A SIFT-based image registration method, device, system and storage medium

Abstract

The application discloses a kind of SIFT-based image registration method, device, system and storage medium, comprising: obtaining image to be registered and reference image, pre-processing image to be registered, construct Gaussian difference scale space;Obtain the extreme point in different scale space in the Gaussian difference scale space, according to the extreme point selects feature point, and generates feature descriptor for the feature point;Calculate the local information entropy of feature point field, according to the local information entropy and feature descriptor determine the final feature descriptor of feature point;According to genetic algorithm optimization feature point matching, generate matching point pair, and correct matching point pair, complete the image registration of image to be registered.The application optimizes matching efficiency and the speed of algorithm by reducing the number of feature points needing to be matched and the area needing to be matched, improves the search efficiency of SIFT algorithm and registration accuracy.

Description

[0001] Technical field

[0002] The present application relates to the technical field of image processing, more particularly, to a SIFT-based image registration method, device, system and storage medium. BACKGROUND

[0003] The SIFT algorithm has good stability and is widely used in feature-based image matching algorithms. The SIFT descriptor improves the image scale invariance and rotation invariance of the original algorithm, and the features have high uniqueness, which can maintain good matching robustness under different angles, different shooting distances, different affine transformations and different lighting conditions.

[0004] However, the traditional SIFT algorithm still has defects such as high computational complexity, long matching time, inability to calculate in real time, inability to meet the increasing demand for matching accuracy, and poor noise resistance, so it is a mainstream research direction to improve and optimize the SIFT algorithm in terms of reducing the complexity of the algorithm, improving the operation speed of the algorithm, improving the matching accuracy and noise resistance of the algorithm.

[0005] On the basis of existing theories, an improved algorithm is proposed to solve the problems of high computational complexity, long matching time and inability to calculate in real time in the SIFT algorithm. By reducing the number of feature points to be matched and the region to be matched and the dimension of the descriptor vector, the matching efficiency and the speed of the algorithm are improved. On the basis of existing theories, an improved algorithm is proposed to solve the problems of low accuracy and poor noise resistance in the SIFT algorithm. The improved algorithm is more suitable for the needs of today's society. SUMMARY

[0006] To solve the above technical problems, the present application provides a SIFT-based image registration method, device, system and storage medium.

[0007] In a first aspect, the present application provides a SIFT-based image registration method, comprising the following steps:

[0008] Obtain the to-be-registered image and the reference image, pre-process the to-be-registered image, and construct a Gaussian difference scale space;

[0009] Obtain extreme points in different scale spaces in the Gaussian difference scale space, select feature points according to the extreme points, and generate feature descriptors for the feature points;

[0010] Calculate the local information entropy of the feature point field, and determine the final feature descriptor of the feature point according to the local information entropy and the feature descriptor;

[0011] According to the genetic algorithm optimization feature point matching, the matching point pairs are generated, and the matching point pairs are corrected, so that the image registration of the to-be-registered images is completed.

[0012] Further, the extreme value points in different scale spaces in the Gaussian difference scale space are obtained, and the feature points are selected according to the extreme value points, specifically as follows:

[0013] The Gaussian difference scale space is simplified, and the first level scale space of the first group in the Gaussian difference scale space is removed, and the extreme value points are detected through the simplified Gaussian difference scale space;

[0014] By comparing the position and scale information of the to-be-detected point and the adjacent pixel points, the extreme value points are selected, and the contrast of the extreme value points is obtained;

[0015] A first contrast threshold is preset, the contrast of the extreme value points is compared with the first contrast threshold, and the extreme value points with a contrast greater than the first contrast threshold are selected as the to-be-selected feature points;

[0016] Meanwhile, a second contrast threshold is preset, the second contrast threshold is greater than the first contrast threshold, and the extreme value points with a contrast greater than the second contrast threshold are continuously saved as the to-be-selected feature points;

[0017] The Hessian matrix of the to-be-selected feature points is obtained, the unstable edge response points are removed, and the feature point set is generated according to the to-be-selected feature points.

[0018] Further, the feature descriptor is generated for the feature points, specifically as follows:

[0019] The scale information of the feature points in the feature point set is calculated, the Gaussian smoothing image is obtained, and the amplitude and amplitude angle of the image gradient of the Gaussian smoothing image in the circular neighborhood region of the feature points are obtained;

[0020] The amplitude and gradient direction of the pixels in the neighborhood of the feature points are counted through a histogram, and the peak value of the histogram is obtained as the main direction information of the feature points;

[0021] According to the main direction information of the feature points, the neighborhood of the feature points is rotated, the neighborhood of the feature points is divided into a plurality of sub-regions, the histogram statistics are generated in each sub-region to generate a feature vector, and the feature vector is dimensionally reduced to obtain a feature descriptor.

[0022] Further, the local information entropy of the feature point field is calculated, and the final feature descriptor of the feature points is determined according to the local information entropy and the feature descriptor, specifically as follows:

[0023] The local information entropy of each sub-region in the feature point field is obtained, a local information entropy threshold is preset, and the local information entropy of each sub-region is compared with the local information entropy threshold for judgment;

[0024] If less, the feature point in the corresponding sub-region is discarded, and if greater, a ratio information of a local information entropy of the corresponding sub-region to a total information entropy of the whole scale space is calculated;

[0025] Weight information is generated according to the ratio information, and the weight information is combined with the feature descriptor to generate a final feature descriptor of the feature point.

[0026] Further, the feature point matching is optimized according to the genetic algorithm, the matching point pairs are generated, and the matching point pairs are corrected to complete the image registration of the image to be registered.

[0027] The feature vector of the feature point is represented by chromosome coding according to the final feature descriptor of the feature point, and population parameters are initialized and set;

[0028] The Euclidean distance between the feature points is taken as the fitness function to perform iterative training, obtain the fitness information of the chromosome, and select, cross and mutate to generate the next generation of chromosome.

[0029] The chromosome with the minimum fitness information is decoded to obtain the information of the matching point, and the RANSAC algorithm is used for correction to eliminate the false matching points, thereby completing the image registration of the image to be registered.

[0030] In a second aspect, the present application provides a device for image registration based on SIFT, comprising:

[0031] An acquisition module is configured to acquire an image to be registered and a reference image, pre-process the image to be registered, and construct a Gaussian difference scale space.

[0032] A selection module is configured to acquire extreme points in different scale spaces in the Gaussian difference scale space, select feature points according to the extreme points, and generate feature descriptors for the feature points.

[0033] A final feature descriptor determination module is configured to calculate a local information entropy of a field of the feature points, and determine a final feature descriptor of the feature points according to the local information entropy and the feature descriptor.

[0034] An image registration module is configured to optimize feature point matching according to a genetic algorithm, generate matching point pairs, and correct the matching point pairs to complete the image registration of the image to be registered.

[0035] In a third aspect, the present application provides a system for image registration based on SIFT, which comprises a memory and a processor, the memory comprises a program of a method for image registration based on SIFT, and the program is executed by the processor to implement the following steps:

[0036] The image to be registered and the reference image are acquired, the image to be registered is preprocessed, and a Gaussian difference scale space is constructed;

[0037] Extreme points in different scale spaces in the Gaussian difference scale space are acquired, feature points are selected according to the extreme points, and feature descriptors are generated for the feature points;

[0038] Local information entropy of a feature point field is calculated, and the final feature descriptor of the feature point is determined according to the local information entropy and the feature descriptor;

[0039] Feature point matching is optimized according to a genetic algorithm, matched point pairs are generated, the matched point pairs are corrected, and image registration of the image to be registered is completed.

[0040] Further, the extreme points in the different scale spaces in the Gaussian difference scale space are acquired, and the feature points are selected according to the extreme points, specifically as follows:

[0041] The Gaussian difference scale space is simplified, the first level scale space of the first group in the Gaussian difference scale space is removed, and extreme point detection is performed through the simplified Gaussian difference scale space;

[0042] The position and scale information of the to-be-detected point and adjacent pixel points are compared, the extreme points are selected, and the contrast of the extreme points is acquired;

[0043] A first contrast threshold is preset, the contrast of the extreme points is compared and analyzed with the first contrast threshold, and the extreme points with a contrast greater than the first contrast threshold are taken as to-be-selected feature points;

[0044] Meanwhile, a second contrast threshold is preset, the second contrast threshold is greater than the first contrast threshold, and the extreme points with a contrast greater than the second contrast threshold are continuously saved as to-be-selected feature points;

[0045] A Hessian matrix of the to-be-selected feature points is acquired, unstable edge response points are removed, and a feature point set is generated according to the to-be-selected feature points.

[0046] Further, the feature point matching is optimized according to the genetic algorithm, the matched point pairs are generated, the matched point pairs are corrected, and the image registration of the image to be registered is completed, specifically as follows:

[0047] The final feature descriptor of the feature point is used for chromosome coding to represent a feature vector of the feature point, and population parameters are initialized and set;

[0048] The Euclidean distance between the feature points is used as a fitness function for iterative training, fitness information of the chromosome is acquired, and the next generation of chromosomes is generated through selection, crossover and mutation;

[0049] The chromosome with the minimum fitness information is decoded to obtain the matching point information, and the RANSAC algorithm is used for correction to eliminate the false matching points and perform image registration of the image to be registered.

[0050] In a fourth aspect, the present application provides a computer readable storage medium, characterized in that the computer readable storage medium comprises a SIFT-based image registration method program, and the SIFT-based image registration method program is executed by a processor to implement the steps of any one of the SIFT-based image registration method in the first aspect.

[0051] The present application discloses a SIFT-based image registration method, device, system and storage medium, comprising: obtaining an image to be registered and a reference image, pre-processing the image to be registered, and constructing a Gaussian difference scale space; obtaining extreme points in different scale spaces in the Gaussian difference scale space, selecting feature points according to the extreme points, and generating feature descriptors for the feature points; calculating the local information entropy of the feature point field, determining the final feature descriptor of the feature point according to the local information entropy and the feature descriptor; optimizing the feature point matching according to the genetic algorithm, generating a matching point pair, and correcting the matching point pair to complete the image registration of the image to be registered. The present application optimizes the matching efficiency and the speed of the algorithm by reducing the number of feature points that need to be matched and the area that needs to be matched, and improves the search efficiency and registration accuracy of the SIFT algorithm. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 A flowchart of the SIFT-based image registration method of the present application is shown;

[0053] Figure 2 A flowchart of the method for optimizing feature point matching according to the genetic algorithm of the present application is shown;

[0054] Figure 3 A structural schematic diagram of the device for optimizing feature point matching according to the genetic algorithm of the present application is shown;

[0055] Figure 4 A block diagram of the SIFT-based image registration system of the present application is shown. DETAILED DESCRIPTION

[0056] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0057] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods have not been described in detail in order to avoid obscuring the present application.

[0058] Referring to Figure 1 , Figure 1 A flow chart of a SIFT-based image registration method of the present application is shown.

[0059] As Figure 1 shown, the first aspect of the present application provides a SIFT-based image registration method, comprising:

[0060] S102, obtaining a to-be-registered image and a reference image, pre-processing the to-be-registered image, and constructing a Gaussian difference scale space;

[0061] S104, obtaining extreme points in different scale spaces in the Gaussian difference scale space, selecting feature points according to the extreme points, and generating feature descriptors for the feature points;

[0062] S106, calculating local information entropy of a feature point field, and determining a final feature descriptor of the feature point according to the local information entropy and the feature descriptor;

[0063] S108, optimizing feature point matching according to a genetic algorithm, generating a matching point pair, correcting the matching point pair, and completing image registration of the to-be-registered image.

[0064] It should be noted that due to the influence of factors such as shooting equipment and environment, the to-be-registered image is usually of low quality and is affected by additional factors such as noise. Before feature extraction, the reference image and the to-be-registered image are pre-processed through filtering and denoising methods such as gray scale transformation. In order to improve the efficiency and accuracy of detection, the first group of the first layer of the Gaussian difference scale space with the most useless feature points is skipped, the Gaussian difference scale space is simplified, the first layer scale space of the first group in the Gaussian difference scale space is removed, and the detection of extreme points is performed through the simplified Gaussian difference scale space. By comparing the position and scale information of the to-be-detected point with that of the adjacent pixel points, the extreme points are selected, and the contrast of the extreme points is obtained. A first contrast threshold is preset, the contrast of the extreme points is compared with the first contrast threshold, and the extreme points with a contrast greater than the first contrast threshold are selected as candidate feature points. At the same time, a second contrast threshold is preset, the second contrast threshold is greater than the first contrast threshold, and the extreme points with a contrast greater than the second contrast threshold are continuously saved as candidate feature points. The Hessian matrix of the candidate feature points is obtained, unstable edge response points are removed, and a feature point set is generated according to the candidate feature points.

[0065] It should be noted that the feature descriptor is generated for the feature point, specifically: the scale information of the feature point in the feature point set is calculated, a Gaussian smoothing image is obtained, and the amplitude and argument of the image gradient of the Gaussian smoothing image in the circular neighborhood region of the feature point are obtained; the amplitude and gradient direction of the pixels in the neighborhood of the feature point are counted by histogram, and the peak value of the histogram is obtained as the main direction information of the feature point; the horizontal axis of the histogram is the gradient direction angle, and the vertical axis is the gradient amplitude accumulation corresponding to the gradient direction angle, and the histogram divides the range of 0-360 degrees into 36 columns, one column every 10 degrees; the neighborhood of the feature point is rotated according to the main direction information of the feature point, the neighborhood of the feature point is divided into a plurality of sub-regions, histogram statistics are performed in each sub-region to generate a feature vector, and the feature vector is dimensionally reduced to obtain a feature descriptor.

[0066] It should be noted that the local information entropy of the feature point region is calculated, and the final feature descriptor of the feature point is determined according to the local information entropy and the feature descriptor, specifically: the local information entropy of each sub-region in the feature point region is obtained, a local information entropy threshold is preset, and the local information entropy of each sub-region is compared with the local information entropy threshold to determine; if it is less than, the feature point in the corresponding sub-region is discarded, if it is greater than, the ratio information of the local information entropy of the corresponding sub-region to the total information entropy of the entire scale space is calculated; the weight information is generated according to the ratio information, and the weight information and the feature descriptor are combined to generate the final feature descriptor of the feature point. The local information entropy is used to represent the degree of confusion of the pixel gray value distribution in the neighborhood of the feature point, and the calculation formula of the local information entropy is:

[0067]

[0068] Wherein, s i The local information entropy of the sub-region, p i Indicates the probability of the pixel value equal to the pixel value i in the feature point region.

[0069] It can be understood that the application discloses an image registration method based on SIFT, which comprises the following steps: obtaining a to-be-registered image and a reference image, preprocessing the to-be-registered image, and constructing a Gaussian difference scale space; obtaining extreme points in different scale spaces in the Gaussian difference scale space, selecting feature points according to the extreme points, and generating feature descriptors for the feature points; calculating the local information entropy of the feature point region, determining the final feature descriptor of the feature point according to the local information entropy and the feature descriptor; optimizing the feature point matching according to the genetic algorithm, generating a matching point pair, correcting the matching point pair, and completing the image registration of the to-be-registered image. The application optimizes the matching efficiency and the speed of the algorithm by reducing the number of feature points that need to be matched and the area that needs to be matched, and improves the search efficiency and registration accuracy of the SIFT algorithm.

[0070] Please refer toFigure 2 , Figure 2 The flow chart of the image registration method based on the genetic algorithm is shown.

[0071] According to the embodiment of the present application, the genetic algorithm is used to optimize the matching of the feature points, the matching point pairs are generated, and the matching point pairs are corrected to complete the image registration of the image to be registered.

[0072] In step S202, the feature vectors of the feature points are represented by the chromosome coding according to the final feature descriptors of the feature points, and the population parameters are initialized and set.

[0073] In step S204, the Euclidean distance between the feature points is used as the fitness function to perform iterative training to obtain the fitness information of the chromosomes, and the next generation of chromosomes is generated through selection, crossover and mutation.

[0074] In step S206, the chromosome with the minimum fitness information is selected to decode to obtain the information of the matching points, and the RANSAC algorithm is used to correct and eliminate the false matching points to perform the image registration of the image to be registered.

[0075] It should be noted that the matching of the feature points is determined by calculating the similarity through the Euclidean distance, a certain feature point is searched in the image to be registered, the feature points with the Euclidean distance meeting the preset condition are obtained, the matching point pairs are obtained, the genetic algorithm is used to optimize the matching of the feature points, the feature vectors of the feature points are represented by the length of the entity coding chromosome, the parameters are initialized, the population size is set to 20, the maximum number of iterations is set to 200, the Euclidean distance calculation formula of the feature points is used as the fitness function to calculate the fitness of the chromosomes in the population, the chromosomes with high fitness are removed, and the chromosomes are crossed, hybridized and mutated between each other in the iterative training, so that the feature points with the minimum Euclidean distance are quickly found.

[0076] Please refer to Figure 3 , Figure 3 is a schematic structural diagram of an image registration device based on SIFT in an embodiment of the present application, as shown in Figure 3 The image registration device 3 based on SIFT includes:

[0077] The acquisition module 301 is configured to acquire the image to be registered and the reference image, pre-process the image to be registered, and construct a Gaussian difference scale space.

[0078] The selection module 302 is configured to acquire extreme points in different scale spaces in the Gaussian difference scale space, select feature points according to the extreme points, and generate feature descriptors for the feature points.

[0079] A final feature descriptor determining module 303 is configured to calculate local information entropy of the feature point field, and determine a final feature descriptor of the feature point according to the local information entropy and the feature descriptor.

[0080] An image registration module 304 is configured to optimize feature point matching according to a genetic algorithm, generate a matching point pair, correct the matching point pair, and complete image registration of the image to be registered.

[0081] As to the device for SIFT-based image registration in the above embodiment, the specific manner in which each module performs the operation has been described in detail in the above embodiment of the related method, and thus will not be described in detail here.

[0082] Please refer to Figure 4 , Figure 4 A block diagram of a SIFT-based image registration system is shown.

[0083] The second aspect of the present application further provides a SIFT-based image registration system 4, which comprises a memory 401 and a processor 402, wherein the memory comprises a SIFT-based image registration method program, and the SIFT-based image registration method program is executed by the processor to implement the following steps:

[0084] Obtaining an image to be registered and a reference image, pre-processing the image to be registered, and constructing a Gaussian difference scale space;

[0085] Obtaining extreme points in different scale spaces in the Gaussian difference scale space, selecting feature points according to the extreme points, and generating feature descriptors for the feature points;

[0086] Calculating local information entropy of the feature point field, and determining a final feature descriptor of the feature point according to the local information entropy and the feature descriptor;

[0087] Optimizing feature point matching according to a genetic algorithm, generating a matching point pair, correcting the matching point pair, and completing image registration of the image to be registered.

[0088] It should be noted that due to the influence of factors such as shooting equipment and environment, the image to be registered is usually of low quality, and is affected by additional factors such as noise. Before feature extraction, the reference image and the image to be registered are preprocessed by filtering and denoising methods such as gray scale transformation. In order to improve the efficiency and accuracy of detection, the first group of the first layer of the Gaussian difference scale space with the most useless feature points is skipped, the Gaussian difference scale space is simplified, and the first level scale space in the Gaussian difference scale space is removed. The extreme points are detected through the simplified Gaussian difference scale space; by comparing the position and scale information of the to-be-detected point with the adjacent pixel points, the extreme points are selected, and the contrast of the extreme points is obtained; a first contrast threshold is preset, the contrast of the extreme points is compared with the first contrast threshold, and the extreme points with a contrast greater than the first contrast threshold are selected as candidate feature points; at the same time, a second contrast threshold is preset, the second contrast threshold is greater than the first contrast threshold, and the extreme points with a contrast greater than the second contrast threshold are continuously saved as candidate feature points; the Hessian matrix of the candidate feature points is obtained, the unstable edge response points are removed, and the feature point set is generated according to the candidate feature points.

[0089] It should be noted that a feature descriptor is generated for the feature points, specifically: the scale information of the feature points in the feature point set is calculated, a Gaussian smoothing image is obtained, and the amplitude and argument of the image gradient of the Gaussian smoothing image in the circular neighborhood region of the feature points are obtained; the amplitude and gradient direction of the pixels in the neighborhood of the feature points are counted through a histogram, and the peak value of the histogram is obtained as the main direction information of the feature points; the horizontal axis of the histogram is the gradient direction angle, and the vertical axis is the gradient amplitude accumulation corresponding to the gradient direction angle. The histogram divides the range of 0-360 degrees into 36 columns, with one column every 10 degrees; the neighborhood of the feature points is rotated according to the main direction information of the feature points, the neighborhood of the feature points is divided into a plurality of sub-regions, and a feature vector is generated by histogram statistics in each sub-region. The feature vector is dimensionally reduced to obtain a feature descriptor.

[0090] It should be noted that the local information entropy of the feature point region is calculated, and the final feature descriptor of the feature point is determined according to the local information entropy and the feature descriptor, specifically: the local information entropy of each sub-region in the feature point region is obtained, a local information entropy threshold is preset, and the local information entropy of each sub-region is compared with the local information entropy threshold; if it is less than, the feature points in the corresponding sub-region are discarded, and if it is greater than, the ratio information of the local information entropy of the corresponding sub-region to the total information entropy of the entire scale space is calculated; the weight information is generated according to the ratio information, and the final feature descriptor of the feature point is generated by combining the weight information and the feature descriptor. The local information entropy is used to represent the degree of confusion of the pixel gray value distribution in the neighborhood of the feature point, and the calculation formula of the local information entropy is:

[0091]

[0092] Wherein, s i Indicates the local information entropy of the sub-region, p i Indicates the probability of the pixel value equal to the feature point pixel value i in the feature point field.

[0093] According to the embodiment of the application, the feature point matching is optimized according to the genetic algorithm, the matching point pairs are generated, and the matching point pairs are corrected to complete the image registration of the image to be registered, specifically

[0094] The feature vector of the feature point is represented by the chromosome coding according to the final feature descriptor of the feature point, and the population parameters are initialized and set;

[0095] The Euclidean distance between the feature points is taken as the fitness function to perform iterative training, obtain the fitness information of the chromosome, and select, cross and mutate to generate the next generation of chromosomes;

[0096] The chromosome with the minimum fitness information is decoded to obtain the information of the matching points, and the RANSAC algorithm is used for correction to eliminate the false matching points, and the image registration of the image to be registered is performed.

[0097] It should be noted that the matching of the feature points is judged by calculating the similarity through the Euclidean distance, a certain feature point is searched in the image to be registered, the feature points with the Euclidean distance meeting the preset condition are obtained, the matching point pairs are obtained, the feature point matching is optimized according to the genetic algorithm, the feature vector of the feature point is represented by the length of the entity coding chromosome, the parameters are initialized, the population size is set to 20, the maximum number of iterations is set to 200, the Euclidean distance calculation formula of the feature points is taken as the fitness function, the fitness of the chromosomes in the population is calculated, the chromosomes with large fitness are removed, and the chromosomes are crossed, hybridized and mutated between each other in the iterative training, so that the feature points with the minimum Euclidean distance are quickly found.

[0098] The fourth aspect of the application further provides a computer readable storage medium, wherein the computer readable storage medium comprises a SIFT-based image registration method program, and when the SIFT-based image registration method program is executed by a processor, the steps of the SIFT-based image registration method according to any one of the above are realized.

[0099] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other manners. The described device embodiments are merely illustrative. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed coupling, or direct coupling or communication connection between the components can be indirect coupling or communication connection through some interfaces, devices, or units, and can be electrical, mechanical, or in other forms.

[0100] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units; they can be located in one place, or distributed on multiple network units; and some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0101] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0102] Those of ordinary skill in the art can understand that all or part of the steps of the above-described method embodiments can be completed by a program instructing related hardware, and the foregoing program can be stored in a computer readable storage medium, and when the program is executed, the steps of the method embodiments are executed; and the foregoing storage medium includes: mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and various media that can store program codes.

[0103] Alternatively, the integrated units of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes: mobile storage devices, ROM, RAM, magnetic disks or optical disks, and various media that can store program codes.

[0104] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A SIFT-based image registration method, characterized by, The method comprises the following steps: obtaining a to-be-registered image and a reference image, pre-processing the to-be-registered image, and constructing a Gaussian difference scale space; obtaining extreme points in different scale spaces in the Gaussian difference scale space, selecting feature points according to the extreme points, and generating feature descriptors for the feature points; calculating local information entropy of a feature point field, and determining a final feature descriptor of the feature point according to the local information entropy and the feature descriptor; the calculation of the local information entropy of the feature point field and the determination of the final feature descriptor of the feature point according to the local information entropy and the feature descriptor are specifically as follows: obtaining local information entropy of each sub-region in the feature point field, presetting a local information entropy threshold, and comparing the local information entropy of each sub-region with the local information entropy threshold; if the local information entropy is less than the local information entropy threshold, the feature points in the corresponding sub-region are discarded; if the local information entropy is greater than the local information entropy threshold, the ratio information of the local information entropy of the corresponding sub-region to the total information entropy of the entire scale space is calculated; and the weight information is generated according to the ratio information, and the final feature descriptor of the feature point is generated by combining the weight information and the feature descriptor; optimizing feature point matching according to a genetic algorithm, generating a matching point pair, correcting the matching point pair, and completing image registration of the to-be-registered image; the optimization of the feature point matching according to the genetic algorithm, the generation of the matching point pair, and the correction of the matching point pair to complete the image registration of the to-be-registered image are specifically as follows: the feature vectors of the feature points are represented by chromosome coding according to the final feature descriptors of the feature points, and population parameters are initialized and set; the Euclidean distance between the feature points is used as an adaptive function for iterative training to obtain adaptive information of the chromosome, and the next generation of chromosome is generated by selection, crossover and mutation; the adaptive information of the chromosome with the smallest adaptive information is decoded to obtain the information of the matching point, and the matching point is corrected according to the RANSAC algorithm to eliminate the false matching points, thereby completing the image registration of the to-be-registered image.

2. The SIFT-based image registration method of claim 1, wherein, The obtaining of the extreme points in different scale spaces in the Gaussian difference scale space and the selection of the feature points according to the extreme points are specifically as follows: skipping a first group of a first layer of the Gaussian difference scale space in which the most useless feature points are generated, simplifying the Gaussian difference scale space, removing the first group of the first layer scale space in the Gaussian difference scale space, and detecting the extreme points through the simplified Gaussian difference scale space; comparing the position and scale information of the to-be-detected point with those of the adjacent pixel points, selecting the extreme points, and obtaining the contrast of the extreme points; presetting a first contrast threshold, comparing and analyzing the contrast of the extreme points with the first contrast threshold, and taking the extreme points with a contrast greater than the first contrast threshold as the to-be-selected feature points; simultaneously, presetting a second contrast threshold, the second contrast threshold being greater than the first contrast threshold, and continuing to save the extreme points with a contrast greater than the second contrast threshold as the to-be-selected feature points; obtaining a Hessian matrix of the to-be-selected feature points, removing unstable edge response points, and generating a feature point set according to the to-be-selected feature points.

3. The SIFT-based image registration method of claim 1, wherein, generating feature descriptors for the feature points, specifically as follows: The scale information of the feature points in the feature point set is calculated to obtain a Gaussian smoothing image, and the amplitude and the argument of the image gradient of the Gaussian smoothing image in a circular neighborhood region of the feature points are obtained; The amplitude and the gradient direction of the pixels in the neighborhood of the feature points are counted by using a histogram, and the peak value of the histogram is obtained as the main direction information of the feature points; The neighborhood of the feature points is rotated according to the main direction information of the feature points, the neighborhood of the feature points is divided into a plurality of sub-regions, the histogram is counted in each sub-region to generate a feature vector, and the feature vector is reduced to obtain a feature descriptor.

4. An apparatus for SIFT-based image registration, characterized by It comprises: An acquisition module is configured to acquire a to-be-registered image and a reference image, pre-process the to-be-registered image, and construct a Gaussian difference scale space; A selection module is configured to acquire extreme points in different scale spaces in the Gaussian difference scale space, select feature points according to the extreme points, and generate feature descriptors for the feature points; A final feature descriptor determination module is configured to calculate the local information entropy of the feature point field, determine the final feature descriptor of the feature points according to the local information entropy and the feature descriptor, and specifically, acquire the local information entropy of each sub-region in the feature point field, compare the local information entropy of each sub-region with a preset local information entropy threshold, discard the feature points in the corresponding sub-region if the local information entropy is less than the local information entropy threshold, calculate the ratio information of the local information entropy of the corresponding sub-region to the total information entropy of the entire scale space if the local information entropy is greater than the local information entropy threshold, generate weight information according to the ratio information, and combine the weight information with the feature descriptor to generate the final feature descriptor of the feature points; An image registration module is configured to optimize feature point matching according to a genetic algorithm, generate matching point pairs, correct the matching point pairs, and complete the image registration of the to-be-registered image. Specifically, the feature vector of the feature points is represented by chromosome coding according to the final feature descriptor of the feature points, and population parameters are initialized and set; the Euclidean distance between the feature points is used as an adaptability function to perform iterative training, acquire the adaptability information of the chromosome, and select, cross and mutate to generate the next generation of chromosome; the information of the matching points is acquired by decoding the chromosome with the minimum adaptability information, and the RANSAC algorithm is used for correction to eliminate the false matching points and perform the image registration of the to-be-registered image.

5. A system for SIFT-based image registration, characterized in that, The system comprises a memory and a processor, the memory comprises a SIFT-based image registration method program, and the SIFT-based image registration method program is executed by the processor to implement the following steps: Acquire a to-be-registered image and a reference image, pre-process the to-be-registered image, and construct a Gaussian difference scale space; Acquire extreme points in different scale spaces in the Gaussian difference scale space, select feature points according to the extreme points, and generate feature descriptors for the feature points; The local information entropy of the feature point region is calculated, and the final feature descriptor of the feature point is determined according to the local information entropy and the feature descriptor; the local information entropy of the feature point region is calculated, and the final feature descriptor of the feature point is determined according to the local information entropy and the feature descriptor, specifically: the local information entropy of each sub-region in the feature point region is obtained, a preset local information entropy threshold is compared with the local information entropy of each sub-region, if the local information entropy of each sub-region is less than the local information entropy threshold, the feature points in the corresponding sub-region are discarded, if the local information entropy of each sub-region is greater than the local information entropy threshold, the ratio information of the local information entropy of the corresponding sub-region to the total information entropy of the entire scale space is calculated; the weight information is generated according to the ratio information, and the final feature descriptor of the feature point is generated by combining the weight information and the feature descriptor; The feature point matching is optimized according to the genetic algorithm, the matching point pairs are generated, the matching point pairs are corrected, and the image registration of the to-be-registered image is completed; the feature point matching is optimized according to the genetic algorithm, the matching point pairs are generated, the matching point pairs are corrected, and the image registration of the to-be-registered image is completed, specifically: the feature vector of the feature point is represented by chromosome coding according to the final feature descriptor of the feature point, and the population parameters are initialized and set; the Euclidean distance between the feature points is used as the fitness function for iterative training to obtain the fitness information of the chromosome, and the next generation of chromosomes is generated by selection, crossover and mutation; the information of the matching points is obtained by decoding the chromosome with the smallest fitness information, and the RANSAC algorithm is used for correction to eliminate the false matching points, and the image registration of the to-be-registered image is performed.

6. The SIFT-based image registration system of claim 5, wherein, The extreme points in different scale spaces in the Gaussian difference scale space are obtained, and the feature points are selected according to the extreme points, specifically: The first group of the first layer is skipped in the Gaussian difference scale space in which the most useless feature points are generated, the Gaussian difference scale space is simplified, the first group of the first level scale space in the Gaussian difference scale space is removed, and the extreme point detection is performed through the simplified Gaussian difference scale space; The position and scale information of the to-be-detected point and the adjacent pixel points are compared, the extreme points are selected, and the contrast of the extreme points is obtained; A first contrast threshold is preset, the contrast of the extreme points is compared with the first contrast threshold, and the extreme points with a contrast greater than the first contrast threshold are selected as the to-be-selected feature points; Meanwhile, a second contrast threshold is preset, the second contrast threshold is greater than the first contrast threshold, and the extreme points with a contrast greater than the second contrast threshold are continuously saved as the to-be-selected feature points; The Hessian matrix of the to-be-selected feature points is obtained, the unstable edge response points are removed, and the feature point set is generated according to the to-be-selected feature points.

7. The SIFT-based image registration system of claim 5, wherein, The feature point matching is optimized according to the genetic algorithm, the matching point pairs are generated, the matching point pairs are corrected, and the image registration of the to-be-registered image is completed, specifically The feature vector of the feature point is represented by chromosome coding according to the final feature descriptor of the feature point, and the population parameters are initialized and set; The Euclidean distance between the feature points is used as the fitness function for iterative training to obtain the fitness information of the chromosome, and the next generation of chromosomes is generated by selection, crossover and mutation; The chromosome with the minimum fitness information is decoded to obtain the matching point information, and the RANSAC algorithm is used for correction to eliminate the false matching points and perform image registration of the image to be registered.

8. A computer-readable storage medium, characterized in that: The computer readable storage medium comprises a SIFT-based image registration method program, and the SIFT-based image registration method program is executed by the processor to realize the steps of the SIFT-based image registration method in any one of claims 1 to 3.

Images