with continued reference to Fig. 2, shifting of the screening plant 2 based on image subject to registration Yu the marginal point of preliminary registration image
Dynamic range information is screened to thick matching double points, so as to obtain accurate matching double points.
Preferably, the moving distance information includes image subject to registration maximum corresponding with the marginal point of preliminary registration image
Displacement and Minimum sliding distance, the screening plant 2 further comprise apart from computing device (not shown) and son screening dress
Put (not shown).
Apart from computing device calculate image subject to registration maximum moving distance corresponding with the marginal point of preliminary registration image and
Minimum sliding distance；Then, sub- screening plant is screened to thick matching double points so that the movement of the characteristic point pair after screening away from
From between the maximum moving distance and Minimum sliding distance.
Obtained for example, referring to Fig. 4, the quadrangle ABCD as image subject to registration shown in Fig. 4 by rotation, translation transformation
It is used as the quadrangle A ' B ' C ' D ' of preliminary registration image.As can be seen that point on two image borders is the distance between corresponding always
More than the distance of the point between two image inside, and the distance between 4 summits in respective edges and the point on remaining edge
The distance between be equal, i.e. maximum moving distance dmaxWith Minimum sliding distance dminIt is present on edge.Therefore, it can choose
Try to achieve maximum moving distance d in four summits on edgemaxWith Minimum sliding distance dmin.That is maximum dmaxWith minimum range dminI.e.
Refer to the maximal and minmal value in distance between four boundary points A and A ', point B and B ', point C and C ', point D and D '.Registration apparatus
Based on constraints dmin<d<dmaxThick matching double points are screened, further to eliminate the mispairing point pair in thick matching double points.
Illustrated with continued reference to Fig. 2, coalignment 3 is based on the accurate matching double points, compute repeatedly device 1 and screening
The operation of device 2, calculates new transformation model matrix to act on the preliminary registration image, to obtain new preliminary registration figure
Picture, and the marginal point based on new preliminary registration image and former preliminary registration image moving distance information to the accurate match point
To screening, to obtain more accurate matching double points.
Preferably, coalignment 3 is based on the accurate matching double points, by way of iteration, constantly computes repeatedly device 1
With the operation of screening plant 2, calculate new transformation model matrix to act on the preliminary registration image that previous process conversion is obtained,
To obtain new preliminary registration image, and based on the new preliminary registration image and the preceding former preliminary registration image once obtained
Moving distance information between marginal point is constantly screened to the accurate match point, until meeting predetermined iteration ends bar
Part.
For example, when carrying out nth iteration, coalignment 3 calculates new transformation model matrix to act on (n-1)th time repeatedly
In generation, obtains new preliminary registration image, obtains new preliminary registration image, and new preliminary is matched somebody with somebody based on what the nth iteration was obtained
Quasi- image obtains moving distance information between the marginal point of new preliminary registration image with (n-1)th iteration and this accurate is matched
Point is constantly screened, until meeting predetermined stopping criterion for iteration.
Wherein, the stopping criterion for iteration includes but is not limited to any one of following：
1) iterations exceedes predetermined threshold；For example, iterations is more than 200 times；
2) displacement of the characteristic point pair after screening is less than zero.
Preferably, coalignment 3 is based on the accurate matching double points, and new transformation model is calculated using RANSAC algorithms
Matrix is simultaneously screened to the accurate matching double points, to obtain more accurate matching double points.
It is highly preferred that coalignment 3 iterates to calculate new transformation model matrix using least square method, so as to further carry
Registering precision between high image and standard picture subject to registration.
According to a preferred embodiment of the present invention, matching double points collection of the coalignment 3 after being screened by screening plant 2
4 preferable matching double points are selected at random in conjunction, transformation matrix H1 are obtained, and be designated as model M 1.Then, calculating is compared remaining
Under matching double points and above-mentioned model M 1 between error, if the error be less than certain threshold value, by the matching double points add in
Point set C；If optimal interior point set C ' is less than number in current interior point set C, assignment C '=C, while updating iterations K；
Repeat the above steps, until iterations is more than 200.
According to the solution of the present invention, based on image subject to registration by between the corresponding marginal point before and after registration several times
Range information the matching double points obtained by SIFT algorithms are screened, so as to obtain more accurate matching double points, with
The general scheme based on SIFT algorithms is compared, and the feature of image subject to registration and standard picture is improved according to the solution of the present invention
The accuracy of matching, realizes the registration of not accurate enough the high score image in geometric position and standard video, so as to remote sensing shadow
Preferably registration effect is reached as match somebody with somebody on time.
The software program of the present invention can realize steps described above or function by computing device.Similarly, originally
The software program (including related data structure) of invention can be stored in computer readable recording medium storing program for performing, for example, RAM is deposited
Reservoir, magnetically or optically driver or floppy disc and similar devices.In addition, some steps or function of the present invention can employ hardware to reality
It is existing, for example, as coordinating with processor so as to performing the circuit of each function or step.
In addition, the part of the present invention can be applied to computer program product, such as computer program instructions, when its quilt
When computer is performed, by the operation of the computer, the method according to the invention and/or technical scheme can be called or provided.
And the programmed instruction of the method for the present invention is called, it is possibly stored in fixed or moveable recording medium, and/or pass through
Broadcast or the data flow in other signal bearing medias and be transmitted, and/or be stored according to described program instruction operation
In the working storage of computer equipment.Here, including a device according to one embodiment of present invention, the device includes using
In the memory and processor for execute program instructions of storage computer program instructions, wherein, when the computer program refers to
When order is by the computing device, method and/or skill of the plant running based on foregoing multiple embodiments according to the present invention are triggered
Art scheme.
It is obvious to a person skilled in the art that the invention is not restricted to the details of above-mentioned one exemplary embodiment, Er Qie
In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, embodiment all should be regarded as exemplary, and be nonrestrictive, the scope of the present invention is by appended power
Profit is required rather than described above is limited, it is intended that all in the implication and scope of the equivalency of claim by falling
Change is included in the present invention.Any reference in claim should not be considered as to the claim involved by limitation.This
Outside, it is clear that the word of " comprising " one is not excluded for other units or step, and odd number is not excluded for plural number.That is stated in system claims is multiple
Unit or device can also be realized by a unit or device by software or hardware.The first, the second grade word is used for table
Show title, and be not offered as any specific order.
Claims (13)
1. it is a kind of be used for image carry out registration method, wherein, methods described using SIFT algorithms obtain image subject to registration with
The characteristic point of standard picture simultaneously obtains thick matching double points by screening, the described method comprises the following steps：
Multiple characteristic points pair in thick matching double points of a based on image subject to registration and standard picture, calculating acts on figure subject to registration
The transformation model matrix of picture, to obtain corresponding preliminary registration image；
Moving distance informations of the b based on image subject to registration and the marginal point of preliminary registration image is screened to thick matching double points,
So as to obtain accurate matching double points；
Wherein, it the described method comprises the following steps：
C is based on the accurate matching double points, and repeat step a and b operation, the new transformation model matrix of calculating are described to act on
Preliminary registration image, to obtain new preliminary registration image, and based on new preliminary registration image and former preliminary registration image
The moving distance information of marginal point is screened to the accurate matching double points, to obtain more accurate matching double points.
2. according to the method described in claim 1, wherein, the moving distance information includes image subject to registration and preliminary registration figure
The corresponding maximum moving distance of marginal point and Minimum sliding distance of picture, the step b comprise the following steps：
- calculate image subject to registration maximum moving distance corresponding with the marginal point of preliminary registration image and Minimum sliding distance；
- thick matching double points are screened so that the displacement of the characteristic point pair after screening is between the maximum moving distance
Between Minimum sliding distance.
3. according to the method described in claim 1, wherein, the step c comprises the following steps：
- the accurate matching double points are based on, by way of iteration, continuous repeat step a and b operation calculates new conversion
Model matrix acts on the preliminary registration image that previous process conversion is obtained, and to obtain new preliminary registration image, and is based on
Moving distance information between the marginal point of the new preliminary registration image and the preceding former preliminary registration image once obtained is to this
Accurate match point is constantly screened, until meeting predetermined stopping criterion for iteration.
4. method according to claim 3, wherein, the stopping criterion for iteration includes any one of following：
- iterations exceedes predetermined threshold；
The displacement of characteristic point pair after-screening is less than zero.
5. method according to claim 3, wherein, the step c comprises the following steps：
- the accurate matching double points are based on, new transformation model matrix is calculated using RANSAC algorithms and to the accurate matching
Point is to screening, to obtain more accurate matching double points.
6. according to the method described in claim 1, wherein, methods described represents the position of characteristic point using three-dimensional homogeneous coordinates,
The transformation model matrix is homography matrix, and the step a comprises the following steps：
- translation transformation relation and rotation transformation relation based on image subject to registration and standard picture, calculating acts on figure subject to registration
The transformation model matrix of picture.
7. according to the method described in claim 1, wherein, the described method comprises the following steps：
- use SIFT algorithms extraction image subject to registration and multiple characteristic points and feature point description of standard picture；
- by carrying out screening the thick matching double points for obtaining image subject to registration and standard picture to each characteristic point.
8. a kind of registration apparatus for being used to carry out image registration, wherein, the registration apparatus obtains waiting to match somebody with somebody using SIFT algorithms
The characteristic point of quasi- image and standard picture simultaneously obtains thick matching double points by screening, and the registration apparatus includes：
Computing device, for multiple characteristic points pair in the thick matching double points based on image subject to registration and standard picture, calculates and makees
For the transformation model matrix of image subject to registration, to obtain corresponding preliminary registration image；
Screening plant, for the moving distance information based on image subject to registration and the marginal point of preliminary registration image to thick match point
To screening, so as to obtain accurate matching double points；
Coalignment, for based on the accurate matching double points, computing repeatedly the operation of device and screening plant, calculates new change
Change model matrix to act on the preliminary registration image, to obtain new preliminary registration image, and based on new preliminary registration
The moving distance information of the marginal point of image and former preliminary registration image is screened to the accurate matching double points, to obtain more
Accurate matching double points.
9. registration apparatus according to claim 8, wherein, the moving distance information includes image subject to registration with tentatively matching somebody with somebody
The corresponding maximum moving distance of marginal point and Minimum sliding distance of quasi- image, the screening plant include：
Apart from computing device, for calculate image subject to registration maximum moving distance corresponding with the marginal point of preliminary registration image and
Minimum sliding distance；
Sub- screening plant, for being screened to thick matching double points so that the displacement of the characteristic point pair after screening is between institute
State between maximum moving distance and Minimum sliding distance.
10. registration apparatus according to claim 8, wherein, the coalignment is used for：
- the accurate matching double points are based on, by way of iteration, constantly repeat the computing device and the screening plant
Operation, calculates new transformation model matrix to act on the preliminary registration image that previous process conversion is obtained, with obtain it is new just
The registering image of step, and between the marginal point based on the new preliminary registration image and the preceding former preliminary registration image once obtained
Moving distance information is constantly screened to the accurate match point, until meeting predetermined stopping criterion for iteration.
11. registration apparatus according to claim 10, wherein, the stopping criterion for iteration includes any one of following：
- iterations exceedes predetermined threshold；
The displacement of characteristic point pair after-screening is less than zero.
12. registration apparatus according to claim 10, wherein, the coalignment is used for：
- the accurate matching double points are based on, new transformation model matrix is calculated using RANSAC algorithms and to the accurate matching
Point is to screening, to obtain more accurate matching double points.
13. registration apparatus according to claim 8, wherein, the registration apparatus represents feature using three-dimensional homogeneous coordinates
The position of point, the transformation model matrix is homography matrix, and the computing device is used for：
- translation transformation relation and rotation transformation relation based on image subject to registration and standard picture, calculating acts on figure subject to registration
The transformation model matrix of picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710278075.XA CN107170001A (en) | 2017-04-25 | 2017-04-25 | Method and apparatus for carrying out registration to image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710278075.XA CN107170001A (en) | 2017-04-25 | 2017-04-25 | Method and apparatus for carrying out registration to image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107170001A true CN107170001A (en) | 2017-09-15 |
Family
ID=59812963
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710278075.XA Withdrawn CN107170001A (en) | 2017-04-25 | 2017-04-25 | Method and apparatus for carrying out registration to image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107170001A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108765277A (en) * | 2018-06-04 | 2018-11-06 | 上海联影医疗科技有限公司 | Image split-joint method, device, computer equipment and storage medium |
| CN110070090A (en) * | 2019-04-25 | 2019-07-30 | 上海大学 | A kind of logistic label information detecting method and system based on handwriting identification |
| WO2019233422A1 (en) * | 2018-06-04 | 2019-12-12 | Shanghai United Imaging Healthcare Co., Ltd. | Devices, systems, and methods for image stitching |
| CN110647301A (en) * | 2018-06-27 | 2020-01-03 | 光宝电子(广州)有限公司 | Augmented reality system and image display method thereof |
| CN110751680A (en) * | 2019-10-17 | 2020-02-04 | 中国科学院云南天文台 | Image processing method with fast alignment algorithm |
| CN111429493A (en) * | 2020-03-20 | 2020-07-17 | 青岛联合创智科技有限公司 | Method for matching feature points among multiple images |
| WO2020259365A1 (en) * | 2019-06-27 | 2020-12-30 | Oppo广东移动通信有限公司 | Image processing method and device, and computer-readable storage medium |
| CN113100941A (en) * | 2021-04-12 | 2021-07-13 | 中国科学院苏州生物医学工程技术研究所 | Image registration method and system based on SS-OCT (scanning and optical coherence tomography) surgical navigation system |
| CN114694040A (en) * | 2022-05-31 | 2022-07-01 | 潍坊绘圆地理信息有限公司 | Data identification method for optical remote sensing data block registration based on dynamic threshold |
| CN114897947A (en) * | 2022-04-08 | 2022-08-12 | 苏州大学 | Thermal infrared and visible light image synchronous registration method based on time-space unification |
| CN116452647A (en) * | 2023-06-15 | 2023-07-18 | 广东工业大学 | Dynamic image registration method, system and device based on matching pursuit |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104660900A (en) * | 2013-10-30 | 2015-05-27 | 株式会社摩如富 | Image Processing Device, Image Processing Method And Recording Medium |
-
2017
- 2017-04-25 CN CN201710278075.XA patent/CN107170001A/en not_active Withdrawn
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104660900A (en) * | 2013-10-30 | 2015-05-27 | 株式会社摩如富 | Image Processing Device, Image Processing Method And Recording Medium |
Non-Patent Citations (1)
| Title |
|---|
| 王新年等: "改进的基于SIFT和RANSAC的图像配准方法", 《新型工业化》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019233422A1 (en) * | 2018-06-04 | 2019-12-12 | Shanghai United Imaging Healthcare Co., Ltd. | Devices, systems, and methods for image stitching |
| CN108765277A (en) * | 2018-06-04 | 2018-11-06 | 上海联影医疗科技有限公司 | Image split-joint method, device, computer equipment and storage medium |
| US12079953B2 (en) | 2018-06-04 | 2024-09-03 | Shanghai United Imaging Healthcare Co., Ltd. | Devices, systems, and methods for image stitching |
| CN108765277B (en) * | 2018-06-04 | 2021-05-07 | 上海联影医疗科技股份有限公司 | Image splicing method and device, computer equipment and storage medium |
| US11763424B2 (en) | 2018-06-04 | 2023-09-19 | Shanghai United Imaging Healthcare Co., Ltd. | Devices, systems, and methods for image stitching |
| CN110647301B (en) * | 2018-06-27 | 2023-08-08 | 光宝电子(广州)有限公司 | Augmented reality system and image display method thereof |
| CN110647301A (en) * | 2018-06-27 | 2020-01-03 | 光宝电子(广州)有限公司 | Augmented reality system and image display method thereof |
| CN110070090A (en) * | 2019-04-25 | 2019-07-30 | 上海大学 | A kind of logistic label information detecting method and system based on handwriting identification |
| CN110070090B (en) * | 2019-04-25 | 2021-06-25 | 上海大学 | Logistics label information detection method and system based on handwritten character recognition |
| WO2020259365A1 (en) * | 2019-06-27 | 2020-12-30 | Oppo广东移动通信有限公司 | Image processing method and device, and computer-readable storage medium |
| CN110751680A (en) * | 2019-10-17 | 2020-02-04 | 中国科学院云南天文台 | Image processing method with fast alignment algorithm |
| CN111429493A (en) * | 2020-03-20 | 2020-07-17 | 青岛联合创智科技有限公司 | Method for matching feature points among multiple images |
| CN113100941B (en) * | 2021-04-12 | 2022-03-08 | 中国科学院苏州生物医学工程技术研究所 | Image registration method and system based on SS-OCT (scanning and optical coherence tomography) surgical navigation system |
| CN113100941A (en) * | 2021-04-12 | 2021-07-13 | 中国科学院苏州生物医学工程技术研究所 | Image registration method and system based on SS-OCT (scanning and optical coherence tomography) surgical navigation system |
| CN114897947A (en) * | 2022-04-08 | 2022-08-12 | 苏州大学 | Thermal infrared and visible light image synchronous registration method based on time-space unification |
| CN114694040A (en) * | 2022-05-31 | 2022-07-01 | 潍坊绘圆地理信息有限公司 | Data identification method for optical remote sensing data block registration based on dynamic threshold |
| CN116452647A (en) * | 2023-06-15 | 2023-07-18 | 广东工业大学 | Dynamic image registration method, system and device based on matching pursuit |
| CN116452647B (en) * | 2023-06-15 | 2023-12-08 | 广州安特激光技术有限公司 | Dynamic image registration method, system and device based on matching pursuit |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107170001A (en) | Method and apparatus for carrying out registration to image | |
| TWI720447B (en) | Image positioning method and system thereof | |
| CN109272537B (en) | Panoramic point cloud registration method based on structured light | |
| CN104867126B (en) | Based on point to constraint and the diameter radar image method for registering for changing region of network of triangle | |
| JP3435084B2 (en) | Optical flow estimation method and image composition method | |
| Sarlin et al. | Lamar: Benchmarking localization and mapping for augmented reality | |
| CN107194959A (en) | The method and apparatus that image registration is carried out based on section | |
| Lee et al. | Real-time disparity estimation algorithm for stereo camera systems | |
| CN104463859B (en) | A kind of real-time video joining method based on tracking specified point | |
| CN106934824B (en) | Global non-rigid registration and reconstruction method for deformable object | |
| CN111553985B (en) | O-graph pairing European three-dimensional reconstruction method and device | |
| CN108961385B (en) | SLAM composition method and device | |
| CN107220996A (en) | A kind of unmanned plane linear array consistent based on three-legged structure and face battle array image matching method | |
| CN111126418A (en) | Oblique image matching method based on planar perspective projection | |
| Wan et al. | Drone image stitching using local mesh-based bundle adjustment and shape-preserving transform | |
| CN115830135A (en) | Image processing method and device and electronic equipment | |
| CN115564639A (en) | Background blurring method and device, computer equipment and storage medium | |
| CN113642397A (en) | Object length measuring method based on mobile phone video | |
| CN106408654B (en) | A kind of creation method and system of three-dimensional map | |
| Tsai et al. | Polygon‐based texture mapping for cyber city 3D building models | |
| Kim et al. | Ep2p-loc: End-to-end 3d point to 2d pixel localization for large-scale visual localization | |
| CN112002007A (en) | Model obtaining method and device based on air-ground image, equipment and storage medium | |
| Zhou et al. | Stn-homography: estimate homography parameters directly | |
| JP3540696B2 (en) | Image synthesizing method, image synthesizing device, recording medium storing image synthesizing program | |
| Yu et al. | Multi-view 2D–3D alignment with hybrid bundle adjustment for visual metrology |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20180514 Address after: 100085 Beijing Haidian District Xueyuan Road armor 5, 2 square B North -1081 Applicant after: Beijing star science and Technology Co., Ltd. Address before: 100085 Beijing Haidian District Xueyuan Road a 5, 2 cottage B-1011 Applicant before: Beijing to gather information technology Co. Ltd. |
|
| TA01 | Transfer of patent application right | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20170915 |
|
| WW01 | Invention patent application withdrawn after publication |