CN107067419A - The method for registering images of application enhancements gravitation search - Google Patents

Abstract

The invention discloses an image registration method using improved gravitational search. The present invention uses the improved gravitational search to optimize the parameters of image registration. In Modified Gravity Search, the parameters of the image registration are encoded as the individual's current location. In the search process, the basic operation operator of gravitational search is executed first, and then an individual is randomly selected and chaotic local search operation is performed on it, so as to improve the local search ability of the algorithm, accelerate the convergence speed, and improve the accuracy of image registration parameters. The invention can accelerate the convergence speed of image registration and improve the precision of image registration.

Description

Image Registration Method Using Improved Gravity Search

technical field

The invention relates to the field of image processing, in particular to an image registration method using improved gravitational search.

Background technique

Image registration is a common technique in the field of image processing, and it plays a very important role in the fields of remote sensing image analysis, medical image diagnosis, robot vision, image fusion and pattern recognition. The essence of image registration is given a reference image and an image to be registered, and it is required to find suitable spatial transformation parameters to perform spatial transformation on the image to be registered, so that the transformed image to be registered and a certain image in the reference image Some specific features achieve corresponding matching to a certain extent. It can be seen that the core of image registration is to optimize the parameters of image registration.

Since the image registration problem is a complex optimization problem in nature, traditional optimization algorithms based on the mathematical properties of the objective function are often difficult to solve effectively. Evolutionary algorithm is a heuristic optimization algorithm that simulates the evolution law of nature, and it shows great potential performance in solving complex optimization problems. Therefore, many researchers use evolutionary algorithms to optimize the parameters of image registration. For example, Sun Jun et al. invented a multi-resolution medical image registration method based on quantum behavior particle swarm algorithm (patent number: 200810019451.4); Zhang Xiujie et al. proposed a method to optimize image registration parameters by using harmonic quantum genetic algorithm Fang Method (Xiujie Zhang, Shiyong Li, Yi Shen, Shenmin Song. Application of Harmony Quantum Genetic Algorithm in Image Registration[J]. Systems Engineering and Electronic Technology, 2012,34(10):2152-2156); Invented by Jiao Licheng et al. A medical image registration method based on quantum evolutionary calculation and B-spline transformation (patent number: 201310516236.6); Xu Yingqiang et al. proposed an image registration algorithm based on genetic algorithm (Xu Yingqiang, Shi Qinghua, Qu Yongdong, Wang Cuizhi, Zhu Pan, Ye Yu.Research on an Image Registration Algorithm Based on Genetic Algorithm[J].Computer Science,2016,43(11A):229-232); Liu Hailing and Pei Lianqun proposed a medical registration algorithm based on improved genetic algorithm Image registration method (Liu Hailing, Pei Lianqun. Improvement of medical image registration method based on genetic algorithm [J]. Automation and Instrumentation, 2016, (09): 218-220).

The gravitational search algorithm is a newly proposed evolutionary algorithm, which shows superior performance in solving engineering optimization problems, and has been widely used in the field of image processing. However, the traditional gravitational search algorithm is prone to the disadvantages of slow convergence speed and insufficient solution accuracy when optimizing the parameters of image registration.

Contents of the invention

The object of the present invention is to provide an image registration method using improved gravitational search. To a certain extent, it overcomes the disadvantages of slow convergence speed and insufficient solution precision of the traditional gravity search algorithm when optimizing the parameters of image registration. The invention can accelerate the convergence speed of image registration and improve the accuracy of image registration.

The technical solution of the present invention: an image registration method using improved gravitational search, comprising the following steps:

Step 1, input the reference image RIM, and then input the image to be registered FIM;

Step 2, the user initializes the population size Popsize, the maximum number of evaluations MAX_FEs;

Step 3, let the current evolution algebra t=0, the current evaluation times FEs=0, and the number of image registration parameters D=3;

Step 4, setting the lower bound LB _j and upper bound UB _j of D optimization parameters, where the dimension subscript j=1,2,3;

Step 5, randomly initialize the population Wherein individual subscript i=1,2,...,Popsize; is the i-th individual in the population _Pt , and its random initialization formula is:

where j=1,2,3; Indicates the location of the i-th individual, and stores three parameter values for image registration, namely is the horizontal offset for image registration, is the vertical offset of the image registration, is the rotation angle for image registration; Indicates the speed of the i-th individual in each dimension, rand(0,1) is a random real number generation function between [0,1];

Step 6, calculate each individual in the population P _t The fitness value of , where the individual subscript i=1,2,...,Popsize, calculate the individual The method of the fitness value is: the The position of is decoded into the parameters of image registration, and the obtained image registration parameters are used to perform spatial transformation on the image FIM to obtain the image CIM, and the mutual information of the image RIM and the image CIM is used as the individual the fitness value;

Step 7, make the current evaluation times FEs=FEs+Popsize;

Step 8, save the best individual Best ^{t in the population P t} _;

Step 9, execute the basic operation operator of gravitational search;

Step 10, calculate the fitness value of each individual in the population P _t ;

Step 11, make the current evaluation times FEs=FEs+Popsize;

Step 12, randomly initialize the chaos factor tr, the specific steps are as follows:

Step 12.1, make the mixing times Num=300+200×rand(0,1);

Step 12.2, randomly generate a real number itr not equal to 0.25, 0.5 and 0.75 between [0,1];

Step 12.3, make the counter ki=0, and let the iteration factor tml=itr;

Step 12.4, if the counter ki is greater than Num, then go to step 13, otherwise go to step 12.5;

Step 12.5, make chaos factor tr=4.0×tml×(1-tml);

Step 12.6, let iteration factor tml=tr;

Step 12.7, make counter ki=ki+1, go to step 12.4;

Step 13, randomly select an individual and perform a chaotic local search operation on it, the specific steps are as follows:

Step 13.1, randomly generate a positive integer RT1 between [1, Popsize];

Step 13.2, randomly generate a positive integer RT2 not equal to RT1 between [1, Popsize], then make the iteration factor tml=tr;

Step 13.3, make chaos factor tr=4.0×tml×(1-tml);

Step 13.4, generate individual LU ^t according to the following formula:

The value of the reverse factor LK is a randomly generated real number between [0,1], the value of the hybridization factor IK is a randomly generated real number between [0,1], MEI ^t is the average value of the positions of all individuals in the population ;

Step 13.5, calculate the fitness value of the individual LU ^t , make the current evaluation times FEs=FEs+1;

Step 13.6, if the fitness value of individual LU ^t is better than If the fitness value is , go to step 13.7, otherwise go to step 14;

Step 13.7,

Step 13.8, go to step 13.2;

Step 14, make the current evolution algebra t=t+1;

Step 15, saving the best individual Best ^{t in the population P t} _;

Step 16, repeating steps 9 to 15 until the current evaluation times FEs reaches MAX_FEs, then the best individual Best ^t obtained during the execution process is decoded into image registration parameters to realize image registration.

The present invention uses the improved gravitational search to optimize the parameters of image registration. In Modified Gravity Search, the parameters of the image registration are encoded as the individual's current location. In the search process, the basic operation operator of gravitational search is executed first, and then an individual is randomly selected and chaotic local search operation is performed on it, so as to improve the local search ability of the algorithm, accelerate the convergence speed, and improve the accuracy of image registration parameters. The invention can accelerate the convergence speed of image registration and improve the precision of image registration.

Description of drawings

Figure 1 is a reference image.

Figure 2 is the image to be registered.

Fig. 3 is an image of the registration result of the application of the present invention.

detailed description

The technical solutions of the present invention will be further specifically described below through the embodiments and in conjunction with the accompanying drawings.

Example:

Step 1, input the reference image RIM as shown in Figure 1, and then input the image to be registered FIM as shown in Figure 2;

Step 2, the user initializes the population size Popsize=50, the maximum number of evaluations MAX_FEs=10000;

Step 3, let the current evolution algebra t=0, the current evaluation times FEs=0, and the number of image registration parameters D=3;

Step 4, setting the lower bound LB _j and upper bound UB _j of D optimization parameters, where the dimension subscript j=1,2,3;

Step 5, randomly initialize the population Wherein individual subscript i=1,2,...,Popsize; is the i-th individual in the population _Pt , and its random initialization formula is:

where j=1,2,3; Indicates the location of the i-th individual, and stores three parameter values for image registration, namely is the horizontal offset for image registration, is the vertical offset of the image registration, is the rotation angle for image registration; Indicates the speed of the i-th individual in each dimension, rand(0,1) is a random real number generation function between [0,1];

Step 6, calculate each individual in the population P _t The fitness value of , where the individual subscript i=1,2,...,Popsize, calculate the individual The method of the fitness value is: the The position of is decoded into the parameters of image registration, and the obtained image registration parameters are used to perform spatial transformation on the image FIM to obtain the image CIM, and the mutual information of the image RIM and the image CIM is used as the individual the fitness value;

Step 7, make the current evaluation times FEs=FEs+Popsize;

Step 8, save the best individual Best ^{t in the population P t} _;

Step 9, execute the basic operation operator of gravitational search;

Step 10, calculate the fitness value of each individual in the population P _t ;

Step 11, make the current evaluation times FEs=FEs+Popsize;

Step 12, randomly initialize the chaos factor tr, the specific steps are as follows:

Step 12.1, make the mixing times Num=300+200×rand(0,1);

Step 12.2, randomly generate a real number itr not equal to 0.25, 0.5 and 0.75 between [0,1];

Step 12.3, make the counter ki=0, and let the iteration factor tml=itr;

Step 12.4, if the counter ki is greater than Num, then go to step 13, otherwise go to step 12.5;

Step 12.5, make chaos factor tr=4.0×tml×(1-tml);

Step 12.6, let iteration factor tml=tr;

Step 12.7, make counter ki=ki+1, go to step 12.4;

Step 13, randomly select an individual and perform a chaotic local search operation on it, the specific steps are as follows:

Step 13.1, randomly generate a positive integer RT1 between [1, Popsize];

Step 13.2, randomly generate a positive integer RT2 not equal to RT1 between [1, Popsize], then make the iteration factor tml=tr;

Step 13.3, make chaos factor tr=4.0×tml×(1-tml);

Step 13.4, generate individual LU ^t according to the following formula:

The value of the reverse factor LK is a randomly generated real number between [0,1], the value of the hybridization factor IK is a randomly generated real number between [0,1], MEI ^t is the average value of the positions of all individuals in the population ;

Step 13.5, calculate the fitness value of the individual LU ^t , make the current evaluation times FEs=FEs+1;

Step 13.6, if the fitness value of individual LU ^t is better than If the fitness value is , go to step 13.7, otherwise go to step 14;

Step 13.7,

Step 13.8, go to step 13.2;

Step 14, make the current evolution algebra t=t+1;

Step 15, saving the best individual Best ^{t in the population P t} _;

Step 16, repeat steps 9 to 15 until the current evaluation times FEs reaches MAX_FEs, then decode the optimal individual Best ^t obtained during the execution process into image registration parameters, and perform spatial transformation on the image to be registered FIM to obtain the following: Figure 3 shows the image registration results.

The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which the present invention belongs can make various modifications or supplements to the described specific embodiments or adopt similar methods to replace them, but they will not deviate from the spirit of the present invention or go beyond the definition of the appended claims range.

Claims (1)

1. An image registration method using improved gravitational search, characterized in that, comprising the following steps: Step 1, input the reference image RIM, and then input the image to be registered FIM; Step 2, the user initializes the population size Popsize, the maximum number of evaluations MAX_FEs; Step 3, let the current evolution algebra t=0, the current evaluation times FEs=0, and the number of image registration parameters D=3; Step 4, setting the lower bound LB _j and upper bound UB _j of D optimization parameters, where the dimension subscript j=1,2,3; Step 5, randomly initialize the population Wherein individual subscript i=1,2,...,Popsize; is the i-th individual in the population _Pt , and its random initialization formula is: <mrow> <msubsup> <mi>A</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>1</mn> <mo>,</mo> <mi>j</mi> </mrow> <mi>t</mi> </msubsup> <mo>=</mo> <msub> <mi>LB</mi> <mi>j</mi> </msub> <mo>+</mo> <mi>r</mi> <mi>a</mi> <mi>n</mi> <mi>d</mi> <mrow> <mo>(</mo> <mn>0</mn> <mo>,</mo> <mn>1</mn> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <mrow> <mo>(</mo> <msub> <mi>UB</mi> <mi>j</mi> </msub> <mo>-</mo> <msub> <mi>LB</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>A</mi> <mrow> <mi>i</mi> <mo>,</mo> <mn>2</mn> <mo>,</mo> <mi>j</mi> </mrow> <mi>t</mi> </msubsup> <mo>=</mo> <mn>0</mn> </mrow> where j=1,2,3; Indicates the location of the i-th individual, and stores three parameter values for image registration, namely is the horizontal offset for image registration, is the vertical offset of the image registration, is the rotation angle for image registration; Indicates the speed of the i-th individual in each dimension, rand(0,1) is a random real number generation function between [0,1]; Step 6, calculate each individual in the population P _t The fitness value of , where the individual subscript i=1,2,...,Popsize, calculate the individual The method of the fitness value is: the The position of is decoded into the parameters of image registration, and the obtained image registration parameters are used to perform spatial transformation on the image FIM to obtain the image CIM, and the mutual information of the image RIM and the image CIM is used as the individual the fitness value; Step 7, make the current evaluation times FEs=FEs+Popsize; Step 8, save the best individual Best ^{t in the population P t} _; Step 9, execute the basic operation operator of gravitational search; Step 10, calculate the fitness value of each individual in the population P _t ; Step 11, make the current evaluation times FEs=FEs+Popsize; Step 12, randomly initialize the chaos factor tr, the specific steps are as follows: Step 12.1, make the mixing times Num=300+200×rand(0,1); Step 12.2, randomly generate a real number itr not equal to 0.25, 0.5 and 0.75 between [0,1]; Step 12.3, make the counter ki=0, and let the iteration factor tml=itr; Step 12.4, if the counter ki is greater than Num, then go to step 13, otherwise go to step 12.5; Step 12.5, make chaos factor tr=4.0×tml×(1-tml); Step 12.6, let iteration factor tml=tr; Step 12.7, make counter ki=ki+1, go to step 12.4; Step 13, randomly select an individual and perform a chaotic local search operation on it, the specific steps are as follows: Step 13.1, randomly generate a positive integer RT1 between [1, Popsize]; Step 13.2, randomly generate a positive integer RT2 not equal to RT1 between [1, Popsize], then make the iteration factor tml=tr; Step 13.3, make chaos factor tr=4.0×tml×(1-tml); Step 13.4, generate individual LU ^t according to the following formula: <mrow> <msubsup> <mi>LU</mi> <mn>1</mn> <mi>t</mi> </msubsup> <mo>=</mo> <mo>&amp;lsqb;</mo> <msubsup> <mi>A</mi> <mrow> <mi>R</mi> <mi>T</mi> <mn>2</mn> <mo>,</mo> <mn>1</mn> </mrow> <mi>t</mi> </msubsup> <mo>+</mo> <mi>t</mi> <mi>r</mi> <mo>&amp;times;</mo> <mrow> <mo>(</mo> <msubsup> <mi>Best</mi> <mn>1</mn> <mi>t</mi> </msubsup> <mo>-</mo> <msubsup> <mi>A</mi> <mrow> <mi>R</mi> <mi>T</mi> <mn>2</mn> <mo>,</mo> <mn>1</mn> </mrow> <mi>t</mi> </msubsup> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> <mo>&amp;times;</mo> <mi>I</mi> <mi>K</mi> <mo>+</mo> <mo>&amp;lsqb;</mo> <mrow> <mo>(</mo> <msubsup> <mi>Best</mi> <mn>1</mn> <mi>t</mi> </msubsup> <mo>+</mo> <msup> <mi>MEI</mi> <mi>t</mi> </msup> <mo>)</mo> </mrow> <mo>&amp;times;</mo> <mi>L</mi> <mi>K</mi> <mo>-</mo> <msubsup> <mi>A</mi> <mrow> <mi>R</mi> <mi>T</mi> <mn>2</mn> <mo>,</mo> <mn>1</mn> </mrow> <mi>t</mi> </msubsup> <mo>&amp;rsqb;</mo> <mo>&amp;times;</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <mi>I</mi> <mi>K</mi> <mo>)</mo> </mrow> </mrow> <mrow> <msubsup> <mi>LU</mi> <mn>2</mn> <mi>t</mi> </msubsup> <mo>=</mo> <msubsup> <mi>A</mi> <mrow> <mi>R</mi> <mi>T</mi> <mn>2</mn> <mo>,</mo> <mn>2</mn> </mrow> <mi>t</mi> </msubsup> </mrow> 1 The value of the reverse factor LK is a randomly generated real number between [0,1], the value of the hybridization factor IK is a randomly generated real number between [0,1], MEI ^t is the average value of the positions of all individuals in the population ; Step 13.5, calculate the fitness value of the individual LU ^t , make the current evaluation times FEs=FEs+1; Step 13.6, if the fitness value of individual LU ^t is better than If the fitness value is , go to step 13.7, otherwise go to step 14; Step 13.7, Step 13.8, go to step 13.2; Step 14, make the current evolution algebra t=t+1; Step 15, saving the best individual Best ^{t in the population P t} _; Step 16, repeating steps 9 to 15 until the current evaluation times FEs reaches MAX_FEs, then the best individual Best ^t obtained during the execution process is decoded into image registration parameters to realize image registration.