CN101294954B - Image registration method based on information potential - Google Patents

Abstract

The invention discloses an image registration method based on the information potential, and a nearest point iterative nearest point algorithm is measured based on the measure of the information potential and Euclidean distance; the information potential in the information theory is used for measuring the similarity of gray level distribution of two protein points in two images and measuring the similarity of the two protein points; the gray level information of the protein points is combined with geometric information, thus obtaining a searching nearest point used for matching the images with bad distortion automatically; the iterative nearest point algorithm is a point-matching algorithm; the traditional iterative nearest point algorithm has high requirement for initial transformation; the invention provides a measure based on the information potential and Euclidean distance, thus improving the robustness of registration algorithm and being applied to the image analysis field, particularly gel electrophoresis images; the method of the invention is very effective for the registration of the gel electrophoresis images, and can also be applied to various clinics and basic research widely.

Description

A kind of method for registering images based on information potential

Technical field

The invention belongs to biometric image registration technique field, relate to a kind of method for registering images,, biometric image data is handled based on the ICP framework based on information potential.

Background technology

Two dimensional gel electrophore-sis is a kind of application protein spots isolation technics very widely in proteinology research.Based on protein molecular weight and electric charge size, the two dimensional gel electrophore-sis technology can be separated different protein, generates width of cloth two dimension point diagram, wherein a kind of protein of each some expression.The difference that the researcher often need carry out between the different gels compares.This need find the corresponding protein spots of representing same protein in the different gels.Because the gel electrophoresis process is comparatively complicated, even same sample, asynchronism(-nization), the gel difference that obtains also can be very big.If data are very big, only with the naked eye be difficult to relatively.Therefore, the registration Algorithm by means of computing machine is very much necessary.

The gel electrophoresis images method for registering that forefathers propose can be divided into three kinds: based on unique point, based on two kinds of pixel grey scale, combination fronts.Present most gel electrophoresis images method for registering is based on the method for registering of unique point.Generally speaking, at first extract the unique point in the image based on the method for unique point, the barycenter of protein spots is then based on these unique point computational transformations.CAROL matees the border through trigonometric ratio, looks for the correspondence of protein spots.People such as F.A.Potra (F.A.Potra., X.Liu, F.Seillier-Moiseiwitsch, A.Roy; Y.Hang, et al., Protein image alignment via piecewise affine transformations; Journal of Computational Biology, 13 (2006)), different with forefathers; Be not between two width of cloth images, to mate, but structure can be represented the template of set of diagrams picture, then each width of cloth image and template in this picture group picture is mated.In people's such as F.A.Potra the method, unique point and correspondence thereof are known.

Method for registering based on pixel grey scale directly utilizes the gray-scale value of image to estimate the conversion of being asked.This method biggest advantage is exactly need not cut apart.Smilansky (Z.Smilansky; Automatic registration for images of two-dimensional protein gels, Electrophoresis22 (2001) 1616-1626.) propose for the first time and will be applied on the gel electrophoresis images based on the method for registering of pixel grey scale.The basic thought of the method for Smilansky is: based on pixel grey scale, based on regional area, show based on complementary false colour colour display screen.

Veeser (S.Veeser; M.J.Dunn; G.Z.Yang; Mul tiresolutionimage registration for two-dimensional gel electrophoresis, Proteomics 1 (2001) 856-870.) etc. people's method for registering is based on multiresolution and BFGS (Broyden-Fletcher-Goldfarb-Shanno) optimized Algorithm thereof.

The somebody is applied to the method for registering that unique point combines with pixel grey scale information on the gel electrophoresis images registration.Rohr (K.Rohr; P.Cathier; S.

; Elasticregistration of electrophoresis images using intensityinformation and point landmarks; Pattern Recognition, 37 (2004) 1035-1048.) etc. the people is applied to the gel electrophoresis images registration with PASTAGA (PASha Treating Additional GeometricAttribute) method.In original PASTAGA method, three-dimensional curve is used as unique point.In people's such as Rohr the work, the barycenter of protein spots is as unique point.In people's such as Carlo the work, SSD (sum of square differences) estimates as tolerance, and four summits of image are as unique point.Wherein unique point not only limits deformation field, but also goes on foot the starting condition of optimizing as each.People's emphasis such as Carlo mate the image with " smileeffect " metaboly.

Summary of the invention

The objective of the invention is to solve the problem that the conventional art iterative closest point algorithms is had relatively high expectations to initial transformation; For this reason; The present invention proposes a kind of method for registering images based on information potential, is a kind of new iterative closest point method, and it is applied on the two dimensional gel electrophore-sis image.Combine effectively through the half-tone information and the geological information of information potential, so method of the present invention is robust more protein spots.

In order to realize the object of the invention, proposition is following based on the technical scheme of the method for registering images of information potential:

Step S1: utilize computer equipment, measure the iterative closest point algorithms of closest approach based on estimating of information potential and Euclidean distance;

Step S2: measure the similarity of two protein site intensity profile in two width of cloth images with the information potential in the information theory, weigh the similarity of two protein spots;

Step S3: the half-tone information and the geological information of protein spots are combined, obtain the search closest approach, be used for mating automatically to being out of shape serious image.

According to embodiments of the invention, two protein spots P in said calculating two width of cloth images _i, Q _j, be used to obtain Euclidean distance

According to embodiments of the invention, calculate two protein spots P _i, Q _j, be used for the acquired information gesture and do

According to embodiments of the invention, with the Euclidean distance that obtains

With the information potential that obtains

Normalization obtains normalized Euclidean distance E _EucWith information potential E _Ip, and with normalized Euclidean distance E _EucWith information potential E _IpLinear junction lumps together, obtain between the protein spots apart from IP-Euc.

According to embodiments of the invention,, find nearest point set according to the distance between the protein spots.

According to embodiments of the invention,, estimate geometric transformation between two width of cloth images according to the nearest point set that finds.

According to embodiments of the invention, if end condition is satisfied in the geometric transformation that obtains, just with it as final result of matching; Otherwise, draw the result who satisfies end condition and end.

According to embodiments of the invention, said information potential

For: two protein spots P _i, Q _jInformation potential energy between two groups of grey scale pixel values that the place image-region comprises.

According to embodiments of the invention, said Euclidean distance E _EucCombination coefficient and information potential E _IpCombination coefficient all be variable, and the perseverance of adding up is 1.

Described method for registering images based on information potential is used for gel electrophoresis images to be handled.

Description of drawings

Fig. 1 is the process flow diagram of the method for registering images implementation procedure based on information potential of the present invention.

Fig. 2 a is the gel electrophoresis of protein image of not registration

Fig. 2 b is to use the registration result of method of the present invention to the gel electrophoresis of protein image.

Embodiment

To combine accompanying drawing that the present invention is specified below, and be to be noted that described embodiment only is intended to be convenient to understanding of the present invention, and it is not played any qualification effect,, be not difficult to find out the beneficial effect that the present invention produces through statement to embodiment.

The inventive method is based on traditional iterative closest point (Iterative Closest Point) algorithm; With information potential (Information potential) estimating as the measurement closest approach; Thereby the half-tone information and the geological information of protein spots are combined; Utilize the half-tone information and the geological information of protein spots, look for nearest point set; Iterative closest point algorithms is an a kind of some matching algorithm, and traditional iterative closest point algorithms is had relatively high expectations to initial transformation, and the present invention provides based on the estimating of information potential and Euclidean distance, thereby, improved the robustness of registration Algorithm.

Core of the present invention is, iterative closest point algorithms of measuring closest approach based on estimating of information potential and Euclidean distance of the present invention; This method has no hypothesis to the intensity profile of protein spots, measures the similarity of two protein site intensity profile in two width of cloth images with the information potential in the information theory, thereby weighs the similarity of two protein spots; Because this method has been utilized the half-tone information and the geological information of protein spots, so have the ability of powerful search closest approach; Therefore for the comparatively serious image of distortion good matching effect can be arranged; This method can be used for the registration of gel electrophoresis images; Especially in anamorphose serious situation comparatively, have very accurately and the matching result of robust.

The characteristic of the inventive method is, utilizes computer equipment, adopts based on the estimating of information potential and Euclidean distance, and use iterative closest point algorithms framework matees image automatically; The image geometry characteristic that traditional iterative closest point algorithms utilization has extracted finds nearest point set based on Euclidean distance, through optimizing a certain objective function, finds required conversion then; The present invention proposes information potential is combined with Euclidean distance, as the nearest measure of a pointset of a kind of measurement; Proposed by the invention estimating is primarily aimed at the special card of gel electrophoresis images, in conjunction with the geological information and the half-tone information of protein spots; From the anamorphose degree, the not enough robust of traditional iterative closest point algorithms; Algorithm robustness of the present invention is more intense; Method of the present invention is very effective for images match, and especially for the more serious image of deformation extent, the result who obtains is well more a lot of than commonsense method; This method can be widely used in various clinical and the fundamental researchs.The present invention is applied to art of image analysis, especially for gel electrophoresis images.

Implement the process flow diagram based on the method for registering images implementation procedure of information potential referring to Fig. 1, basic performing step is following:

Step 1: calculate two protein spots P according to formula (1) _i, Q _jEuclidean distance:

$M_{\mathrm{ij}}^{\mathrm{euc}}=\left|\right|P_i{Q}_j\left|\right|$ (1)

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter Expression protein spots P _i, Q _jEuclidean distance.With the Euclidean distance normalization that obtains, with formula (2):

$E_{\mathrm{euc}}=\frac{M_{\mathrm{ij}}^{\mathrm{euc}}}{\underset{\stackrel{i\∈\{1,...,N_1\}}{j\∈\{1,...,N_2\}}}{\max \left(M_{\mathrm{ij}}^{\mathrm{euc}}\right)}}---\left(2\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter

Expression protein spots P _i, Q _jEuclidean distance, parameter E _EucExpression protein site P _i, Q _jEuclidean distance after the normalization, parameter N ₁The number of the protein that splits of expression floating image, parameter N ₂The number of the protein that splits of expression reference picture.

Step 2: calculate two protein spots P according to formula (3) _i, Q _jInformation potential:

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter K ₁Expression protein spots P _iThe pixel size that the region comprises, parameter K ₂Expression protein spots Q _jThe pixel size that the region comprises,

The area of expression protein spots region, parameter ii representes protein spots P _jIi pixel of region, parameter jj representes protein spots Q _jJj pixel of region, parameter x _IiExpression protein spots P _iThe gray-scale value of ii pixel of region, parameter x _JjExpression protein spots region Q _jThe gray-scale value of jj pixel, parameter

Expression protein spots P _i, Q _jInformation potential.

With the information potential normalization that obtains with formula (4):

$E_{\mathrm{ip}}=\frac{M_{\mathrm{ij}}^{\mathrm{ip}}}{\underset{\stackrel{i\∈\{1,...,N_1\}}{j\∈\{1,...,N_2\}}}{\max \left(M_{\mathrm{ij}}^{\mathrm{ip}}\right)}}---\left(4\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter Expression protein spots P _i, Q _jInformation potential, parameter E _IpExpression protein site P _i, Q _jInformation potential after the normalization, parameter N ₁The number of the protein that splits of expression floating image, parameter N ₂The number of the protein that splits of expression reference picture.

Step 3: the Euclidean distance E after the normalization that step 1 is obtained _EucWith the information potential E after step 2 normalization _IpLinear junction lumps together, obtain between the protein spots apart from IP-Euc:

IP-Fuc(P _i，Q _i)＝(1-λ)E _euc+λE _ip (5)

Wherein, parameter lambda is represented the numerical value between one 0 to 1, the different different between the two proportion of λ value representation difference.

Step 4: according to the distance between the protein spots, find nearest point set, at the corresponding protein spots of looking for the protein spots on the floating image on the reference picture, distance is minimum just thinks the protein spots of asking.

Step 5: after getting rid of out-of-bounds, estimate geometric transformation between two width of cloth images.

Step 6: if end condition is satisfied in the geometric transformation that step 5 obtains, just with it as final result of matching.Otherwise, continue step 1 to step 5, up to, draw the result step 7 that satisfies end condition and finish.

Said procedure uses the C Plus Plus establishment under the VC translation and compiling environment, CPU frequency 2.4G under the situation under the internal memory 256MB, has realized method of the present invention.

Illustrate implementation procedure of the present invention below:

Step 1: at first the image that needs registration is cut apart, obtained two protein spots set P and Q, set P belongs to floating image, and set Q belongs to reference picture, and set P is respectively N with the size of set Q ₁And N ₂

Step 2, calculate protein spots P according to formula (1-4) _i(i=1 ..., N ₁) and protein spots Q _i(j=1., N ₂) Euclidean distance, information potential, and carry out normalization.

Step 3: calculate protein spots P according to formula (5) _i(i=1 ..., N ₁) and protein spots Q _j(j=1 ..., N ₂) apart from IP-Euc, the λ value is 0.5 in this instance.

Step 4: according to the distance between the protein spots, find nearest point set, promptly corresponding protein spots, at the corresponding protein spots of looking for the protein spots on the floating image on the reference picture, distance is minimum just thinks the protein spots of asking.

Step 5: get rid of out-of-bounds point, get rid of and out-of-bounds can be divided into two steps here: at first, the distance that protein spots is right is during greater than 2.5 times of the variance of the distance of all corresponding protein spots, just with these two protein spots eliminatings; Then, it is right to get rid of 20% the poorest protein site, and here, the poorest right Euclidean distance of protein spots that refers to is maximum.

Step 6: floating image is implemented affined transformation.

Step 7: after judging the geometric transformation that floating image implementation step six is obtained, whether the mean value of the Euclidean distance quadratic sum of corresponding point less than less than thresholding 0.002, if satisfy, just with it as final result of matching.Otherwise, continue step 1 to step 6, up to, the mean value of the Euclidean distance quadratic sum of corresponding point whether less than less than till the thresholding 0.002.

In a word, the method for registering images based on information potential has very application prospects to gel images.It has following advantage: (1) to the unusual robust of noise, (2) are insensitive to starting condition.

Fig. 2 a is the gel electrophoresis of protein image of not registration.In image, in the rectangle frame, can find out that from figure corresponding protein spots does not overlap.

Fig. 2 b is to use the registration result of method of the present invention to the gel electrophoresis of protein image.In the rectangle frame, corresponding protein spots coincides together, and explains, uses method of the present invention and has successfully found corresponding protein spots from figure.

The above; Be merely the embodiment among the present invention, but protection scope of the present invention is not limited thereto, anyly is familiar with this technological people in the technical scope that the present invention disclosed; Can understand conversion or the replacement expected; All should be encompassed in of the present invention comprising within the scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (4)

1. the method for registering images based on information potential is characterized in that, comprises as follows:

Step 1: calculate two protein spots P _i, Q _jEuclidean distance represent as follows:

$M_{\mathrm{ij}}^{\mathrm{euc}}=\left|\right|P_i{Q}_j\left|\right|---\left(1\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter

Expression protein spots P _i, Q _jEuclidean distance; The Euclidean distance normalization that obtains is represented as follows:

$E_{\mathrm{euc}}=\frac{M_{\mathrm{ij}}^{\mathrm{euc}}}{\underset{j\∈\{1,...,N_2\}}{\underset{i\∈\{1,...,N_1\}}{\max \left(M_{\mathrm{ij}}^{\mathrm{euc}}\right)}}}---\left(2\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter Expression protein spots P _i, Q _jEuclidean distance, parameter E _EucExpression protein site P _i, Q _jEuclidean distance after the normalization, parameter N ₁The number of the protein that splits of expression floating image, parameter N ₂The number of the protein that splits of expression reference picture;

Step 2: calculate two protein spots P _i, Q _jInformation potential represent as follows:

$M_{\mathrm{ij}}^{\mathrm{ip}}=\frac{K_1{K}_2\·e^{\frac{2\left(\right|P_i|-|Q_j\left|\right)}{\left|P_i\right|+\left|Q_i\right|}}}{\underset{\mathrm{ii}=1}{\overset{K_1}{\mathrm{\Σ}}}\underset{\mathrm{jj}=1}{\overset{K_2}{\mathrm{\Σ}}}G(x_{\mathrm{ii}}-x_{\mathrm{jj}},2{\mathrm{\σ}}^{2}I)}---\left(3\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter K ₁Expression protein spots P _iThe pixel size that the region comprises, parameter K ₂Expression protein spots Q _jThe pixel size that the region comprises, || the area of expression protein spots region, parameter ii representes protein spots P _iIi pixel of region, parameter jj representes protein spots Q _jJj pixel of region, parameter x _IiExpression protein spots P _iThe gray-scale value of ii pixel of region, parameter x _JjExpression protein spots region Q _jThe gray-scale value of jj pixel, parameter

Expression protein spots P _i, Q _jInformation potential;

The information potential normalization that obtains is represented as follows:

$E_{\mathrm{ip}}=\frac{M_{\mathrm{ij}}^{\mathrm{ip}}}{\underset{j\∈\{1,...,N_2\}}{\underset{i\∈\{1,...,N_1\}}{\max \left(M_{\mathrm{ij}}^{\mathrm{ip}}\right)}}}---\left(4\right)$

Wherein, parameter i representes i protein spots in the floating image, and parameter j representes j protein spots in the reference picture, parameter

Expression protein spots P _i, Q _jInformation potential, parameter E _IpExpression protein site P _i, Q _jInformation potential after the normalization, parameter N ₁The number of the protein that splits of expression floating image, parameter N ₂The number of the protein that splits of expression reference picture;

Step 3: the Euclidean distance E after the normalization that step 1 is obtained _EucWith the information potential E after step 2 normalization _IpLinear junction lumps together, and obtains representing as follows apart from IP-Euc between the protein spots:

IP-Euc(P _i，Q _j)＝(1-λ)E _euc+λE _ip (5)

Wherein, parameter lambda is represented the numerical value between one 0 to 1, the different different between the two proportion of λ value representation difference;

Step 4: according to the distance between the protein spots, find nearest point set, at the corresponding protein spots of looking for the protein spots on the floating image on the reference picture, distance is minimum just thinks the protein spots of asking;

Step 5: after getting rid of out-of-bounds,, estimate geometric transformation between two width of cloth images according to the nearest point set that finds; Said eliminating is out-of-bounds put and is divided into two steps: at first, the distance that protein spots is right is just got rid of these two protein spots during greater than 2.5 times of the variance of the distance of all corresponding protein spots; Then, it is right to get rid of 20% the poorest protein site, and here, the poorest right Euclidean distance of protein spots that refers to is maximum;

Step 6: if end condition is satisfied in the geometric transformation that step 5 obtains, just with it as final result of matching; Otherwise, continue step 1 to step 5, up to, draw the result step 7 that satisfies end condition and finish.

2. method for registering images according to claim 1 is characterized in that, said information potential

For: two protein spots P _i, Q _jInformation potential energy between two groups of grey scale pixel values that the place image-region comprises.

3. method for registering images according to claim 1 is characterized in that, said Euclidean distance E _EucCombination coefficient and information potential E _IpCombination coefficient all be variable, and the perseverance of adding up is 1.

4. the described method for registering images based on information potential of claim 1 is used for the gel electrophoresis images processing.

Images