CN103310453B - A kind of fast image registration method based on subimage Corner Feature - Google Patents

Abstract

The invention provides a kind of fast image registration method based on subimage Corner Feature, concrete steps comprise: step one, choose with reference to subgraph and subgraph subject to registration; From reference picture, choose a subgraph as with reference to subgraph, from image subject to registration, choose the coordinate space subgraph identical with reference subgraph as subgraph subject to registration; Step 2, extraction are with reference to subgraph and subgraph angle point subject to registration; Step 3, carry out feature interpretation to reference to the angle point that subgraph and subgraph subject to registration extract, obtain the proper vector of each angle point; Step 4, carries out similarity measurement and characteristic matching by subgraph to be matched with reference to the proper vector of angle point on subgraph, finally obtains K matching double points; Step 5, based on K matching double points, adopt least square method to calculate transformation matrix H between image subject to registration and reference picture, based on transformation matrix H by image registration subject to registration on reference picture.The present invention greatly can improve the matching speed of image while meeting images match precision.

Description

A kind of fast image registration method based on subimage Corner Feature

Technical field

The invention belongs to image registration techniques field, be specifically related to a kind of fast image registration method based on subimage Corner Feature.

Background technology

The application of image registration widely, as fields such as pattern-recognition, self-navigation, medical diagnosis, computer visions.In the registration of image, carry out many research work at present, propose multiple method for registering images.The research of current great majority to image registration concentrates on feature extraction, feature interpretation, similarity measurement, the comparison etc. of multiple method for registering, and the real-time of less concern registration.

Common method for registering can be divided into two classes: the method for registering of feature based, as Harris cornerpoints method, SIFT method etc.; Based on the method for registering in region, as mutual information, FMT etc.Wherein do not need to extract feature based on the method for registering in region, be applicable to the situation that half-tone information is greater than structural information, and require that the gamma function of two width images must similar or at least statistical correlation; From geometric angle, it only can process the situation of translation and small angle rotation, and big angle rotary or scaling must mean the raising of computation complexity and time complexity, and therefore the scope of application is narrower.And the method for registering of feature based can the diverse image of registration two width natural quality, and adapt to geometry complicated between two width images and optical distortion, therefore become the focus of Recent study.But its bottleneck how correctly to detect characteristic of correspondence, and carry out the feature interpretation of low complex degree, robust, to improve the efficiency of match search.Due to shooting environmental and scenery distribution, make to take the image that obtains in contrast, text structure, the aspect distributed poles such as textural characteristics are uneven.When directly carrying out feature extraction on large figure (being called for short large figure method), the noise spot of correct coupling may be become in the unique point of feature Fuzzy extracted region on the contrary, cause mismatch; In addition, a large amount of unique points extends the scope of search volume, causes search efficiency and real-time to decline.

The technology of current this problem of solution has: 1. " a kind of method for registering combined with template matches by mutual information ", mutual information is adopted to be the template matches that similarity criteria carries out image, obtain the Matching sub-image of candidate, by the spatial relationship of large figure remainder subject to registration and template and subgraph, obtain the large figure after registration, calculate the mutual information according to the figure subject to registration in each candidate's subgraph registration situation respectively, obtain the subgraph corresponding to maximum mutual information, determine final registration result.But the method, when the image little to gray scale difference carries out registration, easily occur mismatch phenomenon, and the registration time is long.2. based on the method for registering images of wavelet transformation, utilize wavelet coefficient to choose effective subgraph, and utilize wavelet transformation that image is divided into some levels, utilize cross-correlation coefficient as similarity measure, realize the registration of image finally by iterative refinement algorithm.But the method relates to Calculation of correlation factor, wavelet coefficient subgraph is chosen and the more step consuming time such as iteration refinement, makes algorithm complex high, realizes difficulty large, poor real.

Summary of the invention

Given this, the present invention proposes a kind of rapid registering method based on subimage Corner Feature on the basis of improving feature based method for registering, improves robustness and the real-time of registration under being intended to meet the prerequisite of registration accuracy.

In order to solve the problems of the technologies described above, the present invention is achieved in that

Based on a fast image registration method for subimage Corner Feature, concrete steps comprise:

Step one, choose with reference to subgraph and subgraph subject to registration;

From reference picture, choose a subgraph as with reference to subgraph, from image subject to registration, choose the coordinate space subgraph identical with reference subgraph as subgraph subject to registration;

The angle point of step 2, extraction reference subgraph and subgraph subject to registration;

Step 3, carry out feature interpretation to reference to the angle point that subgraph and subgraph subject to registration extract, obtain the proper vector of each angle point;

Step 4, carry out similarity measurement and characteristic matching by subgraph to be matched with reference to the proper vector of angle point on subgraph, finally obtain K matching double points;

The detailed process of this step is:

1) for the angle point p that each extracts _i, find and p _ip contiguous point forms p _idistance neighborhood, i=1,2 ... N, N by two width subgraphs total number of extraction angle point;

2) calculate each Corner Feature vector in subgraph subject to registration successively and, to the mahalanobis distance with reference to all Corner Feature vectors in subgraph, mahalanobis distance is less than setting threshold value d _mth1two angle points be defined as matching double points, multiple matching double points forms coupling queue;

3) in coupling queue, reject not at the matching double points in respective distances field, obtain K matching double points;

Step 5, based on K matching double points, adopt least square method to calculate transformation matrix H between image subject to registration and reference picture, utilize described transformation matrix H by image registration subject to registration on reference picture.

Further, of the present invention is strong, the obvious width subgraph of architectural feature of contrast on reference picture with reference to subgraph.

Further, the process of choosing with reference to subgraph of the present invention is:

Secondly first become with reference to Iamge Segmentation the subgraph that n size is identical, calculate entropy and the average gradient of each subgraph, then select entropy and the maximum subgraph of average gradient sum as with reference to subgraph.

Further, the present invention extracts with reference to subgraph identical with the method for subgraph angle point subject to registration, and detailed process is:

First, based on Harris angular-point detection method, detect subgraph angle point; Secondly, the method taking neighborhood non-maximum restraining and total amount to suppress is screened the angle point that initial detecting goes out, and extracts top n angle point; Then, remove the angle point be positioned on subgraph borderline region, thus extract required angle point.

Further, subgraph borderline region of the present invention is distance border width is the rectangular area of 4* σ, and wherein σ is the Gaussian smoothing factor.

Further, the present invention preferably utilizes 12 of angle point dimension gradient vectors to represent the feature interpretation matrix of angle point.

Further, the present invention is at calculating mahalanobis distance d _min the process of (i, j), utilize orthogonal matrix P and diagonal matrix D to represent the inverse matrix C of covariance matrix C ^-1, convert the calculating of mahalanobis distance to Euclidean distance d _fcalculating;

$d_M(i,j)=\sqrt{{(X_i^1-X_j^2)}^t{C}^{-1}(X_i^1-X_j^2)}$

$=\sqrt{{(X_i^1-X_j^2)}^t{P}^t\sqrt{D}\·\sqrt{D}P(X_i^1-X_j^2)}$

$=d_E(\sqrt{D}{\mathrm{PX}}_i^1,\sqrt{D}P{X}_j^2)$

Wherein, represent the proper vector of i-th angle point on subgraph subject to registration, represent the proper vector with reference to a jth angle point on subgraph.

Beneficial effect:

The first, the present invention is by large figure subject to registration with extract subgraph with reference on large figure, and is arranged the matching double points extracting high matching degree by threshold value, thus makes the present invention can coupling fast, accurately between subject to registration and reference diagram.

The second, the present invention is at large figure subject to registration with reference to the subgraph large figure choosing contrast and clear in structure, then subsequent treatment is carried out according to the method for registering of feature based, namely the operation that the computation complexity such as extract minutiae and match search is high is directly carried out on subgraph instead of large figure, improve the accuracy that character pair point detects, more accurately estimate transformation matrix sooner, finally improve precision and the real-time of registration.

Three, select the Harris angular-point detection method based on improving, what it detected is gray scale and the violent maximal point of graded in angle point subrange., matching primitives complexity excessive, slow-footed problem more for extraction angle point number, the method that the present invention takes neighborhood non-maximum restraining and total amount to suppress is screened the angle point that initial detecting goes out, and remove in 4* σ in border in R (σ is the Gaussian smoothing factor) institute a little, object makes the angle point of reservation as far as possible at the middle position of entire image, avoid because rotation, translation, convergent-divergent etc. make part angle point shift out, improve the repetition rate extracting angle point.

Four, the present invention carries out Steerable filter to each corner pixels, provides the total derivative along gradient direction; For the gray difference that reply causes because of affined transformation, removing first order derivative, the proper vector being angle point with 12 dimension gradient vectors.Compare, based on the method for SIFT, its descriptor is 128 dimensions, and the PCA-SIFT of improvement is 36 dimensions.The present invention has larger minimizing in the complicacy and calculated amount of algorithm.

Five, mahalanobis distance is solved the calculating converting simple Euclidean distance to by the present invention, can avoid the inversion operation of redundancy, reduce the space requirement to internal memory when hardware implementing, improves the real-time of algorithm.

Accompanying drawing explanation

Fig. 1 is subgraph method registration schematic flow sheet.

Embodiment

To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.

First suppose that between image, distortion model is affined transformation, its mathematical notation is as follows:

$\left[\begin{array}{c}{x_i}^{\′}\\ {y_i}^{\′}\end{array}\right]=s\left[\begin{array}{cc}\cos \mathrm{\θ}& -\sin \mathrm{\θ}\\ \sin \mathrm{\θ}& \cos \mathrm{\θ}\end{array}\right]*\left[\begin{array}{c}{x}_i\\ y_i\end{array}\right]+\left[\begin{array}{c}{t}_x\\ t_y\end{array}\right]=\left[\begin{array}{cc}{a}_1& a_2\\ a_3& a_4\end{array}\right]*\left[\begin{array}{c}{x}_i\\ y_i\end{array}\right]+\left[\begin{array}{c}{t}_x\\ t_y\end{array}\right]---\left(1\right)$

Wherein, s is scale factor, and θ is rotation angle, t _xfor x direction translational movement, t _yfor y direction translational movement, (x _i', y _i') be the point on image after distortion, (x _i, y _i) be the point on image before distortion.

As shown in Figure 1, based on the method for registering images of Sub-Image Feature, concrete steps are:

Step one, to choose with reference to subgraph and subgraph subject to registration, from reference picture, namely choose a subgraph as with reference to subgraph, from image subject to registration, choose the coordinate space subgraph identical with reference subgraph as subgraph subject to registration.

The present invention preferably extracts two width subgraphs in the following ways:

First the image that contrast is strong, architectural feature is obvious, size is identical is chosen in the upper same coordinate space of image subject to registration and reference picture (image subject to registration is the image after reference picture distorts) as subgraph subject to registration with reference to subgraph.

Subgraph choose the method adopting entropy and gradient-norm to combine:

First become with reference to Iamge Segmentation the subgraph that n (generally getting n >=4) individual size is identical, calculate entropy and the average gradient of each subgraph again, and select entropy and the maximum subgraph of average gradient sum as effectively with reference to subgraph, then choose a subgraph as effective subgraph subject to registration in image same coordinate space subject to registration.

Entropy can be used as the tolerance of image local area information, is normally defined

$E=-\underset{i=1}{\overset{L}{\mathrm{\Σ}}}P\left(\mathrm{bi}\right)\ln P\left(\mathrm{bi}\right)$

Wherein, the pixel progression of L-subgraph, the probability of P (bi)-i-th grade pixel brightness value.

Average gradient reflects the readability of image, and also reflect minor detail contrast and texture transformation feature in image, its formula is simultaneously:

$\▿\stackrel{\‾}{G}=\frac{1}{M\×N}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{[\mathrm{\Δxf}{(i,j)}^2+\mathrm{\Δyf}{(i,j)}^2]}^{1/2}$

Wherein, Δ xf (i, j) and Δ yf (i, j) represents the first order difference of pixel (i, j) on x direction and y direction respectively, and in this formula, M and N represents line number and the columns of subimage respectively.

According to entropy and the average gradient of each subgraph in above-mentioned formulae discovery image, and get entropy and the maximum effective registration subgraph of conduct of average gradient sum.

Step of registration is below except final step, and remaining is all carry out on subgraph.

The angle point of step 2, extraction reference subgraph and subgraph subject to registration.

That use the present invention below, existing angle point grid technology is described:

Harris Corner Detection Algorithm only relates to the first order derivative of image, first defines matrix M:

$M=G\&CircleTimes;\left[\begin{array}{cc}{I_x}^2& I_x{I}_y\\ I_x{I}_y& {I_y}^2\end{array}\right]=\left[\begin{array}{cc}<{I_x}^2>& <I_x{I}_y>\\ <I_x{I}_y>& <{I_y}^2>\end{array}\right]$

Wherein I _xfor the gradient in the x direction of image I; I _yfor the gradient in the y direction of image I; G is Gaussian template; <> represents Gaussian template function convolution: $<I_x^2>=G\&CircleTimes;I_x^2,$ $<I_y^2>=G\&CircleTimes;I_y^2,<I_x{I}_y>=G\&CircleTimes;I_x{I}_y,$ represent convolution.The angle point response function CRF adopting Nobel to propose defines:

$\mathrm{CRF}=\frac{\mathrm{trace}\left(M\right)}{\det \left(M\right)}=\frac{<I_x^2>+<I_y^2>}{<I_x^2>\&CenterDot;<I_y^2>-{<I_x{I}_y>}^2}---\left(16\right)$

Wherein, det is determinant of a matrix; Trace is matrix trace; The Local modulus maxima of CRF is angle point.

The angle point number proposed for existing method is more, excessive, the slow-footed problem of the matching primitives complexity caused, the method that the present invention takes neighborhood non-maximum restraining and total amount to suppress is screened the angle point that initial detecting goes out, and namely gets a front topN angle point by the method for the threshold value and sequence that arrange CRF and forms last angle point collection R; Then the institute in removing R neutron image border 4* σ (σ is the Gaussian smoothing factor) a little, object makes the angle point of reservation as far as possible at the middle position of entire image, avoid because rotation, translation, convergent-divergent etc. make part angle point shift out, improve the repetition rate extracting angle point.

Step 3, carry out feature interpretation to reference to the angle point that subgraph and subgraph subject to registration extract, obtain the proper vector of each angle point;

The detailed process of this step is:

Utilize 12 of each angle point dimension gradient vectors to be used as the proper vector of angle point in the present invention, because it is 12 dimension data, therefore greatly can improve registration speed of the present invention.Concrete principle is described as follows:

The nervous physiology experiment that Young (1987) carries out shows, human retina and cerebral cortex receptive field section can be simulated by Gaussian derivative.Therefore the present invention is to each angle point of trying to achieve, and asks its 4 rank Gauss's partial derivatives

G＝[G _x，G _y，G _xx，G _xy，G _yy，G _xxx，G _xxy，G _xyy，G _yyy，G _xxxx，G _xxxy，G _xxyy，G _xyyy，G _yyyy]。

Wherein G ₁=[G _x, G _y], ask according to G1

$\cos \left(\mathrm{\&theta;}\right)=\frac{G_x}{\left|G_1\right|}$

$\sin \left(\mathrm{\&theta;}\right)=\frac{G_y}{\left|G_1\right|}$

Rotation matrix:

$L=\left[\begin{array}{cc}\cos \mathrm{\&theta;}& \sin \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\end{array}\right]$

G ₂＝[G _xx，G _xy，G _yy]

The symmetric tensor of G2: g ₂=[G _xx, G _xy, G _xy, G _yy]

Tensor transition matrix:

$M_2=\left[\begin{array}{cccc}{\cos }^2\mathrm{\&theta;}& \cos \mathrm{\&theta;}\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\sin \mathrm{\&theta;}& {\sin }^2\mathrm{\&theta;}\\ -\cos \mathrm{\&theta;}\sin \mathrm{\&theta;}& -{\sin }^2\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}& \cos \mathrm{\&theta;}\sin \mathrm{\&theta;}\\ {\sin }^2\mathrm{\&theta;}& -\cos \mathrm{\&theta;}\sin \mathrm{\&theta;}& -\cos \mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\end{array}\right]$

D (1: 3)-M ₂* g ₂(Matrix Multiplication)

G ₃＝[G _xxx，G _xxy，G _xyy，G _yyy]

The symmetric tensor of G3: g ₃=[G _xxx, G _xxy, G _xxy, G _xyy, G _xxy, G _xyy, G _xyy, G _yyy]

Tensor transition matrix:

$M_3=\left[\begin{array}{cccccccc}{\cos }^3\mathrm{\&theta;}& -{\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& -{\sin }^3\mathrm{\&theta;}\\ -{\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& -\cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& -\cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& {\sin }^3\mathrm{\&theta;}& -{\cos }^3\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}\\ \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& -{\sin }^3\mathrm{\&theta;}& -{\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& -\cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& -{\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^3\mathrm{\&theta;}& -{\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}\\ -{\sin }^3\mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}{\sin }^2\mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^2\mathrm{\&theta;}\sin \mathrm{\&theta;}& {\cos }^3\mathrm{\&theta;}\end{array}\right]$

D(4:7)-M ₃*g ₃

The like, obtain D (8:12), using the descriptor of D (1:12) as angle point.Compare, its descriptor of method based on SIFT is 128 dimensions, and its descriptor of PCA-SIFT method of improvement is 36 dimensions.Description of the invention is 12 dimensions, therefore in the complicacy and calculated amount of algorithm, has larger minimizing.

Step 4: carry out similarity measurement and characteristic matching by subgraph subject to registration with reference to the proper vector of angle point on subgraph, obtain K matching double points;

1) for the angle point p that each extracts _i, find and p _ip contiguous point forms p _idistance neighborhood, i=1,2 ... N, N by two width subgraphs total number of extraction angle point; Carried out step 1) after, two width subimages respectively define N number of distance neighborhood altogether.

2) calculate each Corner Feature vector in subgraph subject to registration successively and, to the mahalanobis distance with reference to all Corner Feature vectors in subgraph, mahalanobis distance is less than setting threshold value d _mth1two angle points be defined as matching double points, multiple matching double points forms coupling queue 1.

3) in 1, reject not at the matching double points in respective distances field, obtain K matching double points in coupling queue;

Detailed process is: find out matching double points distance neighborhood no and nm separately successively, and forms matching double points matrix M (p*p element) between two; In M, utilize coupling queue 1, finding out can a worthy angle point in distance field nm and distance field no, and add up in nm the angle point number m meeting matching relationship, if m >=p (p is parameter preset), then definition distance neighborhood no and nm is corresponding distance field, and the matching double points on definition distance neighborhood no and nm is final matching double points, otherwise reject not at the matching double points (namely deleting the matching double points on distance neighborhood no and nm from coupling queue 1) in respective distances field, the like, queue 2 must be mated; K matching double points before finally getting in coupling queue 2, the K an obtained matching double points is the matching double points that similarity is higher on the whole, and K=4 is to matching double points for the present invention's better reservation, while guarantee matching precision, can improve matching speed like this.

The present invention utilizes the affine-invariant features of mahalanobis distance to carry out angle point invariant features similarity measurement.

Mahalanobis distance be solved to prior art, existing it to be briefly described: for the sample space (being feature space here) be made up of n point (t represents transposition, m representation dimension), wherein any sample point to another sample space $X^2=\{{(x_{11}^2,\&CenterDot;\&CenterDot;\&CenterDot;x_{1m}^2)}^t,\&CenterDot;\&CenterDot;\&CenterDot;,{(x_{\mathrm{nm}}^2,\&CenterDot;\&CenterDot;\&CenterDot;x_{\mathrm{nm}}^2)}^t$ In arbitrary sample point $X_j^2={(x_{j1}^2,\&CenterDot;\&CenterDot;\&CenterDot;x_{\mathrm{jm}}^2)}^t$ Mahalanobis distance be:

$d_M(i,j)=\sqrt{{(X_i^1-X_j^2)}^t{C}^{-1}(X_i^1-X_j^2)}$

Wherein, C represents covariance matrix; C ^-1represent the inverse matrix of C.

Assuming that with being respectively subgraph subject to registration and the proper vector with reference to one group of matching double points between subgraph, is 12 dimension invariant features vectors here.

Because covariance matrix is a real symmetric tridiagonal matrices, the present invention decomposes as follows to mahalanobis distance: C ^-1=P ^tdP, P is orthogonal matrix here, and D is diagonal matrix, then:

$\begin{array}{c}{d}_M(i,j)=\sqrt{{(X_i^1-X_j^2)}^t{P}^t\sqrt{D}\&CenterDot;\sqrt{D}P(X_i^1-X_j^2)}\\ =d_E(\sqrt{D}P{X}_i^1,\sqrt{D}P{X}_j^2)\end{array}---\left(18\right)$

Two angle points are calculated according to above formula with corresponding mahalanobis distance, the method is by mahalanobis distance d _mbe converted to simple Euclidean distance d _ecalculating, the inversion operation of redundancy can be avoided when hardware implementing, reduce the space requirement to internal memory, improve the real-time of algorithm.

Different from the method for traditional search matching strategy, the algorithm of this Binding distance neighborhood and threshold value, achieves the second degree matches process of " by thick to smart ", search volume and calculated amount is reduced with the change tread of topN threshold value, makes matching rate higher.

Step 5, determine subgraph subject to registration and with reference to the matching relationship between angle point in subgraph after, based on K matching double points, adopt least-squares algorithm, directly calculate the transformation matrix H between conversion image subject to registration and reference picture, afterwards based on transformation matrix H, adopt bilinear interpolation to treat registering images and rebuild.

Such as: establish (x _i, y _i) and (x _i', y _i') (i=1,2,3,4) be coupling point set A and B final on image A subject to registration and template image B respectively; The method of affine transformation parameter is to utilize least square method to estimate:

A*H＝B

$A=\left[\begin{array}{cccccc}{x}_1& y_1& 0& 0& 1& 0\\ x_2& y_2& 0& 0& 1& 0\\ x_3& y_3& 0& 0& 1& 0\\ x_4& y_4& 0& 0& 1& 0\\ 0& 0& x_1& y_1& 0& 1\\ 0& 0& x_2& y_2& 0& 1\\ 0& 0& x_3& y_3& 0& 1\\ 0& 0& x_4& y_4& 0& 1\end{array}\right]$

B＝[x ₁′；x ₂′；x ₃′；x ₄′；y ₁′；y ₂′；y ₃′y ₄′；]

Therefore, H=pinv (A) * B

Visible, the present invention is by reasonably choosing subgraph, and in feature extraction, the method for registering of feature based is improved in the aspects such as invariant features describes, similarity measurement, and these methods is applied to effective subgraph of extraction, thus carries out the registration of image sooner more accurately.

In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (7)

1. based on a fast image registration method for subimage Corner Feature, it is characterized in that, concrete steps comprise:

Step one, choose with reference to subgraph and subgraph subject to registration;

From reference picture, choose a subgraph as with reference to subgraph, from image subject to registration, choose the coordinate space subgraph identical with reference subgraph as subgraph subject to registration;

The angle point of step 2, extraction reference subgraph and subgraph subject to registration;

Step 3, carry out feature interpretation to reference to the angle point that subgraph and subgraph subject to registration extract, obtain the proper vector of each angle point;

Step 4, carry out similarity measurement and characteristic matching by subgraph to be matched with reference to the proper vector of angle point on subgraph, finally obtain K matching double points;

The detailed process of this step is:

1) for the angle point p that each extracts _i, find and p _ip contiguous point forms p _idistance neighborhood, i=1,2 ... N, N by two width subgraphs total number of extraction angle point;

2) calculate each Corner Feature vector in subgraph subject to registration successively and, to the mahalanobis distance with reference to all Corner Feature vectors in subgraph, mahalanobis distance is less than setting threshold value d _mth1two angle points be defined as matching double points, multiple matching double points forms coupling queue 1;

3) in coupling queue 1, reject not at the matching double points in respective distances field, obtain K matching double points;

Step 5, based on K matching double points, adopt least square method to calculate transformation matrix H between image subject to registration and reference picture, utilize described transformation matrix H by image registration subject to registration on reference picture;

The described process of choosing with reference to subgraph is:

Secondly first become with reference to Iamge Segmentation the subgraph that n size is identical, calculate entropy and the average gradient of each subgraph, then select entropy and the maximum subgraph of average gradient sum as with reference to subgraph.

2. according to claim 1 based on the fast image registration method of subimage Corner Feature, it is characterized in that, described n >=4.

3. according to claim 1 based on the fast image registration method of subimage Corner Feature, it is characterized in that, extract with reference to subgraph identical with the method for subgraph angle point subject to registration, detailed process is:

First, based on Harris angular-point detection method, detect subgraph angle point; Secondly, the method taking neighborhood non-maximum restraining and total amount to suppress is screened the angle point that initial detecting goes out, and extracts top n angle point; Then, remove the angle point be positioned on subgraph borderline region, thus extract required angle point.

4. according to claim 3 based on the fast image registration method of subimage Corner Feature, it is characterized in that, described subgraph borderline region is distance border width is the rectangular area of σ, and wherein σ is the Gaussian smoothing factor.

5. according to claim 1 based on the fast image registration method of subimage Corner Feature, it is characterized in that, described proper vector is 12 dimension gradient vectors of angle point.

6. according to claim 1 based on the fast image registration method of subimage Corner Feature, it is characterized in that, mahalanobis distance d _m(i, j), in the process calculated, utilizes orthogonal matrix P and diagonal matrix D to represent the inverse matrix C of covariance matrix C ^-1, convert the calculating of mahalanobis distance to Euclidean distance d _ecalculating;

$\begin{array}{c}{d}_M(i,j)=\sqrt{{(X_i^1-X_j^2)}^t{C}^{-1}(X_i^1-X_j^2)}\\ =\sqrt{{(X_i^1-X_j^2)}^t{P}^t\sqrt{D}\&CenterDot;\sqrt{D}P(X_i^1-X_j^2)}\\ =d_E(\sqrt{D}{\mathrm{PX}}_i^1,\sqrt{D}{\mathrm{PX}}_j^2)\end{array}$

Wherein, represent the proper vector of i-th angle point on subgraph subject to registration, represent the proper vector with reference to a jth angle point on subgraph.

7. according to claim 1 based on the fast image registration method of subimage Corner Feature, it is characterized in that, the number K=4 of described matching double points.