CN101556692A - Image mosaic method based on neighborhood Zernike pseudo-matrix of characteristic points - Google Patents

Abstract

The invention relates to an image mosaic method based on neighborhood Zernike pseudo-matrix of characteristic points, which leads image statistical information (Zernike pseudo-matrix) to be interrelated with the characteristic points of an image by using the following steps: firstly, extracting interest points of an inference image and an input image by utilizing a Harris angle detector, and taking rectangular neighborhoods using the interest points as a center as a local characteristic region with characteristic matching; secondly, calculating the Zernike pseudo-matrix to the rectangular characteristic regions to be used as descriptors of the characteristic region, and realizing the matching of the characteristic points through comparing the Euclidean distance of the characteristic vector of the descriptor of each characteristic region. The matching can have less wrong matching points which are eliminated through a RANSAC (RANdom Sample Consensus) algorithm, and the right matching relation can be calculated to realize the image mosaic. The invention can effectively realize the image registration and mosaic with geometric transformation relations of translation, rotation, small scale zooming, and the like, and can be used for image treatment and image synthesis of fields such as communication, multimedia technology, and the like.

Description

Image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points

Technical field

The present invention relates to a kind of image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points, this method can realize having translation effectively, rotation, the image registration and the splicing of what transformation relation of youngsters such as convergent-divergent can be used for the image registration and the splicing in a plurality of fields such as virtual reality, Medical Image Processing, remote sensing technology.

Background technology

The image mosaic technology is the hot research direction of computer vision, Flame Image Process and computer graphics always.It can be used for setting up high-definition picture with great visual angle, all is widely used in virtual reality field, field of medical image processing, remote sensing technology field and military field.Image mosaic be exactly with a series of pictures that overlaps mutually at Same Scene be spliced into significantly, wide visual angle, with original image near and distortion little, do not have a tangible sutural high-definition picture.Image mosaic is as one of emphasis of image studies aspect in these years, and the researchist has also proposed a lot of stitching algorithms both at home and abroad.The quality of image mosaic, the registration accuracy of main dependency graph picture, so the registration of image is the core and the key of stitching algorithm.According to the difference of image matching method, generally image split-joint method can be divided into following two types:

(1) based on the relevant joining method in zone

This is the most traditional and general algorithm.Based on the method for registering in zone is gray-scale value from image to be spliced, treating the zone in the stitching image and the zone of the same size in the reference picture uses least square method or other mathematical method to calculate the difference of its gray-scale value, this diversity ratio was judged more afterwards the similarity degree in the doubling of the image to be spliced zone and reference picture zone, obtain the scope and the position in the doubling of the image to be spliced zone thus, thereby realize image mosaic.Also can be by the FFT conversion with image by spatial transform to frequency domain, and then carry out registration.

When with the difference of two regional picture element gray-scale values during as discrimination standard, the simplest a kind of method is directly the difference of each point gray scale to be added up.This method effect is not fine, and usually variation and other reason owing to brightness, contrast causes the splicing failure.Another kind method is to calculate the related coefficient of the corresponding picture element gray-scale value in two zones, and related coefficient is big more, and then the matching degree of two blocks of images is high more.The splicing effect of this method increases, but exists convergent-divergent and rotation the time to tend to because the rotation of matching area and convergent-divergent make the bigger change of content generation of the matching area that the matching area that mates image to be spliced is corresponding with reference picture cause the failure of mating when image.Based on the frequency domain method of FFT because use entire image to mate as the zone, to the overlapping proportion requirement of image than higher.

(2) based on the stitching algorithm of feature

Method for registering based on feature is not a pixel value of directly utilizing image, but the feature by the pixel deduced image, be standard then with the characteristics of image, search matched is carried out in the character pair zone of doubling of the image part, such algorithm has reasonable robustness and robustness.

Method for registering based on feature has two processes: feature extraction and characteristic matching.At first from two width of cloth images, extract features such as the tangible point of grey scale change, line, zone and form characteristic set.In the set of two width of cloth image characteristics of correspondence, utilize the characteristic matching algorithm will exist the feature of corresponding relation as much as possible then to extracting.A main class is an angle point in the feature point extraction algorithm that extensively is used at present, and Corner Detection Algorithm is modal Moravec Corner Detection Algorithm, SUSAN Corner Detection Algorithm and Harris Corner Detection Algorithm.Unique point as herein described is the Harris angle point.

Image registration algorithm advantage based on feature is mainly reflected in three aspects:

● the unique point of image is lacked a lot than the picture element of image, therefore significantly reduced the calculated amount of matching process;

● the matching degree value of unique point is relatively more responsive to the variation of position, can improve the levels of precision of coupling greatly;

● the leaching process of unique point can reduce The noise, to greyscale transformation, anamorphose and block adaptive faculty is preferably all arranged.

Summary of the invention

The objective of the invention is to overcome the deficiency of above-mentioned prior art, propose a kind of image split-joint method, to realize image mosaic with factors such as rotation, translation, convergent-divergent and noise based on neighborhood Zernike pseudo-matrix of characteristic points.

The object of the present invention is achieved like this:

The present invention makes full use of the method for registering images based on feature, its technical scheme is that image statistics information (Zernike pseudo-matrix) is combined with the unique point of image, utilize the Harris angle detector to extract the point of interest of reference picture and image to be spliced, calculate the descriptor of the Zernike pseudo-matrix of point of interest neighborhood window then as this characteristic area, and realize the coupling of unique point by the Euclidean distance of each characteristic area descriptor proper vector relatively, it is right to reject incorrect match point by RANSAC (RANdomSample Consensus) algorithm at last, and calculates correct matching relationship to realize the splicing of image.

Described image split-joint method comprises the steps:

(1) utilize the Harris angular-point detection method to reference picture and image extract minutiae to be spliced, note is made p={p respectively ₁, p ₂, p ₃..., p _mAnd p '=p ' ₁, p ' ₂, p ' ₃..., p ' _n, be that the characteristic area of the zone of rectangle as this unique point got at the center with the unique point;

(2) in all characteristic areas, calculate this regional Zernike pseudo-matrix respectively, and with the proper vector that the generates descriptor as this unique point;

(3) by relatively between the point of interest Euclidean distance of proper vector S can obtain corresponding relation between two width of cloth image points of interest.Promptly for each the point of interest S in the reference picture _p, calculate the Euclidean distance of each point of interest proper vector in the proper vector of this point and the image to be spliced respectively, distance is more little to illustrate that then the proper vector between 2 is similar more, gets the minimum point of distance as with reference to image mid point S _pMatch point in image to be spliced, note is made S ' _p

(4) by above-mentioned the 3rd step, can calculate the match point of each point of interest in image to be spliced in the reference picture, it is right that certain such coupling centering can comprise the pseudo-coupling of part, uses the RANSAC algorithm and estimate the right geometric transformation model T of these match points, and it is right to reject the puppet coupling simultaneously;

$T=\left[\begin{array}{ccc}s\cos \mathrm{\θ}& -s\sin \mathrm{\θ}& t_x\\ s\sin \mathrm{\θ}& s\cos \mathrm{\θ}& t_y\\ 0& 0& 1\end{array}\right]$

Wherein s is the change of scale parameter, and θ is the angle of rotation, t _x, t _yIt is respectively the translational movement on x direction and the y direction.

(5) image to be spliced is transformed on the coordinate system of reference picture according to geometric transformation model T;

x＝Tx′

Wherein x ' is the original coordinates vector of image to be spliced, and T is a geometric transformation, and x is the new coordinate vector after transforming on the reference picture coordinate system

(6) use method of weighted mean that image co-registration is carried out in the overlapping region of two width of cloth images, the overlapping region seamlessly transits in the realization stitching image.

$f(x,y)=\left\{\begin{array}{cc}{f}_1(x,y)& (x,y)\∈f_1\\ d_1{f}_1(x,y)+d_2{f}_2(x,y)& (x,y)\∈(f_1\∩f_2)\\ f_2(x,y)& (x,y)\∈f_2\end{array}\right.$

D wherein ₁, d ₂The expression weighted value, relevant with the width of overlapping region, and d ₁+ d ₂=1,0≤d ₁, d ₂≤ 1.D in the overlapping region ₁By 1 gradual change to 0, d ₂By 0 gradual change to 1, be implemented in thus in the overlapping region by f ₁Be smoothly transitted into f ₂f ₁, f ₂Be two images to be spliced, f is the image after merging.

Above-mentioned image split-joint method, described in (1) is that the characteristic area of the zone of rectangle as this unique point got at the center with each unique point, carries out as follows:

1) utilizes the Harris angular-point detection method to extract the unique point of original image, be designated as unique point set p and p ';

2) be the center with each unique point among unique point set p and the p ' respectively, getting the length of side is the characteristic area of the rectangular area of L as this unique point;

Above-mentioned image split-joint method, (2) described in all rectangular areas, when calculating this regional Zernike pseudo-matrix respectively, the point of interest that selects image is polar initial point, with the pixel mapping in the unit circle is polar coordinates, and the outer pixel of unit circle is not considered when calculating.And precision and speed in order to improve coupling, need be optimized selection to the Zernike pseudo-matrix that calculates, undertaken by following principle:

1) exponent number when square is higher than some value N _MaxThe time, calculated amount is big and calculating is no longer accurate, gets 10 rank in the experiment.

2) multiplicity be m=4i (i=0,1,2 ...) and square to calculate be inaccurate, should remove.And since square grip symmetry altogether, it is independently that remaining square has only half.The square set that note is chosen is S (being proper vector), S={A so _Nm, n≤N _Max, m 〉=0, m ≠ 4i}.

Above-mentioned image split-joint method when the Euclidean distance with proper vector described in (3) is weighed the similarity degree of two proper vectors, is calculated as follows:

$\mathrm{Dis}\tan \mathrm{ce}=\sqrt{\underset{i=1}{\overset{N_{\max }}{\mathrm{\Σ}}}{(Z_i-Z_i^{\′})}^2}$

Z wherein _i(i=1,2,3 ..., N _Max), Z ' _i(=1,2,3 ..., N _Max) represent the proper vector of two unique points respectively, in order to increase the robustness of coupling, certain threshold value need be set, and the point of interest in two proper vector minor increments are thought reference picture during all greater than this threshold value can not find the point with his coupling in image to be spliced, should reject this point.

Above-mentioned image split-joint method, the geometric transformation model T described in (4), the affine matrix that adopts 4 parameters here is as the transformation model between the image.

$T=\left[\begin{array}{ccc}s\cos \mathrm{\θ}& -s\sin \mathrm{\θ}& t_x\\ s\sin \mathrm{\θ}& s\cos \mathrm{\θ}& t_y\\ 0& 0& 1\end{array}\right]$

The present invention has following effect:

(1) owing to adopt the local feature zone of image to carry out characteristic matching, can reduce the calculated amount of coupling.

(2) because the matching degree value of unique point is relatively more responsive to the variation of position, can improve the levels of precision of coupling greatly.

(3) because the leaching process of unique point can reduce The noise, to greyscale transformation, anamorphose and block adaptive faculty is preferably all arranged.

(4) owing to adopt in the local feature zone of image with the descriptor of Zernike pseudo-matrix as this unique point, and when calculating this regional Zernike pseudo-matrix, the point of interest that selects image is polar initial point, with the pixel mapping in the unit circle is polar coordinates, and the outer pixel of unit circle is not considered when calculating.Overcome like this because the variation of the caused rectangular characteristic area contents of rotation, change of scale of image has improved the precision of mating.

(5) owing to adopt the descriptor of Zernike pseudo-matrix as unique point, the amplitude of Zernike pseudo-matrix has the character of invariable rotary, makes this descriptor have the character of invariable rotary.No matter the rotation relationship how many angles reference picture and image to be spliced have, descriptor still has very high accuracy, has improved the precision of coupling.

(6) owing to adopt the descriptor of Zernike pseudo-matrix as unique point, Zernike pseudo-matrix has good noise robustness, makes this descriptor have good robustness for noise, has improved the noise robustness of coupling.

Description of drawings

Fig. 1 is the techniqueflow block diagram of algorithm of the present invention.

Fig. 2 is a trial image, and wherein 2 (a) are reference picture, and 2 (b) are image to be spliced.

Fig. 3 is for extracting the point of interest image, and wherein 3 (a) are the point of interest extraction of reference picture, and 3 (b) are that the point of interest of image to be spliced extracts.

The match point that Fig. 4 obtains for characteristic matching is to image.

Fig. 5 rejects the correct match point image that pseudo-coupling is kept the back for using the RANSAC method.

Image behind Fig. 6 reference picture and the image mosaic to be spliced.

Fig. 7 is the stitching image through seamlessly transitting.

Embodiment

It is following that the present invention is described in further detail with reference to accompanying drawing.

Image split-joint method step of the present invention is as follows:

The first step: extract point of interest.

To reference picture Fig. 2 (a) and image graph to be spliced 2 (b), utilize Harris angular-point detection method extract minutiae, the unique point of initial extraction is shown in Fig. 3 (a) and 3 (b).To piece image I (x, y), the extraction step of Harris unique point is specific as follows:

(1) utilize following formula compute gradient image:

$\left\{\begin{array}{c}X=I\⊗(-\mathrm{1,0,1})=\∂I/\∂X\\ Y=I\⊗{(-\mathrm{1,0,1})}^T=\∂I/\∂Y\end{array}\right.$

Wherein

The expression convolution, X represents the gradient image of horizontal direction, Y represents the gradient image of vertical direction.

(2) structure autocorrelation matrix:

Order $\left\{\begin{array}{c}A=X^2\⊗w\\ B=Y^2\⊗w\\ C=\left(\mathrm{XY}\right)\⊗w\end{array}\right.$

W=exp ((X wherein ²+ y ²)/2 σ ²) be Gauss's smoothing windows function.

Autocorrelation matrix then $M=\left[\begin{array}{cc}A& C\\ C& B\end{array}\right].$

(3) extract minutiae:

Order $\left\{\begin{array}{c}{D}_{\mathrm{et}}\left(M\right)=\mathrm{AB}-C^2\\ T_{\mathrm{race}}\left(M\right)=A+B\end{array}\right.$

Then Harris unique point response is

R _H＝D _et(M)-k·T _race ²(M)

Wherein, constant k is got between the 0.04-0.06 usually.With R _HCompare with a threshold value, then regarding as greater than this threshold value is unique point, and this threshold value is according to the unique point number setting that will detect, generally more than or equal to 1000.

Second step: the Zernike pseudo-matrix in calculated characteristics zone, and as this unique point descriptor

1) calculating of Zernike pseudo-matrix

The Zernike pseudo-matrix of image obtains image mapped to one group of basis function, be called the base of Zernike pseudo-matrix, is designated as { V _Nm(x, y) }.This group base has constituted unit circle (x ²+ y ²≤ 1) Nei one group of complete orthogonal set, it is defined as follows:

V _nm(x，y)＝V _nm(ρ，θ)＝R _nm(ρ)exp(jmθ)

Wherein n is a nonnegative integer, and m is an integer, and both satisfy: | m|≤n.ρ, θ are respectively the radius and the angle of pixel under the polar coordinates.R _NmBe radial polynomial (Radial polynomial) (ρ), be defined as follows:

$R_{\mathrm{nm}}\left(\mathrm{\ρ}\right)=\underset{s=0}{\overset{n-\left|m\right|}{\mathrm{\Σ}}}\frac{{(-1)}^s(2n+1-s)!{\mathrm{\ρ}}^{n-s}}{s!(n+|m|+1-s)!(n-|m|-s)!}$

These polynomial expressions are mutually orthogonal, satisfy following relation:

$\underset{x^2+y^2\≤1}{\∫\∫}{V}_{\mathrm{nm}}^{*}(x,y)V_{\mathrm{pq}}(x,y)\mathrm{dxdy}=\frac{\mathrm{\π}}{n+1}{\mathrm{\δ}}_{\mathrm{np}}{\mathrm{\δ}}_{\mathrm{mq}}$

Wherein, ${\mathrm{\δ}}_{\mathrm{st}}=\left\{\begin{array}{cc}1& s=t\\ 0& \mathrm{otherwise}\end{array}\right.$

Consecutive image f (x, y), (n, m) rank Zernike pseudo-matrix definition:

$A_{\mathrm{nm}}=\frac{n+1}{\mathrm{\π}}\underset{x^2+y^2\≤1}{\∫\∫}f(x,y)V_{\mathrm{nm}}^{*}(x,y)\mathrm{dxdy}$

For digital picture, under polar coordinates, following formula becomes:

$A_{\mathrm{nm}}=\frac{n+1}{\mathrm{\π}}\underset{\mathrm{\ρ}\≤1}{\mathrm{\Σ}}\underset{0\≤\mathrm{\θ}\≤2\mathrm{\π}}{\mathrm{\Σ}}f(\mathrm{\ρ},\mathrm{\θ})V_{\mathrm{nm}}^{*}(\mathrm{\ρ},\mathrm{\θ})\mathrm{\ρ}$

$=\frac{n+1}{\mathrm{\π}}\underset{\mathrm{\ρ}\≤1}{\mathrm{\Σ}}\underset{0\≤\mathrm{\θ}\≤2\mathrm{\π}}{\mathrm{\Σ}}f(\mathrm{\ρ},\mathrm{\θ})R_{\mathrm{nm}}\left(\mathrm{\ρ}\right)\exp (-\mathrm{jm\θ})\mathrm{\ρ}$

During the Zernike pseudo-matrix of computed image, the point of interest that selects image is polar initial point, is polar coordinates with the pixel mapping in the unit circle, and the outer pixel of unit circle is not considered when calculating.

2) character of Zernike pseudo-matrix

The amplitude of Zernike pseudo-matrix has the character of invariable rotary.Be located under the polar coordinates original image and be f (ρ, θ), the image f after its rotation alpha angle then ^r(ρ, Zernike pseudo-matrix A θ) _Nm ^rZernike pseudo-matrix A with original image _NmThe pass is:

$A_{\mathrm{nm}}^r=A_{\mathrm{nm}}\exp (-\mathrm{jm\α})$

So, $\left|A_{\mathrm{nm}}^r\right|=\left|A_{\mathrm{nm}}\right|.$ Therefore, the amplitude of Zernike pseudo-matrix has rotational invariance.

3) selection of Zernike pseudo-matrix

The pixel discrete nature of digital picture makes the calculating of Zernike pseudo-matrix produce error, not the counting accuracy difference of same order square.Therefore, must be optimized selection to square.Mainly consider 2 points when selecting square: a) exponent number when square is higher than some value N _MaxThe time, calculated amount is big and calculating is no longer accurate, gets 10 rank in the experiment; B) multiplicity be m=4i (i=0,1,2 ...) and square to calculate be inaccurate, should remove.And since square grip symmetry altogether, it is independently that remaining square has only half.The square set that note is chosen is S, i.e. S={A _Nm, n≤N _Max, m 〉=0, m ≠ 4i}.

The 3rd step: it is right to extract initial characteristics point coupling.

By relatively between the point of interest Euclidean distance of proper vector S can obtain corresponding relation between two width of cloth image points of interest.Promptly for each the point of interest S in the reference picture _p, calculate the Euclidean distance of each point of interest proper vector in the proper vector of this point and the image to be spliced respectively, distance is more little to illustrate that then the proper vector between 2 is similar more, gets the minimum point of distance as with reference to image mid point S _pMatch point in image to be spliced, note is made S ' _pReferring to Fig. 4.

When weighing the similarity degree of two proper vectors, be calculated as follows:

$\mathrm{Dis}\tan \mathrm{ce}=\sqrt{\underset{i=1}{\overset{N_{\max }}{\mathrm{\Σ}}}{(Z_i-Z_i^{\′})}^2}$

Z wherein _i(i=1,2,3 ..., N _Max), Z ' _i(i=1,2,3 ..., N _Max) represent that respectively the proper vector of two unique points is in order to increase the robustness of coupling, certain threshold value need be set, point of interest in two proper vector minor increments are thought reference picture during all greater than this threshold value can not find the point with his coupling in image to be spliced, should reject this point.

The 4th step: the estimation of image geometry transformation model.

The inevitable wrong match point of the match point centering that obtains from previous step is rapid is right, they are called as " exterior point " (point of green among Fig. 5), corresponding, " interior point " (cross of redness among Fig. 5) is real right corresponding to the correct Feature Points Matching of same point in the scene.The existence of exterior point is asked for very big negative effect to the image transformation matrix parameter, needs to reject exterior point (referring to Fig. 5).

RANSAC (Random Sample Consensus) is that most widely used exterior point is rejected algorithm, this algorithm is by constantly extracting fixed sample point computation model all unique point centerings, statistics meets the interior point of model, the model of point is the image transformation model in obtaining at most, reject exterior point simultaneously, kept interior point.When the extraction number of times is abundant, can guarantee the accuracy of model and interior point with big probability.Roughly step is as follows:

1) select N match point to as sample at random whole M match point centerings;

2) by the parameter x of N match point to estimated image transformation matrix T;

3) calculate sample size K in the subclass that whole M match point centerings meet transformation parameter x;

4) if K is enough big, then finish to calculate, parameter current x is the transformation matrix parameter;

5) if K does not satisfy condition, repeating step 1)～4), repeat altogether L this;

The 5th step: image transformation.

Image to be spliced is transformed on the coordinate system of reference picture according to geometric transformation model T.

x＝Tx′

Wherein x ' is the original coordinates vector of image to be spliced, and T is a geometric transformation, and x is the new coordinate vector (referring to Fig. 6, can see that tangible seam is arranged on the stitching image) after transforming on the reference picture coordinate system.

The 6th step: image co-registration.

Use method of weighted mean that image co-registration is carried out in the overlapping region of two width of cloth images, the overlapping region seamlessly transits in the realization stitching image.

$f(x,y)=\left\{\begin{array}{cc}{f}_1(x,y)& (x,y)\∈f_1\\ d_1{f}_1(x,y)+d_2{f}_2(x,y)& (x,y)\∈(f_1\∩f_2)\\ f_2(x,y)& (x,y)\∈f_2\end{array}\right.$

D wherein ₁, d ₂The expression weighted value, relevant with the width of overlapping region, and d ₁+ d ₂=1,0≤d ₁, d ₂≤ 1.D in the overlapping region ₁By 1 gradual change to 0, d ₂By 0 gradual change to 1, be implemented in thus in the overlapping region by f ₁Be smoothly transitted into f ₂f ₁, f ₂Be two images to be spliced, f is the image (referring to Fig. 7, seam is eliminated) after merging.

Claims (5)

1. the image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points comprises the steps:

(1) utilize the Harris angular-point detection method to reference picture and image extract minutiae to be spliced, note is made p={p respectively ₁, p ₂, p ₃..., p _mAnd p '=p ' ₁, p ' ₂, p ' ₃..., p ' _n, be that the characteristic area of the zone of rectangle as this unique point got at the center with the unique point;

(2) in all characteristic areas, calculate this regional Zernike pseudo-matrix respectively, and with the proper vector that the generates descriptor as this unique point;

(3) by relatively between the point of interest Euclidean distance of proper vector S can obtain corresponding relation between two width of cloth image points of interest.Promptly for each the point of interest S in the reference picture _p, calculate the Euclidean distance of each point of interest proper vector in the proper vector of this point and the image to be spliced respectively, distance is more little to illustrate that then the proper vector between 2 is similar more, gets the minimum point of distance as with reference to image mid point S _pMatch point in image to be spliced, note is made S ' _p

(4) by above-mentioned the 3rd step, can calculate the match point of each point of interest in image to be spliced in the reference picture, it is right that certain such coupling centering can comprise the pseudo-coupling of part, uses the RANSAC algorithm and estimate the right geometric transformation model T of these match points, and it is right to reject the puppet coupling simultaneously;

$T=\left[\begin{array}{ccc}s\cos \mathrm{\θ}& -s\sin \mathrm{\θ}& t_x\\ s\sin \mathrm{\θ}& s\cos \mathrm{\θ}& t_y\\ 0& 0& 1\end{array}\right]$

Wherein s is the change of scale parameter, and θ is the angle of rotation, t _x, t _yIt is respectively the translational movement on x direction and the y direction.

(5) image to be spliced is transformed on the coordinate system of reference picture according to geometric transformation model T;

x＝Tx′

Wherein x ' is the original coordinates vector of image to be spliced, and T is a geometric transformation, and x is the new coordinate vector after transforming on the reference picture coordinate system

(6) use method of weighted mean that image co-registration is carried out in the overlapping region of two width of cloth images, realize seamlessly transitting of overlapping region in the stitching image,

$f(x,y)=\left\{\begin{array}{cc}{f}_1(x,y)& (x,y)\∈f_1\\ d_1{f}_1(x,y)+d_2{f}_2(x,y)& (x,y)\∈(f_1\∩f_2)\\ f_2(x,y)& (x,y)\∈f_2\end{array}\right.$

D wherein ₁, d ₂The expression weighted value, relevant with the width of overlapping region, and d ₁+ d ₂=1,0≤d ₁, d ₂≤ 1.D in the overlapping region ₁By 1 gradual change to 0, d ₂By 0 gradual change to 1, be implemented in thus in the overlapping region by f ₁Be smoothly transitted into f ₂, f ₁, f ₂Be two images to be spliced, f is the image after merging.d ₁And d ₂Be calculated as follows: the horizontal ordinate of supposing current pixel is x _i, the horizontal ordinate on border, the left and right sides, overlapping region is respectively x _lAnd x _r, so

$d_1=\frac{x_r-x_i}{x_r-x_l},$ $d_2=1-d_1=\frac{x_i-x_l}{x_r-x_l}.$

2. the image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points according to claim 1 is characterized in that described in 1. (1) being that the characteristic area of the zone of rectangle as this unique point got at the center with each unique point, carries out as follows:

(a) utilize the Harris angular-point detection method to extract the unique point of original image, be designated as unique point set p and p ';

(b) be the center with each unique point among unique point set p and the p ' respectively, getting the length of side is the characteristic area of the characteristic area of L as this unique point.

3. the image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points according to claim 1, it is characterized in that described in 1. (2) in all characteristic areas, when calculating this regional Zernike pseudo-matrix respectively, the point of interest that selects image is polar initial point, with the pixel mapping in the unit circle is polar coordinates, and the outer pixel of unit circle is not considered when calculating.And precision and speed in order to improve coupling, need be optimized selection to the Zernike pseudo-matrix that calculates, undertaken by following principle:

(a) exponent number when square is higher than some value N _MaxThe time, calculated amount is big and calculating is no longer accurate, gets 10 rank in the experiment.

(b) multiplicity be 4 integral multiple (be m=4i (i=0,1,2 ...)) square to calculate be inaccurate, should remove.And since square grip symmetry altogether, it is independently that remaining square has only half.The square set that note is chosen is S (being proper vector), S={A so _Nm, n≤N _Max, m 〉=0, m ≠ 4i}.

4. the image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points according to claim 1 when it is characterized in that the Euclidean distance with proper vector described in 1. (3) is weighed the similarity degree of two proper vectors, is calculated as follows:

$\mathrm{Dis}\tan \mathrm{ce}=\sqrt{\underset{i=1}{\overset{N_{\max }}{\mathrm{\Σ}}}{(Z_i-Z_i^{\′})}^2}$

Z wherein _i(i=1,2,3 ..., N _Max), Z ' _i(i=1,2,3 ..., N _Max) represent the proper vector of two unique points respectively, in order to increase the robustness of coupling, certain threshold value need be set, and the point of interest in two proper vector minor increments are thought reference picture during all greater than this threshold value can not find the point with his coupling in image to be spliced, should reject this point.

5, the image split-joint method based on neighborhood Zernike pseudo-matrix of characteristic points according to claim 1 is characterized in that the geometric transformation model T described in 1. (4), and the affine matrix that adopts 4 parameters here is as the transformation model between the image

$T=\left[\begin{array}{ccc}s\cos \mathrm{\θ}& -s\sin \mathrm{\θ}& t_x\\ s\sin \mathrm{\θ}& s\cos \mathrm{\θ}& t_y\\ 0& 0& 1\end{array}\right].$

Images