CN105184764B - A kind of method for registering images based on real coding clonal selection algorithm - Google Patents
A kind of method for registering images based on real coding clonal selection algorithm Download PDFInfo
- Publication number
- CN105184764B CN105184764B CN201510229255.XA CN201510229255A CN105184764B CN 105184764 B CN105184764 B CN 105184764B CN 201510229255 A CN201510229255 A CN 201510229255A CN 105184764 B CN105184764 B CN 105184764B
- Authority
- CN
- China
- Prior art keywords
- antibody
- image
- affinity
- collection
- gray level
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Abstract
The invention discloses a kind of method for registering images based on real coding clonal selection algorithm, and it is not high mainly to solve the problem of that the prior art is easily absorbed in local optimum image registration accuracy in majorized function.Implementation step is：(1) input reference chart and the figure that floats；(2) using real number coding method to antibody initialization of population；(3) structure normalized mutual information function is as object function；(4) population affinity of antibody is calculated；(5) antibody selected, cloned and is made a variation；(6) memory antibody collection is formed；(7) memory of antibody；(8) judge whether to reach end condition, if not reaching, jump to (4) and continue to execute, otherwise, perform (9) step；(9) it is registrated image with optimal antibody and exports result.The present invention optimizes mutual information function using the clonal selection algorithm based on real coding, solves the problems, such as that existing method is easily absorbed in local optimum in majorized function, improves the accuracy of image registration.
Description
Technical field
The invention belongs to image registration fields, are related to the optimization method of object function, specifically a kind of to be based on Immune Clone Selection
The method for registering images of algorithm.
Background technology
Image registration techniques have been widely used in computer vision, mould as a very important research topic
Formula matching, medical image analysis and remote sensing image processing.Due to the novel sensor to emerge in an endless stream, people are to the acquisition energy of image
Power fast lifting.Novel sensor has a variety of different characteristics, thus different types of remote sensing images also constantly get up more.
In view of sensor has such as spectral information, geometric size and time order and function in all fields when obtaining image information
Apparent difference, so being difficult to meet actual conditions using single image information.In order to preferably obtain different figures
The various information of picture so as to obtain the remote sensing images of higher resolution, needs to make full use of the advantage of multi-modality images, to utilizing
Image under distinct device difference image-forming condition is merged, and the premise of image co-registration is exactly image registration.In brief, it leads
Syllabus be exactly will be to Same Scene in different moments, the two images of different angle or different sensors shooting carry out pair
Together.
There are mainly two types of types for the method for image registration at this stage：Method for registering images and feature based based on gray scale
Method for registering images.
Method for registering images based on gray scale does not need to carry out image feature extraction, but uses entire image or figure
The subregion of picture estimates the gray consistency between two images, and common method for measuring similarity has：Cross-correlation, phase phase
Pass and mutual information.Although the computation complexity of such method is higher, such method has been proved in image registration
Achieve good effect.Wherein, mutual information has the characteristics that：First, it does not need to carry out ash to the image of different modalities
Spend the hypothesis of correspondence.Second, mutual information has good robustness to noise.Third reaches accuracy registration in two images
When, mutual information obtains maximum value.
The method for registering images of feature based is needed first to the feature of image, and including point, line or region are detected, later
The feature detected is matched one by one, then by the spatial transform relation between matched feature assessment two images, into
And it is registrated image.Common characteristic detection method has：The method that Harris Corner Detections, Canny detective operators, image are divided
It is consistent with phase to extract characteristic point etc., it obtains needing to carry out feature using spatial relationship or constant description after characteristic point
It corresponds.When there is enough features, the method for registering images of feature based can be acquired easily close to the overall situation most
Excellent transformation parameter.
In general, image registration problem can be converted to the optimization problem of function after similarity measures are determined,
Similarity measurement reaches the image being registrated during maximum value.Common function optimization method has genetic algorithm, particle cluster algorithm
Deng.Genetic algorithm is the evolution algorithm to be grown up based on the natural selection of Darwin's biological evolution theory and genetic mechanisms,
It randomly generates initial population, then carries out fitness evaluation to each individual of initial population, and iteration selects to adapt to after starting
Spend high individual intersected with certain probability, mutation operation, generate new individual, then fitness is carried out to new individual and is commented
Valency usually sets certain iterations or iteration stopping after object function reaches a certain threshold value, exports final solution.Its
Shortcoming is that computation complexity is high, easy Premature Convergence, and is easily trapped into local optimum.
Artificial immune system (Artificial Immune system, AIs) is the one of natural imitation function of immune system
Kind intelligent method, it realizes that one kind is inspired by Immune System, the study of the natural defense mechanism by learning external substance
Technology, provide noise restrain oneself, teacherless learning, self-organizing, the evolutionary learnings mechanism such as memory, combine grader, neural network
With some advantages of the systems such as machine inference, therefore with providing the novel potentiality solved the problems, such as.Clone (Clone) is immune to be
The important theory of Immune System theory.Due to heredity and gene mutation of the immunocyte in proliferation, form immune thin
The diversity of born of the same parents, the continuous proliferation of these cells form clone.The vegetative propagation of cell is known as cloning.
Burnet in 1958 etc. proposes famous clonal selective theory, and central idea is：Antibody is natural products, with
The form of receptor is present in cell surface, and antigen can selectively react therewith.Antigen can cause carefully with reacting for corresponding antibodies
The Clonal increment of born of the same parents, the group have identical antibody specificity, and some of which cell clone is divided into antibody-producting cell, separately
Some are formed immunological memory cell and are reacted with the secondary immunity after participating in, and Immune Clone Selection is that organism immune system self-adaption resists
The dynamic process of primary stimuli, in this course, the biological natures such as study, memory, antibody diversity for being embodied are exactly artificial
What immune system was used for reference.Viewpoint based on information processing, it is believed that the essence of Immune Clone Selection be exactly a generation evolution in,
Near candidate disaggregation, according to affinity degree size, the group of a variation solution is generated.Clonal selection algorithm is by antibody one
Antigen affinity degree realizes the competition between individual, and effectively adjusts excessive competition, to keep the diversity of antibody population.
As a kind of new global optimization search, Immune Clonal Selection Algorithm is taken into account the overall situation in algorithm realization and is searched
Rope and local search, and mnemon is constructed, the single optimum individual of memory of genetic algorithm is become into one optimal solution of memory
Group.In general genetic algorithm, intersection is main operators, and variation is background operator, but clonal selection algorithm then exactly phase
Instead, and testing proves Immune Clonal Selection Algorithm performance better than corresponding genetic algorithm.In addition, the selection of Clone cells in itself
Mechanism has memory function, therefore can ensure that algorithm converges to optimal solution with probability l, and the genetic algorithm of standard then cannot.
Immune Clonal Selection Algorithm is applied in image registration by the present invention, and the method for using real coding, according to
The size of fitness function constantly updates the scale of population clone, to acquire optimal image registration parameter, has reached preferable
Image registration effect.By analysis of experimental data, the method for the present invention is better than traditional genetic algorithm.
Invention content
It is an object of the invention to be directed to the deficiency of above-mentioned prior art, a kind of image based on clonal selection algorithm is proposed
Method for registering improves the accuracy estimated registration parameter, realizes the correct registration to reference picture and floating image.
Technical scheme of the present invention step includes as follows：
(1) input resolution ratio identical reference picture imageR and floating image imageS；
(2) antibody population initializes：Antibody population is randomly generated, A represents antibody collection, and antibody collection A is by interim antibody collection Ar
With memory antibody collection AmComposition, N=r+m, r represent interim antibody collection ArIn interim antibody levels, m represents memory antibody collection Am
In memory antibody quantity, N is the total quantity of antibody in antibody collection A；
(3) with A represent reference picture imageR, B represent floating image imageS structures normalized mutual information function MI (A,
B) as object function：
Wherein, H (A) is the edge entropy of image A,pA(i) it is i-stage gray level in image A
The probability of appearance, M are the series of gray level in image A, and H (B) is the edge entropy of image B,pB
(j) be in image B j-th stage gray level occur probability, N be image B in gray level series,
The combination entropy between image A and image B, p (i, j) is i-stage gray level in image A, in image B j-th stage gray level this
To putting the joint probability that same position occurs simultaneously in image A and image B this two images, andh
(i, j) is the joint histogram of image A and image B, and n is the number of image A and image B overlapping regions pixel, wherein joint is straight
Square figure is defined as follows：
H (a, b) (0≤a≤K-1,0≤b≤L-1, K and L are the ranges of gray value in two images) represents ash in image A
Degree grade is a, and gray level is the number of the pixel pair of b in image B；
(4) normalized mutual information function MI (A, B) is optimized using clonal selection algorithm
4a) calculate the value of the affinity of antibody, i.e. normalized mutual information function MI (A, B) of population；
4b) antibody selects：Affinity of antibody according to descending is arranged, the highest preceding k individual of affinity of antibody is chosen and makees
For interim antibody collection Ar；
4c) antibody cloning：The k antibody chosen will by independent cloning, the number of clone be also it is indefinite, specifically
Rule be：Affinity is higher, and affinity serial number i is smaller, and the number being cloned is more, on the contrary, affinity is lower, affinity
Serial number i is bigger, is cloned the total quantity N that number is fewer, and antibody is clonedcSuch as following formula：
Wherein, β is a multiplication factor, and N is the antibody total quantity in collection of antibodies A, the value range of affinity serial number i
It is 1 to k, round () is an independent variable operator, is represented to nearest integer rounding, NcI-thCorresponding the
The scale that i antibody is cloned.
4d) antibody variation：Mutation operation is done to the antibody after clone.The random number between one 0 to 1 is generated, if this
Number is less than preset mutation probability, then to corresponding parameter into row variation：
xnew=x+ δ * λ
Wherein, xnewParameter after making a variation for variable x, δ=0.1* (xmax-xmin) to narrow the region of search factor, xmaxFor
The maximum value of variable x, xminFor the minimum value of variable x, λ is a random number between 0 to 1, for adjusting the step-length of x variations；
4e) form memory antibody collection:Recalculate the adaptation by antibody each in antibody population after above-mentioned mutation operation
The size of angle value i.e. affinity, if being higher than A by the affinity of antibody in population after above-mentioned mutation operationrIn it is corresponding
The affinity of antibody before variation just replaces original A with the antibody after mutation operationrIn, form memory antibody collection Am；
4f) antibody memory：The natural extinction of 5% B cell in artificial immune system bioselection is simulated, i.e., from antibody
The antibody of affinity minimum 5% is deleted in collection A, the new antibodies for the generation 5% that reinitializes with newly generated 5% antibody, replace
For the antibody that 5% affinity being deleted in antibody collection A is minimum.
(5) judge whether to meet evolution conditions：If iterations are not up to pre-defined evolutionary generation gene, jump to
It 4a) continues to execute, otherwise, performs (6)；
(6) when iterations reach pre-defined evolutionary generation gene, the highest individual of affinity is optimal antibody,
Spatial alternation is carried out to floating image imageS using the transformation parameter representated by optimal antibody, is calculated and floating after output transform
Association relationship between motion video and reference picture imageR matches the floating image after reference picture imageR and transformation
Standard, the image after finally output is registrated.
The present invention regards image registration problem as a function optimization problem, wherein, normalized mutual information function conduct
Object function optimizes object function using the clonal selection algorithm based on real coding, reaches maximum with association relationship
When solution carry out spatial alternation, obtain registration image, have the following advantages that compared with prior art：
First, the present invention regards image registration as a function optimization problem, is selected using the clone based on real coding
It selects algorithm to optimize object function mutual information, local optimum is easily trapped into when overcoming traditional genetic algorithm registration image
Shortcoming.
Second, due to the method that the present invention employs antibody real coding, compared with traditional binary coding method
A large amount of coding and decoding step, therefore operational efficiency higher are saved, it is as a result also more accurate.
Third, the main operators that clonal selection algorithm operates mutation operator as it are expanded in certain algebraically and are searched
Rope space.In addition Clone cells have memory function in itself so that algorithm convergence with probability 1 is in globally optimal solution, and standard
Genetic algorithm then cannot.Therefore, the optimizing of registration parameter is carried out by clonal selection algorithm, obtained effect is more accurate.
Description of the drawings
Fig. 1 is the realization flow chart of the present invention；
Fig. 2 is the Yellow River estuary that the present invention obtained RadarSat-2 satellites at 2008 and 2009 with existing method
Image the first width interception image registration result comparison, wherein, Fig. 2 (a) be reference picture, Fig. 2 (b) be floating image, Fig. 2
(c) it is the image after present invention registration, Fig. 2 (d) is registrated image for DE algorithms, and Fig. 2 (e) is registrated image, Fig. 2 (f) for PSO algorithms
Image is registrated for GA algorithms；
Fig. 3 is the Yellow River estuary that the present invention obtained RadarSat-2 satellites at 2008 and 2009 with existing method
The evolution curve of the Average Mutual value of image the first width interception image；
Fig. 4 is the Yellow River estuary that the present invention obtained RadarSat-2 satellites at 2008 and 2009 with existing method
Image the second width interception image registration result comparison, wherein, Fig. 4 (a) be reference picture, Fig. 4 (b) be floating image, Fig. 4
(c) it is the image after present invention registration, Fig. 4 (d) is registrated image for DE algorithms, and Fig. 4 (e) is registrated image, Fig. 4 (f) for PSO algorithms
Image is registrated for GA algorithms；
Fig. 5 is the Yellow River estuary that the present invention obtained RadarSat-2 satellites at 2008 and 2009 with existing method
The evolution curve of the Average Mutual value of image the second width interception image.
Specific embodiment
With reference to Fig. 1, realization step of the invention is as follows：
Step 1, reference picture and floating image are inputted；
Step 2, antibody population initializes：Based on determining antigen, some antibody populations are randomly generated, A represents antibody collection,
Antibody collection A is by interim antibody collection ArWith memory antibody collection AmComposition, N=r+m, r represent interim antibody collection ArIn interim antibody number
Amount, m represent memory antibody collection AmIn memory antibody quantity, N is the total quantity of antibody in antibody collection A；
Step 3, reference picture imageR, B are represented with A and represents floating image imageS structure normalized mutual information functions MI
(A, B) is as object function：
Wherein, H (A) is the edge entropy of image A,pA(i) it is i-stage gray level in image A
The probability of appearance, M are the series of gray level in image A, and H (B) is the edge entropy of image B,pB
(j) be in image B j-th stage gray level occur probability, N be image B in gray level series,
The combination entropy between image A and image B, p (i, j) is i-stage gray level in image A, in image B j-th stage gray level this
To putting the joint probability that same position occurs simultaneously in the picture, andH (i, j) is image A and image B
Joint histogram, n are the numbers of image A and image B overlapping regions pixel, and wherein joint histogram is defined as follows：
H (i, j) (0≤i≤K-1,0≤j≤L-1, K and L are the ranges of gray value in two images) represents ash in image A
Degree grade is i, and gray level is the number of the pixel pair of j in image B；
Step 4, normalized mutual information function MI (A, B) is optimized using clonal selection algorithm, specifically included：
4a) calculate the value of the affinity of antibody, i.e. normalized mutual information function MI (A, B) of initial population；
4b) antibody selects：Affinity of antibody according to descending is arranged, the highest preceding k individual of affinity of antibody is chosen and makees
For interim antibody collection Ar；
4c) antibody cloning：According to the size of affinity, the highest individual of k affinity of selection is cloned；
4d) antibody variation：Mutation operation is done to the antibody after clone.The random number between one 0 to 1 is generated, if this
Number is less than mutation probability, then to xoldInto row variation：
xnew=x+ δ * λ
Wherein, xnewParameter after making a variation for variable x, δ=0.1* (xmax-xmin) to narrow the region of search factor, xmaxFor
The maximum value of variable x, xminFor the minimum value of variable x, λ is a random number between 0 to 1, for adjusting the step of x variations
It is long；
4e) form memory antibody collection：Recalculate the adaptation by antibody each in antibody population after above-mentioned mutation operation
The size of angle value i.e. affinity, if being higher than A by the affinity of antibody after above-mentioned mutation operationrIn corresponding variation it
The affinity of preceding antibody replaces original A with regard to the antibody after mutation operationrIn the antibody that does not make a variation, form memory antibody collection Am；
4f) antibody memory：The natural extinction of 5% B cell in artificial immune system bioselection is simulated, i.e., from antibody
Delete 5% minimum antibody of affinity in collection A, reinitialize generate 5% new antibodies, with newly generated 5% antibody,
Substitute the antibody that part affinity is minimum in antibody collection A.
Step 5, judge whether to meet evolution conditions：If iterations are not up to pre-defined evolutionary generation gene,
Jump to 4a) it continues to execute, otherwise, perform step 6；
Step 6, when iterations reach pre-defined evolutionary generation gene, the highest individual of affinity is optimal anti-
Body carries out spatial alternation to floating image imageS using the transformation parameter representated by optimal antibody, calculates and after output transform
Floating image and reference picture imageR between association relationship, by reference picture imageR and transformation after floating image into
Row registration, the image after finally output is registrated.
The effect of the present invention can be further illustrated by following emulation：
1. simulated conditions
This example is under 7 systems of Intel (R) Core (TM) 2Duo CPU 2.33GHz Windows, Matlab 2014a
On operation platform, the emulation experiment of the present invention and existing method are completed.
2. emulation experiment content
It chooses and is intercepted by two width of the Yellow River estuary image that RadarSat-2 satellites were obtained at 2008 and 2009 respectively
Image (size is 400*400) carries out Experimental comparison, and true registration parameter is unknown.In order to verify the property of the present invention
Can, the method for the present invention is compared analysis with existing several optimization algorithms, is DE (differential evolution algorithm), PSO respectively
(particle cluster algorithm) and GA (genetic algorithm).Parameter setting is as follows in experiment：In four kinds of algorithms, Population Size popsize is
350, iterations 30, in CSA algorithms of the present invention, β=0.16, mutation probability PmIn=0.1, DE algorithm, mutagenic factor FD
It is set as FD=0.7, crossover probability CR is set as CR=0.5, and in PSO algorithms, weights omega is set as linearly passing from 0.9 to 0.4
Subtract, and c1=c2=2.0 is set, in GA algorithms, mutation probability M is set as 0.05, and crossover probability is set as 0.8.The knot of registration
Fruit is evaluated by the size of similarity measurement normalized mutual information first, and value represents to match between 0 to 1, closer to 1
Quasi- effect is better.Secondly the result of registration quality is judged according to visual observation.Following CSA represent the present invention based on clone
The method for registering images of selection, DE represent the method for registering images based on differential evolution algorithm, and PSO is represented based on particle cluster algorithm
Method for registering images, GA represent the method for registering images based on genetic algorithm.
(c), (d), (e) and (f) of Fig. 2 is the experimental result picture of CSA, DE, PSO and GA to first group of data respectively, can
With, it is evident that image can be more accurately registrated by the solution acquired of CSA, in CSA operation result figures, top it is black
Coincidence that color part is completely seamless.And PSO registration effects are worst, it can be seen that be substantially misaligned.Fig. 3 show four kinds of optimization algorithms
First group of data is tested respectively, each algorithm is run 20 times, and the mutual information average value in 20 every generations is taken to be drawn as above-mentioned song
Line.By Fig. 3, it is apparent that CSA can obtain higher fitness value compared to other three kinds of optimization algorithms, so as to obtain
More accurate registration result.
As shown in Figure 3, from the point of view of for the evolution curve of first group of HUANGHE ESTUARY image, four kinds of algorithms registrations, CSA algorithms are the
A generation starts just to show high association relationship, and its association relationship is also maximum in final convergence.Although DE algorithms are restrained
It is early, but association relationship but very little, therefore registration result is also poor.The effect of GA and PSO is between CSA and DE, the two phase
Effect is preferable for DE, but still not as good as CSA, CSA is good for being showed when seeking mutual information it can be seen from curve
Stability and convergence.
Have recorded maximum, minimum peace that four kinds of algorithms are iterated image the fitness value of optimization in detail in table 1
The situation of change of mean value.It can be seen that：The maximum adaptation angle value of CSA (clonal selection algorithm) and average fitness value are above it
His three kinds of algorithms, secondly GA (genetic algorithm), followed by PSO (particle cluster algorithm), compare for, for the diagram, DE is (poor
Point evolution algorithm) effect it is relatively less preferable, be primarily due to remote sensing images with noise and differential evolution algorithm for making an uproar
Caused by acoustic sensing.From above-mentioned chart, we are this it appears that CSA (clonal selection algorithm) has very high adaptability and has
Effect property.
Table 1
| Algorithm | MI(best) | MI(worst) | MI(average) |
| CSA | 0.6314 | 0.4636 | 0.5770 |
| DE | 0.4241 | 0.1967 | 0.3703 |
| PSO | 0.5602 | 0.3316 | 0.4926 |
| GA | 0.6093 | 0.3546 | 0.5374 |
(c), (d), (e) and (f) of Fig. 4 is the experimental result picture of CSA, DE, PSO and GA to second group of data respectively, can
To find out, image can be more accurately registrated by the solution that CSA is acquired, in CSA operation result figures, upper right corner black line
Part seamless coincidence completely, and dark trapezoidal region is also accurately aligned in figure.DE, PSO and GA have it is various it is apparent not
Accurate representation, PSO and GA upper right corner black line part are not aligned completely.It is right respectively that Fig. 5 show four kinds of optimization algorithms
Third group data are tested, each algorithm is run 20 times, and the mutual information average value in 20 every generations is taken to be drawn as above-mentioned curve.By scheming
5 can be seen that CSA can obtain higher fitness value compared to other three kinds of optimization algorithms, more accurately match so as to obtain
Quasi- result.
As shown in Figure 5, from the point of view of for the evolution curve of second group of data HUANGHE ESTUARY image, four kinds of algorithm registrations, four kinds of algorithms
Convergence rate all quickly, and all very stable convergence, but final association relationship, that is, fitness value of CSA is slightly above remaining
Three kinds of algorithms, all algorithms are optimal value in or so 15 generations.From above-mentioned figure：CSA can not only Fast Convergent, and
The precision of image registration can also be improved while convergence rate is ensured.
The maximum of the fitness value that optimization is iterated to image of four kinds of algorithms, minimum and average are had recorded in table 2
The situation of change of value.It can be seen that：The maximum adaptation angle value of CSA (clonal selection algorithm) and average fitness value are above other
Three kinds of algorithms, secondly GA (genetic algorithm), followed by PSO (particle cluster algorithm) and DE (differential evolution algorithm), compare for,
This it appears that CSA (clonal selection algorithm) has higher adaptability and validity.
Table 2
| Algorithm | MI(best) | MI(worst) | MI(average) |
| CSA | 0.6178 | 0.4404 | 0.5936 |
| DE | 0.4172 | 0.2164 | 0.3870 |
| PSO | 0.5828 | 0.4537 | 0.5524 |
| GA | 0.5874 | 0.4216 | 0.5560 |
In short, image registration problem is regarded as the optimization problem of function by the present invention, using normalized association relationship as
Object function optimizes object function using based on immune clonal selection algorithm, conventional method can be overcome easily to be absorbed in
The shortcomings that local optimum, and from the point of view of the convergence curve of registration result figure and mutual information function, the present invention will be better than existing
Four kinds of control methods, this also demonstrates effectiveness of the invention and robustness.
Claims (4)
1. a kind of method for registering images based on real coding clonal selection algorithm, which is characterized in that the method includes following
Step：
(1) input resolution ratio identical reference picture imageR and floating image imageS, and judge that imageR and imageS are
No is gray level image, if it is not, then being converted into gray level image；
(2) antibody population initializes：Antibody population is randomly generated, Z represents antibody collection, and antibody collection Z is by interim antibody collection ArAnd memory
Antibody collection AmComposition, P=r+m, r represent interim antibody collection ArIn interim antibody levels, m represents memory antibody collection AmIn note
Recall antibody levels, P is the total quantity of antibody in antibody collection Z；
(3) reference picture imageR, B are represented with A and represents floating image imageS structure normalized mutual information function MI (A, B) works
For object function：
Wherein, H (A) is the edge entropy of image A,pA(i) it is that i-stage gray level occurs in image A
Probability, M be image A in gray level series, H (B) be image B edge entropy,pB(j)
It is the probability that j-th stage gray level occurs in image B, N is the series of gray level in image B,
The combination entropy between image A and image B, p (i, j) is i-stage gray level in image A, in image B j-th stage gray level this
To putting the joint probability that same position occurs simultaneously in image A and image B this two images, andh(i,
J) be image A and image B joint histogram, n is the number of image A and image B overlapping regions pixel；
(4) normalized mutual information function MI (A, B) is optimized using clonal selection algorithm, specifically included：
4a) calculate the value of the affinity of antibody, i.e. normalized mutual information function MI (A, B) of initial population；
4b) antibody selects：Affinity of antibody according to descending is arranged, the highest preceding k individual of affinity of antibody is chosen and is used as and face
Shi Kangti collection Ar；
4c) antibody cloning：According to the size of affinity, the highest individual of k affinity of selection is cloned, including：
Affinity is higher, and affinity serial number i is smaller, and the number being cloned is more, on the contrary, affinity is lower, affinity serial number i
It is bigger, it is cloned the total quantity N that number is fewer, and antibody is clonedcSuch as following formula：
Wherein, β is a multiplication factor, and P is the antibody total quantity in antibody collection Z, and the value range of affinity serial number i is arrived for 1
K, round () are an independent variable operators, are represented to nearest integer rounding, NcI-thCorresponding i-th anti-
The scale that body is cloned；
4d) antibody variation：Mutation operation is done to the antibody after clone, including：
The random number between one 0 to 1 is generated, if this number is less than preset mutation probability, mutation probability is 0.01
To between 0.1, then to the antibody after clone into row variation：
xnew=x+ δ * λ
Wherein, xnewParameter after making a variation for variable x, δ=0.1* (xmax-xmin) to narrow the region of search factor, xmaxFor variable x
Maximum value, xminFor the minimum value of variable x, λ is a random number in the range of 0 to 1 open interval, after adjusting clone
The step-length of antibody variation；
4e) form memory antibody collection：Recalculate the fitness value by antibody each in antibody population after above-mentioned mutation operation
The namely size of affinity, if being higher than A by the affinity of antibody after above-mentioned mutation operationrIn before corresponding variation
The affinity of antibody just replaces original A with the antibody after mutation operationrIn the antibody that does not make a variation, form memory antibody collection Am；
4f) antibody memory：5% minimum antibody of affinity is deleted from antibody collection Z, the new of the generation 5% that reinitializes resists
Body with newly generated 5% antibody, substitutes the minimum antibody of 5% affinity being deleted in antibody collection Z；
(5) judge whether to meet evolution conditions：If iterations are not up to pre-defined evolutionary generation gene, jump to
It 4a) continues to execute, otherwise, performs (6)；
(6) when iterations reach pre-defined evolutionary generation gene, the highest individual of affinity is optimal antibody, is utilized
Transformation parameter representated by optimal antibody carries out floating image imageS spatial alternation, the floating image after being converted, meter
The association relationship between floating image and reference picture imageR after simultaneously output transform is calculated, by reference picture imageR and transformation
Floating image afterwards is registrated, finally the image after output registration.
2. method for registering images according to claim 1, wherein, the antibody population that randomly generates described in step (2) includes
Following steps：
(2a) determines the range of affine transformation rotation angle, i.e., according to the size of input picture, determines affine transformation in X and Y side
The range shifted up；
(2b) in range, is randomly generating antibody population using real number coding method determined by step (2a).
3. method for registering images according to claim 1, wherein, joint histogram is defined as follows described in step (3)：
H (a, b) represents same position in image A and image B this two images, and gray level is a in image A, gray level in image B
The number of pixel pair for b, wherein, 0≤a≤K-1,0≤b≤L-1, K and L are the gray scale of pixel in image A and image B respectively
The range of grade.
4. method for registering images according to claim 1, wherein, k is equal with the quantity r of interim antibody collection, that is, takes k=r.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510229255.XA CN105184764B (en) | 2015-05-07 | 2015-05-07 | A kind of method for registering images based on real coding clonal selection algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510229255.XA CN105184764B (en) | 2015-05-07 | 2015-05-07 | A kind of method for registering images based on real coding clonal selection algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105184764A CN105184764A (en) | 2015-12-23 |
| CN105184764B true CN105184764B (en) | 2018-06-15 |
Family
ID=54906818
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510229255.XA Active CN105184764B (en) | 2015-05-07 | 2015-05-07 | A kind of method for registering images based on real coding clonal selection algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105184764B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107300562B (en) * | 2017-05-19 | 2020-06-16 | 温州大学 | X-ray nondestructive testing method for measuring contact distance of finished relay product |
| CN111797903B (en) * | 2020-06-12 | 2022-06-07 | 武汉大学 | Multi-mode remote sensing image registration method based on data-driven particle swarm optimization |
| CN116152316B (en) * | 2023-04-17 | 2023-07-07 | 山东省工业技术研究院 | Image registration method based on self-adaptive parameter particle swarm algorithm |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103091676A (en) * | 2013-01-22 | 2013-05-08 | 中国矿业大学 | Mining area surface subsidence synthetic aperture radar interferometry monitoring and calculating method |
| US8731334B2 (en) * | 2008-08-04 | 2014-05-20 | Siemens Aktiengesellschaft | Multilevel thresholding for mutual information based registration and image registration using a GPU |
-
2015
- 2015-05-07 CN CN201510229255.XA patent/CN105184764B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8731334B2 (en) * | 2008-08-04 | 2014-05-20 | Siemens Aktiengesellschaft | Multilevel thresholding for mutual information based registration and image registration using a GPU |
| CN103091676A (en) * | 2013-01-22 | 2013-05-08 | 中国矿业大学 | Mining area surface subsidence synthetic aperture radar interferometry monitoring and calculating method |
Non-Patent Citations (1)
| Title |
|---|
| 基于克隆选择算法的PET-CT医学图像融合的实现;李爽;《万方硕士学位论文数据库》;20091207;正文第9-15、30-45页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105184764A (en) | 2015-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110223324B (en) | Target tracking method of twin matching network based on robust feature representation | |
| Ji et al. | Surfacenet: An end-to-end 3d neural network for multiview stereopsis | |
| CN106780485B (en) | SAR image change detection method based on super-pixel segmentation and feature learning | |
| CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
| CN105930868B (en) | A kind of low resolution airport target detection method based on stratification enhancing study | |
| CN105809693B (en) | SAR image registration method based on deep neural network | |
| CN110263845B (en) | SAR image change detection method based on semi-supervised countermeasure depth network | |
| CN108510532A (en) | Optics and SAR image registration method based on depth convolution GAN | |
| CN104778464B (en) | A kind of clothes position finding and detection method based on depth convolutional neural networks | |
| CN104392223B (en) | Human posture recognition method in two-dimensional video image | |
| CN101980250B (en) | Method for identifying target based on dimension reduction local feature descriptor and hidden conditional random field | |
| CN108564606A (en) | Heterologous image block matching method based on image conversion | |
| CN106950276A (en) | A kind of inversion method of the defect of pipeline depth based on convolutional neural networks | |
| Liu et al. | Remote sensing image change detection based on information transmission and attention mechanism | |
| CN104392228A (en) | Unmanned aerial vehicle image target class detection method based on conditional random field model | |
| CN109766936A (en) | Image change detection method based on information transmitting and attention mechanism | |
| CN105184764B (en) | A kind of method for registering images based on real coding clonal selection algorithm | |
| Quan et al. | Using deep neural networks for synthetic aperture radar image registration | |
| CN104680151B (en) | A kind of panchromatic remote sensing image variation detection method of high-resolution for taking snow covering influence into account | |
| CN110555841B (en) | SAR image change detection method based on self-attention image fusion and DEC | |
| CN110349185A (en) | A kind of training method and device of RGBT target following model | |
| CN107180436A (en) | A kind of improved KAZE image matching algorithms | |
| CN107816987A (en) | A kind of method for recognising star map based on cobweb pattern and convolutional neural networks | |
| CN110334656A (en) | Multi-source Remote Sensing Images Clean water withdraw method and device based on information source probability weight | |
| CN110110694A (en) | A kind of vision SLAM closed loop detection method based on target detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |