CN103279946A - Globally optimized image registration method and system - Google Patents

Abstract

The invention discloses a globally optimized image registration method. The method comprises the steps of reading a deformed image, and calculating the highest frequency of an objective function throughout deformation coefficients; carrying out non-uniform sampling on the deformation coefficients to obtain a deformed image sequence with a sampling frequency which is higher than or equal to the highest frequency multiplied by two; reading a reference image, deforming the deformed image to finish coarse registration through a deformed function corresponding to a minimal objective function value; taking a deformation coefficient value of the image after the coarse rectification as a starting point, searching for a global minimum point of the objective function, and finishing the registration of the image. The invention also discloses a globally optimized image registration system. Due to the adoption of the sampling frequency which is higher than or equal to the highest frequency multiplied by two, the sampling frequency can be regulated according to the highest frequency, a high sampling frequency is adopted in a high frequency region, and the sampling frequency can be relatively reduced in a low frequency region, so the global minimum point of the objective function can be obtained by utilizing fewer sampling points, computational effort can be reduced, and registration time can be shortened.

Description

A kind of method for registering images of global optimization and system

Technical field

The application relates to the graph image conversion method in the plane of delineation, relates in particular to a kind of image elastic registrating method and system of global optimization.

Background technology

Image registration refers to seek a kind of spatial alternation for piece image, makes corresponding point on it and another width of cloth image reach coupling on the space.This coupling refer to content identical o'clock at two matching images identical locus is arranged.

Method for registering images commonly used mainly may further comprise the steps:

(1) constructs the warping function that image is out of shape.Existing method for registering can be divided into two kinds of Rigid Registration method and elastic registrating methods.The warping function that uses in the Rigid Registration method is a kind of rigid deformation function, and it can only carry out simple relatively conversion such as translation, rotation to figure.If pattern distortion is serious, just need to introduce the elastic deformation function during registration, corresponding method for registering is elastic registrating method just.The elastic deformation function has enough versatilities, can approach nonlinear transformation arbitrarily.The elastic deformation function can be divided into two classes substantially: a class is non-parametric model.In this model, image is regarded as the resilient film of a slice, foretells distortion in the effect of external force and internal force, and finally reaches balance.External force is determined that by the difference of reference picture and deformation pattern internal force is determined by the intensity peace slippage degree of film.The most famous in this model is fluid model, also has diffusion model, light stream model etc. in addition.Another kind of is parameterized model, and model need use some parameters and basis function to represent, the computation process of warping function is the CALCULATION OF PARAMETERS process just.The kind of basis function is a lot, and common basis function comprises that polynomial expression, harmonic function, classification basis function, small echo are with B batten etc.

(2) structure needs the similarity criterion of reference picture and the deformation pattern of registration.The similarity criterion is also referred to as objective function sometimes, and it is to weigh similarity between reference picture and the deformation pattern.Objective function commonly used has mutual information, mean square deviation etc.

(3) objective function is optimized calculating, reaches its extreme value up to objective function.This process constantly is out of shape with warping function deformation pattern exactly, and when objective function reached extreme value, when namely the similarity between the deformation pattern after reference picture and the distortion was maximum, registration process had also just been finished.Mostly optimization method commonly used in the above-mentioned registration process is some local optimization methods, as Newton method, half Newton method, common rumble vector method etc.The shortcoming of this method is the global extremum that can not guarantee to find objective function, thereby can not guarantee the correctness of registration results.Optimization method also can use global optimization approach, and corresponding global optimization method has simulated annealing, genetic method, tunneling method, method of exhaustion, method such as a lot of somes methods at random at random.Wherein, simulated annealing and genetic method are oversize computing time, and can not guarantee to obtain at last overall situation solution; The shortcoming of tunneling method is exactly that its optimization object has certain limitation, versatility is bad: and various method based on limit does not at random have definite calculation methods yet on the quantity Calculation of random point, obtaining the accurate overall situation separates, has only the quantity that strengthens random point, this just causes computing time long, can not guarantee the correctness of last registration results simultaneously.

At present, various methods based on limit at random do not have definite calculation methods in the quantity Calculation of sampled point, obtain the accurate overall situation and separate, and have only the quantity that strengthens random point, but after sampled point quantity increased to a certain degree, the quantity that increases sampled point again may be exactly a kind of waste.How to determine the present also suitable algorithm useless of this sampled point quantity.If can determine the lower bound of sampled point quantity, just can in a limited time, obtain correct registration results.This quantity lower bound can obtain according to nyquist sampling theorem.The method of available technology adopting uniform sampling, consistent with high frequency treatment in its sampling density of frequency lower, the sampling work amount is big, and the registration time is long, need to be improved and enhanced.

Summary of the invention

The technical matters that the application will solve is at the deficiencies in the prior art, and the method for registering images of short global optimization of a kind of registration time is provided.

Another technical matters that the application will solve is based on the figure registration system that said method provides a kind of global optimization.

The technical matters that the present application will solve is solved by the following technical programs:

A kind of method for registering images of global optimization comprises:

Read in deformation pattern, calculating target function is at deformation coefficient highest frequency everywhere;

Be that sample frequency is carried out nonuniform sampling to deformation coefficient to be greater than or equal to 2 times of described highest frequency, obtain the deformation pattern sequence;

Read in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, finish thick registration;

Deformation coefficient value to described thick images after registration is starting point, and the global minimum point of ferret out function is finished the registration of image.

Described sample frequency is 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of described deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

The wherein said deformation pattern that reads in, calculating target function comprises at deformation coefficient highest frequency everywhere: read in deformation pattern, use Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

Wherein said deformation coefficient value to described thick images after registration is starting point, the global minimum point of ferret out function, the registration of finishing image comprises: the deformation coefficient value to described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finish the registration of image.

A kind of figure registration system of global optimization comprises and reads in module, sampling module, thick registration module and registration module that the described module of reading in is used for reading in deformation pattern, and calculating target function is at deformation coefficient highest frequency everywhere; Described sampling module is used for sampling and obtains the deformation pattern sequence; Described thick registration module is used for reading in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, and finishes thick registration; The deformation coefficient value that described registration module is used for described thick images after registration is starting point, the global minimum point of ferret out function, finish the registration of image, it is that sample frequency is carried out nonuniform sampling to deformation coefficient that described sampling module also is used for to be greater than or equal to 2 times of described highest frequency, obtains the deformation pattern sequence.

Described sample frequency is 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of described deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

The described module of reading in also is used for reading in deformation pattern, uses Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

The deformation coefficient value that described registration module also is used for described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finishes the registration of image.

Owing to adopted above technical scheme, the beneficial effect that the present invention is possessed is:

In the application's embodiment, owing to adopt the sample frequency more than or equal to 2 times of highest frequencies, both guaranteed the complete recovery of sampled point, have and to adjust sample frequency according to highest frequency, make high-frequency region adopt high sample frequency, and the corresponding reduction sample frequency of low frequency region, so just can utilize still less sampled point to obtain the global minimum point of objective function, thereby reduced amount of calculation, make the registration time shorten.

Description of drawings

Fig. 1 is the process flow diagram of an embodiment of method for registering images of the application's global optimization;

Fig. 2 is the process flow diagram of another embodiment of method for registering images of the application's global optimization;

Fig. 3 is the structural representation of an embodiment of figure registration system of the application's global optimization.

Embodiment

By reference to the accompanying drawings the present invention is described in further detail below by embodiment.

Fig. 1 illustrates the process flow diagram according to an embodiment of method for registering images of the application's global optimization, comprising:

Step 102: read in deformation pattern, calculating target function is at deformation coefficient highest frequency everywhere;

Step 104: be that sample frequency is carried out nonuniform sampling to deformation coefficient to be greater than or equal to 2 times of described highest frequency, obtain the deformation pattern sequence;

Step 106: read in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, finish thick registration;

Step 108: the deformation coefficient value to described thick images after registration is starting point, and the global minimum point of ferret out function is finished the registration of image.

A kind of embodiment, sample frequency are 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

A kind of embodiment, step 102 comprises: read in deformation pattern, use Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

A kind of embodiment, step 108 comprises: the deformation coefficient value to described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finishes the registration of image.

Fig. 2 illustrates the process flow diagram according to another embodiment of the application's method, comprising:

Step 202: read in deformation pattern, carry out interpolation with the B batten, calculate its continuous wavelet transform with DOG small echo or Marr small echo, determine image highest frequency value everywhere.

Step 204: use DOG small echo or Marr small echo calculating target function E (θ) for the wavelet transformation of parameter θ, calculate objective function for parameter θ highest frequency throughout by this wavelet transformation again.This highest frequency is

With frequency

To being that (x θ) carries out nonuniform sampling, asks the maximal value of sampling back similarity criterion, and at this moment Dui Ying θ value is exactly to ask for the warping function g of variable with θ.

Step 206: 2 of the highest frequency of 204 objective functions that obtain times set by step

Nonuniform sampling is carried out in the space of deformation coefficient θ correspondence calculate, (x θ), is out of shape deformation pattern respectively with resulting these warping functions again, obtains deformation pattern series to obtain the corresponding warping function g of a series of different deformation coefficient θ;

Step 208: read in reference picture, calculate the mean square deviation of the deformation pattern series that reference picture and step 206 obtain respectively, obtain objective function Value Data group E (θ), relatively the size of these target function values finds wherein minimum target function value E (θ ^*), with this minimum target functional value corresponding warping function g (x, θ ^*) deformation pattern is out of shape, obtain the deformation pattern of thick registration.

Step 210: with the target function value E (θ of the minimum that step 208 was found ^*) corresponding deformation coefficient value is starting point, with the minimum value E of gradient descent method ferret out function E (θ), (x θ) is out of shape deformation pattern, just obtains the deformation pattern of smart registration to use the corresponding warping function g of this minimum value again.

According to another embodiment of the application's method, the test platform of its registration is PC, is operating system with WindowXP, and Matlab6.5 is programming language.Concrete registration process is as described below:

Global optimization approach is divided two parts in the present embodiment: first is the approximate location of determining similarity criterion global extremum with the method for nonuniform sampling; Second portion is to find out similarity criterion global extremum point with existing Local Optimization Algorithm.

Sample to the similarity criterion, an important condition is exactly to know the highest frequency of similarity criterion.This can be determined by the frequency spectrum that calculates the similarity criterion.If this similarity criterion is got SSD, it the highest as follows by frequency computation part:

If f _t(x) be trial image, f _r(x) be reference picture, (x θ) is warping function to g, and wherein θ is the parameter of asking in generation, and then the similarity criterion is

$E=\frac{1}{\left|\right|I\left|\right|}\underset{x\∈I}{\mathrm{\Σ}}{e}_i^2=\frac{1}{\left|\right|I\left|\right|}\underset{x\∈I}{\mathrm{\Σ}}{(f_t\left(g(x,\mathrm{\θ})\right)-f_r\left(x\right))}^2$

So that similarity criterion E obtains the just corresponding best distortion of warping function g of extreme value, namely ask g=argmin _{G ∈ G}E (g), wherein G makes the corresponding space of warping function g.Because be the frequency spectrum that will calculate similarity criterion E, for representing conveniently to omit the coefficient entry of E.

If be basis function with DOG small echo or Marr small echo, similarity criterion E for the wavelet transformation of parameter θ is

${\mathrm{WT}}_{E\left(\mathrm{\&theta;}\right)}(a,\mathrm{\&tau;})=<E\left(\mathrm{\&theta;}\right),{\mathrm{\&psi;}}_{a,\mathrm{\&tau;}}\left(t\right)>=\frac{1}{\sqrt{a}}{\&Integral;}_{R}E\left(\mathrm{\&theta;}\right)\&CenterDot;\mathrm{\&psi;}\left(\frac{t-\mathrm{\&tau;}}{a}\right)\mathrm{dt}$

$=\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}{\&Integral;}_R\underset{x\&Element;I}{\mathrm{\&Sigma;}}{(f_r\left(x\right)-f_t\left(g(x,\mathrm{\&theta;})\right))}^2\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;}$

$=\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_r^2\left(x\right)\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

$-\frac{2}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_r\left(x\right)f_t\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

$+\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_t^2\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

Following formula has three, wherein first

$\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_r^2\left(x\right)\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

$=\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}\left(f_r^2\left(x\right)\right)\&CenterDot;{\&Integral;}_R\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;}$

$=0$

Second is

$\frac{2}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_r\left(x\right)f_t\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

$=\frac{2}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}(f_r\left(x\right)\&CenterDot;{\&Integral;}_R{f}_t\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

The maximal value of following formula frequency just

Maximal value.Calculate now ${\&Integral;}_R{f}_t\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;}$ Maximal value.

(x θ) in arbitrfary point θ=m place Taylor expansion, gets g

$g(x,\mathrm{\&theta;})=g(x,m)+g^{\&prime;}(x,m)\&CenterDot;(\mathrm{\&theta;}-m)+\frac{g^{\&prime;\&prime;}(x,m)}{2!}{(\mathrm{\&theta;}-m)}^2+\&CenterDot;\&CenterDot;\&CenterDot;$

Use first approximation,

g(x,θ)＝g(x,m)+g′(x,m)·(θ-m)

＝(g(x,m)-g′(x,m)·m)+g′(x,m)·θ

If f _t(θ) highest frequency throughout is respectively ω _Max(θ), f then _t(g ' (x, a) highest frequency throughout of θ+b) is ω _Max(θ) g ' (x, m).Because m is the arbitrfary point, so

Highest frequency throughout is ω _Max(θ) max (g ' (x, m))

(x, definition θ) can know that (x, θ) parameter in is C to g, so (x θ) is g ' by g

$\left\{\begin{array}{c}\frac{{\&PartialD;g}_x(x,y)}{{\&PartialD;C}_{i,j}}={\mathrm{\&beta;}}_n(x/2^w-i){\mathrm{\&beta;}}_n(y/2^w-j)\\ \frac{{\&PartialD;g}_y(x,y)}{{\&PartialD;C}_{i,j}}={\mathrm{\&beta;}}_n(x/2^w-p){\mathrm{\&beta;}}_n(y/2^w-q)\end{array}\right.$

β _n(x/2 ^w-maximal value i) is relevant with the size of n, gets n=3 herein.When n=3, β _n(x/2 ^w-i) maximal value is 2/3, so max (g ' (x, a))=4/9.Can obtain at last

Highest frequency be

The 3rd is

$\frac{1}{\sqrt{a}\&CenterDot;\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}({\&Integral;}_R{f}_t^2\left(g(x,\mathrm{\&theta;})\right)\&CenterDot;\mathrm{\&psi;}\left(\frac{\mathrm{\&theta;}-\mathrm{\&tau;}}{a}\right)\mathrm{d\&theta;})$

Because do not need to ask the frequency spectrum of following formula, just need ask its highest frequency, and f _t(highest frequency of g (x, θ)) is So,

Highest frequency be

Above three maximum frequency, the highest frequency of cost cost function is as can be seen

With frequency

To being that (x θ) carries out nonuniform sampling, asks the maximal value of sampling back similarity criterion, and at this moment Dui Ying θ value is exactly to ask for the warping function g of variable with θ.This process can be written as following formula

I wherein _θBe

Be the sample space of sample frequency after to the θ nonuniform sampling.

The approximate location of global extremum can be determined with the method for sampling in the front, can be starting point with this point just at this moment, comes accurate Calculation global extremum point with Local Optimization Algorithm.What Local Optimization Algorithm adopted is the gradient descent method, and concrete computation process is as follows:

The quadratic sum of the error of same service test image and reference picture is as cost function, namely

$E=\frac{1}{\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}{e}_i^2=\frac{1}{\left|\right|I\left|\right|}\underset{x\&Element;I}{\mathrm{\&Sigma;}}{(f_t\left(g(x,\mathrm{\&theta;})\right)-f_r\left(x\right))}^2$

Select the gradient descent method as optimizing algorithm, concrete registration process is such: in the i step of iterative process, according to existing parameter c _iCalculate the step delta c that upgrades _i=-μ _i▽ _cE (c _i), if be c=c in parameter _i+ Δ c _iSituation under, the value of E is littler, this step upgrades successfully so, the parameter value before replacing with new parameter value, i.e. c _I+1=c _i+ Δ c _i, μ _I+1=μ _iIf the value of E has increased on the contrary, just get μ _I+1=μ _i/ 2, compute gradient value Δ c again _i=-μ _I+1▽ _cE (c _i); This iterative process continues always, until a certain threshold value of Δ c less than prior setting.

In this optimizing process, need to calculate the partial derivative vector ▽ of E _cE (c _i).The single order partial derivative of E is

$\frac{\&PartialD;E}{{\&PartialD;c}_{j,m}}=\frac{1}{\left|\right|I\left|\right|}\underset{i\&Element;I}{\mathrm{\&Sigma;}}\frac{{\&PartialD;e}_i}{{\&PartialD;f}_w\left(i\right)}\frac{{\&PartialD;f}_t^c\left(x\right)}{{\&PartialD;x}_m}|x=g\left(i\right)\frac{{\&PartialD;g}_m\left(i\right)}{{\&PartialD;c}_{j.m}}$

Wherein

$\frac{{\&PartialD;e}_i}{{\&PartialD;f}_w\left(i\right)}=2{(f_w\left(i\right)-f_r\left(i\right))}^2$

$\frac{{\&PartialD;g}_m}{{\&PartialD;c}_{j,m}}={\mathrm{\&beta;}}_{n_m}(x/h-j)$

$\frac{\&PartialD;f_t^c}{{\&PartialD;x}_m}\left(x\right)=\underset{k\&Element;I}{\mathrm{\&Sigma;}}{b}_k{\mathrm{\&beta;}}_n^{\&prime;}(x_m-k_m)\underset{l=1,l\&NotEqual;m}{\overset{N}{\mathrm{\&Pi;}}}{\mathrm{\&beta;}}_n(x_l-k_l)$

Because be out of shape the test effect that can better check this paper algorithm under the known condition, this paper adopts artificial deformation's method to produce deformation pattern.The size of original image is 256 * 256 pixels.The content of distortion comprises along X and Y-axis translation 10 each pixel, the distortion elastic deformation that turns clockwise 10 degree and realize with photoshop respectively.From this laboratory picture library, choose 16 width of cloth CT images as the reference image, respectively they have been carried out identical deformation process.

A lot of somes algorithms at random that selection is all global optimization approach are with reference to algorithm.The warping function number of control points the more is out of shape more meticulously, but at this moment the warping function parameter is more many, and computational complexity is more high.Choosing in the experiment that warping function control counts is 4.For the number of spots that rises in a lot of somes algorithms at random, each parameter is got 2,3,4 and 5 sampled points respectively, at this moment the total sampling number of algorithm is 2 * (4 ²) ², 2 * (4 ²) ³, 2 * (4 ²) ⁴With 2 * (4 ²) ⁵The experimental result contrast as shown in Table 1.

As can be seen from Table 1, use the registration rate of accuracy reached to 93.75% of this paper global optimization approach, more than or equal to the registration accuracy rate of reference algorithm.Increase along with reference algorithm sampling number, registration accuracy rate with reference to algorithm also increases gradually, when each parameter sampling point is reached 4, the registration rate of accuracy reached is to maximal value 93.75%, increase sampling number again, the registration accuracy rate also no longer increases, and at this moment increasing sampling number again is invalid to improving the registration accuracy rate, has the problem of sampled point surplus.Use this paper algorithm a routine registration error still to be arranged at 16 illustrations in as registration, analyze its reason, mainly be the content complexity because of the CT image, frequency spectrum medium-high frequency component composition is many, has given up the cause of too much high fdrequency component for reducing sample frequency when determining highest frequency.Be 2 o'clock at the reference algorithm to each parameter sampling point, a width of cloth CT image of registration error is shown in figure one.

The nonuniform sampling global optimization approach that proposes can calculate the optimum sampling number of spots, so on the basis that guarantees the registration accuracy rate, can avoid the problem of superfluous sampling, reduce calculated amount to greatest extent, reduce the image registration time, improve the robustness of registration Algorithm.This is applied to the clinical vital role of using to registration Algorithm to improving speed and the robustness of medical figure registration.

Result's contrast of table 116 group test

Fig. 3 is the structural representation according to an embodiment of figure registration system of the application's global optimization, comprise: comprise and read in module, sampling module, thick registration module and registration module, the described module of reading in is used for reading in deformation pattern, and calculating target function is at deformation coefficient highest frequency everywhere; Described sampling module is used for sampling and obtains the deformation pattern sequence; Described thick registration module is used for reading in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, and finishes thick registration; The deformation coefficient value that described registration module is used for described thick images after registration is starting point, the global minimum point of ferret out function, finish the registration of image, it is that sample frequency is carried out nonuniform sampling to deformation coefficient that described sampling module also is used for to be greater than or equal to 2 times of described highest frequency, obtains the deformation pattern sequence.

A kind of embodiment, sample frequency are 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

A kind of embodiment reads in module and also is used for reading in deformation pattern, uses Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

The deformation coefficient value that a kind of embodiment, registration module also are used for described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finishes the registration of image.

Above content be in conjunction with concrete embodiment to further describing that the present invention does, can not assert that concrete enforcement of the present invention is confined to these explanations.For the general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, can also make some simple deduction or replace.

Claims (8)

1. the method for registering images of a global optimization is characterized in that, comprising:

Read in deformation pattern, calculating target function is at deformation coefficient highest frequency everywhere;

Be that sample frequency is carried out nonuniform sampling to deformation coefficient to be greater than or equal to 2 times of described highest frequency, obtain the deformation pattern sequence;

Read in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, finish thick registration;

Deformation coefficient value to described thick images after registration is starting point, and the global minimum point of ferret out function is finished the registration of image.

2. the method for registering images of global optimization as claimed in claim 1 is characterized in that, described sample frequency is 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of described deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

3. the method for registering images of global optimization as claimed in claim 1 is characterized in that, the wherein said deformation pattern that reads in, and calculating target function comprises at deformation coefficient highest frequency everywhere:

Read in deformation pattern, use Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

4. as appointing the method for registering images of a described global optimization in the claim 1 to 3, it is characterized in that wherein said deformation coefficient value to described thick images after registration is starting point, the global minimum point of ferret out function, the registration of finishing image comprises:

Deformation coefficient value to described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finishes the registration of image.

5. the figure registration system of a global optimization comprises and reads in module, sampling module, thick registration module and registration module, and the described module of reading in is used for reading in deformation pattern, and calculating target function is at deformation coefficient highest frequency everywhere; Described sampling module is used for sampling and obtains the deformation pattern sequence; Described thick registration module is used for reading in reference picture, with the corresponding warping function of minimum target functional value described deformation pattern is out of shape, and finishes thick registration; The deformation coefficient value that described registration module is used for described thick images after registration is starting point, the global minimum point of ferret out function, finish the registration of image, it is characterized in that: it is that sample frequency is carried out nonuniform sampling to deformation coefficient that described sampling module also is used for to be greater than or equal to 2 times of described highest frequency, obtains the deformation pattern sequence.

6. the figure registration system of global optimization as claimed in claim 5 is characterized in that, described sample frequency is 2 ω _Max(θ) (g ' (x, m)), wherein: the highest frequency of described deformation pattern is ω to max _Max(θ), deformation coefficient is that (x, θ), the similarity criterion is got SSD to g.

7. the figure registration system of global optimization as claimed in claim 6 is characterized in that, the described module of reading in also is used for reading in deformation pattern, uses Fourier window or method of wavelet calculating target function at deformation coefficient highest frequency everywhere.

8. as the figure registration system of each described global optimization in the claim 5 to 7, it is characterized in that, the deformation coefficient value that described registration module also is used for described thick images after registration is starting point, with the global minimum point of gradient descent method ferret out function, finishes the registration of image.

Images