CN106127719A - A kind of novel neutral net Method of Medical Image Fusion - Google Patents

Abstract

A kind of novel neutral net Method of Medical Image Fusion, the method utilizes Shearlet conversion with yardstick, original image (CT image and MR image) can be carried out decomposition in any direction and obtains directional information and the texture information of image.There is the drain capacitance integral weighting summation method of Global-Coupling, variable threshold value, the product coupling of internal act, input first with PCNN, for the selection of Shearlet conversion medium-high frequency sub-band coefficients；The ignition times of each pixel when then calculating the input outside respectively as self adaptation PCNN of CT image and MR image, selects the pixel lighted a fire often in two width image corresponding pixel points as the pixel of both fused image；Again by the Shearlet coefficient obtained is carried out Shearlet inverse transformation, obtain the image after CT image and MR image co-registration.This method is converted by Shearlet and CT image and MR image are merged the quality improving medical image by self adaptation PCNN, compared with using single use Shearlet conversion or self adaptation PCNN, it is reduced data redundancy and improves the advantage of picture contrast.

Description

A kind of novel neutral net Method of Medical Image Fusion

Technical field

The present invention relates to a kind of novel neutral net Method of Medical Image Fusion.This method can improve fusion image matter Amount, reaches comparatively ideal syncretizing effect, will be better than using merely one of which to become in visual effect and objective evaluation index Change, also have a clear superiority in compared with tradition.Have a wide range of applications in field of medical images.

Background technology

In the latter stage seventies, Tenney and Sandell proposes the concept of multi-sensor information fusion.And image co-registration skill The image procossing new technique that art grows up on this basis.The purpose of image co-registration is multiple images of comprehensive same scene Information, such result is exactly to obtain being more suitable for the image of human vision and computer vision, or is more suitable for carrying out Next step image processed.Owing to image fusion technology can effectively utilize complementarity and the redundancy of different images information, Therefore by merging later image compared with the image that any single channel obtains, there is prominent advantage.

In recent years, along with computer technology and the development of computer vision technique, Medical Imaging there has also been considerable sending out Exhibition.Medical Imaging is intersected to form with technology such as graph and image processing technology, computer vision technique and pattern recognitions The i.e. Medical Image Processing of individual new disciplines technology branch, this technology, mainly as a kind of disease detection means, assists doctor couple The cause of disease of patient judges.And wherein, Medical image fusion is exactly an important field of research.

For strengthening multi-focus and the syncretizing effect of Medical image fusion, the present invention propose a kind of based on Shearlet and from Adapt to the image interfusion method that PCNN combines.Similar with wavelet transformation, Shearlet conversion has simple mathematical structure, this One advantage makes it to associate with multiresolution analysis easily.Piece image converts through Shearlet, can be by it Deconstructing by any yardstick and direction, therefore Shearlet conversion ratio traditional wavelet can capture more directional information With other geological informations.For image co-registration, Shearlet conversion is a kind of selection very well.It is special that PCNN has Global-Coupling Property, become the characteristic such as drain capacitance integral weighting summation of threshold property, the product coupled characteristic of internal act, input, can be used In the selection of Shearlet conversion sub-band coefficients, using the ignition times in PCNN method as convergence strategy.Test result indicate that, The method will be better than using merely one of which to convert in visual effect and objective evaluation index, with traditional wavelet and Laplacian Pyramid Transform is compared and is also had a clear superiority in.

Summary of the invention

The most conventional identification image is to be converted by Shearlet or self adaptation PCNN monotechnics, this identification figure image space There is the problem that visual effect is the best in method, the present invention provides a kind of novel neutral net Method of Medical Image Fusion, the method Visual effect and objective evaluation index will be better than use merely one of which to convert, also have obvious gesture compared with tradition； Have a wide range of applications at medical image；

The present invention solves the used technical scheme of its technical problem: a kind of novel neutral net Medical image fusion side Method, it utilizes Shearlet to convert the advantage can decomposed image with yardstick in any direction, obtains more comprehensive With various detailed information, utilize simultaneously PCNN Global-Coupling characteristic, become threshold property, the product coupled characteristic of internal act, The characteristics such as the drain capacitance integral weighting summation of input, can use it for the selection of Shearlet conversion sub-band coefficients, by PCNN side Ignition times in method, as convergence strategy, carries out Shearlet inverse transformation with Shearlet coefficient, obtains fusion image.

Specifically comprising the following steps that of said method

(1) respectively the source images A and source images B being complete registration is carried out Shearlet conversion obtain low frequency coefficient and The high frequency coefficient of all directions yardstick；

(2) low frequency sub-band coefficient is used the convergence strategy averaged, draws the low frequency coefficient of fusion image, respectively will The high frequency coefficient of source images A and source images B, as the input of PCNN1 and PCNN2, obtains two Fire mapping image O_AAnd O_B, the most defeated Entering passage is F, and high frequency coefficient fusion rule is as follows:

(3) the Shearlet coefficient obtained is carried out Shearlet inverse transformation, obtain fusion image.

Based on self adaptation PCNN in above-mentioned Method of Medical Image Fusion based on Shearlet conversion and self adaptation PCNN Fusion mainly comprise the following specific steps that:

(1) initiation parameter.Set population scale NP according to neuron population, then set maximum evolutionary generation G_max, prominent Varying function F, crossing-over rate CR and end function.Randomly choose vector initialising population within the limits prescribed.If response variable t= 0。

(2) assessment population.The fitness value of each individuality in population is calculated according to fitness function.We use image Comentropy is as fitness function.The calculating formula of image entropy H is as follows:

H (p)=-p₁×log₂p₁-p₂log₂p₂

Y during wherein p1 and p2 represents the output of PCNN respectively_ij(n)=0 and Y_ijN probability that ()=1 occurs, meanwhile, record The most optimal individuality

(3)Represent in t secondary response, the value of the i-th vector in population,Represent each target individualSudden change Vector,For intersecting vector,Being the result when (t+1) secondary conversion, f is fitness function.

According to formula

Carry out suddenling change, intersect and select operation.

(4) if to reach stop condition, then the value exported is then optimal parameter, otherwise, makes t=t+1, then repeats The operation of 2-4 step.

Wherein, in Method of Medical Image Fusion based on Shearlet conversion and self adaptation PCNN based on described based on from The convergence strategy of suitable PCNN is when image procossing, and the image participating in merging is A and B, and image A and B makees the input thorn of PCNN Swashing, if each neuron links with surrounding k × k neighborhood neuron in PCNN, two PCNN obtain being output as A and B after running Mapping graph YA and the YB duration of ignition of figure, then utilizes judgement selection opertor to judge, two output results according to igniting Situation it may determine that well-marked target be present in A figure or B figure in.

Image co-registration based on Shearlet conversion mainly comprises the steps:

(1) the two width images completing registration being carried out Shearlet conversion respectively, so, source images is just decomposed It is a low-frequency image and several high frequency direction sub-band images.

(2) respectively with HFS, the low frequency part of image being used different fusion rules, take herein is low frequency Part coefficient is averaged, and HFS coefficient takes big fusion rule.

(3) carry out Shearlet inverse transformation, i.e. can get fusion image.

The present invention compared with prior art, has following features and advantage: the present invention is simultaneously by shearlet and PCNN It is used in combination, carries out image co-registration process.First image is carried out shearlet conversion, it is thus achieved that high frequency coefficient and low frequency coefficient, High frequency coefficient is carried out self adaptation PCNN again and merges the new high frequency of acquisition.Finally carry out shearlet inverse transformation.Obtain and new melt figure Picture；The method will be better than using merely one of which to convert, compared with tradition also in visual effect and objective evaluation index Have a clear superiority in.Have a wide range of applications at medical image.

Accompanying drawing explanation

Fig. 1 is a kind of novel neutral net Method of Medical Image Fusion flow chart of the present invention.

Fig. 2 is the multi-focus image fusion result ratio of a kind of novel neutral net Method of Medical Image Fusion of the present invention Relatively, (a) Wavelet Transform Fusion result in Fig. 2；(b) Laplacian Pyramid Transform fusion results；(c) PCNN fusion results；(d) Shearlet fusion results；(e) Shearlet-PCNN fusion results.

Fig. 3 is the brain CT of a kind of novel neutral net Method of Medical Image Fusion of the present invention and MRI image melts Close results contrast, (a) Wavelet Transform Fusion result in figure；(b) Laplacian Pyramid Transform fusion results；C () PCNN merges Result；(d) Shearlet fusion results；(e) Shearlet-PCNN fusion results.

Fig. 4 and Fig. 5 gives the objective evaluation index of five kinds of method fusion results, Fig. 4 and Fig. 5 have employed comentropy and The value of average gradient weighs the quality of fusion image, and then evaluates the effectiveness of the present embodiment fusion method.

Detailed description of the invention

Shown in Fig. 1, a kind of novel neutral net Method of Medical Image Fusion of the present invention, utilize Shearlet conversion permissible The advantage decomposed image with yardstick in any direction, it is thus achieved that more comprehensive and various detailed information, utilizes simultaneously PCNN Global-Coupling characteristic, change threshold property, the product coupled characteristic of internal act, the drain capacitance integral weighting summation etc. of input Characteristic, uses it for the selection of Shearlet conversion sub-band coefficients, using the ignition times in PCNN method as convergence strategy, and will The Shearlet coefficient obtained carries out Shearlet inverse transformation, obtains fusion image, and implementation process is as follows:

(1) respectively the source images A and source images B being complete registration is carried out Shearlet conversion obtain low frequency coefficient and The high frequency coefficient of all directions yardstick；

(2) low frequency sub-band coefficient is used the convergence strategy averaged, draws the low frequency coefficient of fusion image, respectively will The high frequency coefficient of source images A and source images B, as the input of PCNN1 and PCNN2, obtains this two Fire mapping image O_AAnd O_B, its Middle input channel is F, and high frequency coefficient fusion rule is as follows:

(3) the Shearlet coefficient obtained is carried out Shearlet inverse transformation, so that it may obtain fusion image.

Fig. 2 show the image interfusion method of self adaptation PCNN.Its mathematical formulae represented has:

F_ij(n)=I_ij (1)

U_ij(n)=F_ij(n)(1+βL_ij(n)) (3)

θ_ij(n)=exp (-α_θ)θ_ij(n-1)+V_θY_ij(n-1) (4)

Formula (1) to formula (5) represents that the leak integrators of input domain and link field is in the PCNN neuron models of this simplification Eliminate.Formula (1) is outside stimulus I_ijInput as neuron；Formula (2), to the neuron weighted sum in neighborhood, is then made Connection for neuron inputs；Formula (3) represents that t1 moment neuron firing i.e. has pulse to export, V_θWhen being arranged to pulse output Threshold value, then threshold value by exponential form decay, t2 has again pulse to export during the moment, and threshold value is configured to V again_θ；

Method of Medical Image Fusion Shearlet therein conversion based on Shearlet conversion and self adaptation PCNN, specifically Step is as follows:

(1) the two width images completing registration being carried out Shearlet conversion respectively, so, source images is just decomposed It is a low-frequency image and several high frequency direction sub-band images.

(2) respectively with HFS, the low frequency part of image being used different fusion rules, take herein is low frequency Part coefficient is averaged, and HFS coefficient takes big fusion rule.

(3) carry out Shearlet inverse transformation, i.e. can get fusion image.

Shearlet principle mathematical equation is:

OrderMeet following condition:

(1)WhereinFourier transform for Ψ；

(2)For continuous wavelet,

(3)AndOn interval (-1,1), Ψ₂> 0 and ‖ Ψ₂‖=1, then claim by Ψ,AndThe following system generatedFor shearlet system, claim Ψ_ast(x) For shearlet, wherein A_aAnd B_sIt is respectively anisotropic expansion matrix and shears matrix.

The letter that the image interfusion method combined based on Shearlet and PCNN as can be seen from Figures 2 and 3 comprises at image Breath amount and image definition aspect are substantially better than other four kinds of methods, while preserving image information to greatest extent, try one's best again Avoid image fault.

Method of Medical Image Fusion based on Shearlet conversion and self adaptation PCNN can improve fusion image matter, reaches Comparatively ideal syncretizing effect, will be better than using merely one of which to convert, with biography in visual effect and objective evaluation index System is compared and is had a clear superiority in.

Claims (2)

1. a novel neutral net Method of Medical Image Fusion, its method feature: utilize Shearlet conversion in arbitrarily side Carry out decomposing the acquisition directional information of image, texture information etc. to image to yardstick；PCNN is utilized to have Global-Coupling, variable threshold The advantage that value, the product coupling of internal act, the drain capacitance integral weighting of input are sued for peace, uses it in Shearlet conversion high Frequently the selection of sub-band coefficients；Then, each pixel when calculating the input outside respectively as self adaptation PCNN of CT image and MR image The ignition times of point, and select the pixel lighted a fire often in two width image corresponding pixel points as both fused image Pixel；Finally, then by the Shearlet coefficient obtained is carried out Shearlet inverse transformation, so that it may obtain CT image and MR figure As the image after fusion.

A kind of novel neutral net Method of Medical Image Fusion the most according to claim 1, its steps characteristic is:

(1) respectively CT image A and MR image B is carried out Shearlet conversion and obtains low frequency coefficient and the high frequency of all directions yardstick Coefficient；

$\left\{\begin{array}{c}F(i,j)=A(i,j),O_A\&GreaterEqual;O_B\\ F(i,j)=B(i,j),O_A<O_B\end{array}\right.$

(2) low frequency sub-band coefficient is used the convergence strategy averaged, draws the low frequency coefficient of fusion image, respectively CT is schemed As the high frequency coefficient of A and MR image B is as the input of PCNN1 and PCNN2, obtain this two Fire mapping image and, wherein input is logical Road is F, high frequency coefficient fusion rule:

H (p)=-p₁×log₂p₁-p₂log₂p₂

p₁And p₂Represent Y in the output of PCNN respectively_ij(n)=0 and Y_ijN probability that ()=1 occurs, records optimal simultaneously Body

(3) formula is used:

$U_{i,j}^{t+1}=\left\{\begin{array}{c}{X}_{i,j}^{t}rand\left(\right)\&le;CR\\ X_{i,j}^{t}rand\left(\right)>CR\end{array}\right.,j=1,2,...,D,$

$X_i^{t+1}=\left\{\begin{array}{c}{U}_i^{t+1},f\left(U_i^{t+1}\right)\&le;f\left(X_i^t\right)\\ X_i^t,f\left(U_i^{t+1}\right)>f\left(X_i^t\right)\end{array}\right.,$

$V_i^t=X_{best}^t+F(X_{r2}^t-X_{r3}^t)$

Table carries out suddenling change, intersect and select operation, shows in t secondary response, the value of the i-th vector in population, represents each Target individual sudden change vector, for intersecting vector, is the result when (t+1) secondary conversion, is fitness function.