An image matching method based on quadratic signature type

Abstract

The present invention discloses an image matching method based on quadratic signature, including: obtaining a correspondence set; obtaining a first cost function based on the feature points and the neighborhood points of the feature points in the first relation set and the second feature point set; converting the first cost function into a simplified second cost function; obtaining a quadratic signature distance according to the weight vector between the feature points and the neighborhood points and the similarity matrix; obtaining a first quantization distance according to the consistency of the neighborhood topology of the vector angle and length, and obtaining a second quantization distance according to the consistency measure of the structure by the quadratic signature distance; converting the second cost function into a third cost function based on the first quantization distance and the second quantization distance; converting the third cost function into a fourth cost function, and minimizing the simplified fourth cost function to obtain an optimal correspondence set. The present invention can more accurately measure the similarity of local structures, improve the ability to remove incorrect correspondences, and retain more reliable correspondences.

Description

Technical Field

[0001] The present invention belongs to the technical field of image matching, and particularly relates to an image matching method based on a quadratic signature type. Background Art

[0002] Image registration refers to matching two or more images from the same scene but taken at different shooting angles or times through a certain registration algorithm. Image matching techniques are mainly divided into two categories: one is the technique based on pixel gray information, and the other is the image registration technique based on feature points. Since feature points contain clear physical meanings and have good stability, image registration based on point features has been widely studied. Image registration based on point features mainly includes two key steps: establishing corresponding relationships and estimating spatial transformation. Establishing accurate corresponding relationships is the key method for feature-based image registration. Feature-based image registration algorithms can use the global structure information and local structure information of features for registration. For images with non-rigid deformations, local structures can be retained while undergoing non-rigid deformations. Therefore, by imposing geometric constraints on local structures, matching point pairs with inconsistent spatial neighborhood structures can be eliminated. Existing methods use a method of maintaining local structure consistency to remove incorrect corresponding relationships from known corresponding relationships. This method uses the intersection of neighborhood points and the displacement vectors (vector length and direction) between corresponding point pairs to describe the similarity of neighborhood structures, where the starting point and ending point of each vector correspond to the spatial positions of two feature points in two images. However, some outliers and their neighborhood points may still be consistent with inliers in terms of vector length and angle. In addition, due to the influence of noise and outliers, comparing the vector lengths and included angles between corresponding point pairs cannot accurately describe the similarity of two local structures.

[0003] The invention patent with the publication number CN109697692A discloses a feature matching method based on local structure similarity. The implementation steps of this method are as follows: performing feature extraction and initial matching on two images to be matched, establishing a neighborhood affine coefficient matrix of feature points, calculating the difference between the neighborhood affine coefficient matrices of the feature points associated with each match in the initial matching set; optimizing the neighborhood affine coefficient matrix to obtain the degree of local structure difference, setting a comparison threshold according to the local structure difference values of the feature points associated with each match, and determining the final feature matching pairs as the matching relationship results of the images to be matched. Since this method uses the neighborhood affine coefficient matrix as a consistency constraint, and some outliers and their neighborhood points may still be consistent with inliers in terms of vector length and angle, affected by noise and outliers, this method cannot accurately describe the similarity of two local structures by comparing the vector lengths and included angles between corresponding point pairs, and has a poor ability to remove incorrect corresponding relationships.

[0004] When the above - mentioned scheme performs vector consistency constraint, the vector length and angle of the corresponding point pairs are used as measures. There may still be consistency in the vector length and angle between some outliers and their neighboring points and inliers. In addition, due to the influence of noise and outliers, the existing technology cannot accurately describe the problem of the similarity between two local structures, and has poor ability to remove incorrect correspondence relationships. Summary of the Invention

[0005] In order to solve the above - mentioned problems existing in the prior art, the present invention provides an image matching method based on quadratic signature type. The technical problems to be solved by the present invention are realized through the following technical solutions:

[0006] An image matching method based on quadratic signature type, the image matching method includes:

[0007] Step 1: Extract the first feature point set and the second feature point set from the first image and the second image respectively, and remove the mismatched point pairs in the first feature point set and the second feature point set to obtain a correspondence set. The correspondence set is composed of the mutually - matched feature point pairs (x i , y i ) in the first feature point set and the second feature point set, where x i is the feature point in the first feature point set, and y i is the feature point in the second feature point set;

[0008] Step 2: Determine the neighborhood points of each feature point in the first relationship set and the second feature point set, and obtain the first cost function based on the feature points and the neighborhood points of the feature points in the first relationship set and the second feature point set;

[0009] Step 3: Based on the distance between the feature point pairs under the local structure consistency constraint and the association between the feature point pairs and the binary vector P, convert the first cost function into a simplified second cost function;

[0010] Step 4: Establish a similarity matrix according to the expected similarity between two neighborhood points, and obtain the quadratic signature - type distance according to the weight vector between the feature point and the neighborhood points and the similarity matrix;

[0011] Step 5: Based on the neighborhood topological structure consistency constraint, obtain the first quantization distance according to the consistency of the neighborhood topological structure of the vector angle and length, and obtain the second quantization distance according to the consistency measurement of the structure by the quadratic signature - type distance;

[0012] Step 6: Based on the first quantization distance and the second quantization distance, convert the simplified second cost function into a third cost function;

[0013] Step 7: Convert the third cost function into a fourth cost function in a multi - scale neighborhood, and minimize the simplified fourth cost function to obtain the optimal correspondence set.

[0014] In one embodiment of the present invention, the first cost function is expressed as:

[0015]

[0016] where d(x i , x j ) represents the Euclidean distance between feature point x i and feature point x j , d(y i , y j ) represents the Euclidean distance between feature point y i and feature point y j , y j represents the neighborhood point of feature point y i , represents the set of neighborhood points of feature point x i , represents the set of neighborhood points of feature point y i , Ι represents the set of unknown correspondence relations, |·| represents the number of elements in the set, λ represents the coefficient for balancing the two terms and λ > 0, K represents the number of neighborhood points, and N represents the number of matching feature points in the correspondence relation set.

[0017] In one embodiment of the present invention, step 3 includes:

[0018] Step 3.1, Quantify the distance between a feature point and its neighborhood points based on the local structure consistency constraint, and the quantified distance is expressed as:

[0019]

[0020]

[0021] Step 3.2, Associate the correspondence of the feature point pair (x i , y i ) with the binary vector P, where the element p i ∈{0, 1} in the binary vector P represents the matching correctness of the i-th feature point pair (x i , y i ).

[0022] Step 3.3, Based on the quantified distance and the feature point pair (x i , y i ) associated with the binary vector P, convert the first cost function into a second cost function, and the second cost function is expressed as:

[0023]

[0024] where:

[0025]

[0026] where n i represents the number of elements common to the neighborhood point set and the neighborhood point set ;

[0027] Step 3.4. Convert the second cost function obtained in Step 3.3 into a simplified second cost function, and the simplified second cost function is expressed as:

[0028]

[0029] where N represents the number of matching feature points in the correspondence set.

[0030] In an embodiment of the present invention, Step 4 includes:

[0031] Step 4.1. Determine the first weight vector and the second weight vector based on the inverse of the distance from the neighborhood point to the corresponding feature point. Among them, the first weight vector is The second weight vector is represents the weight vector of the a-th neighborhood point and the feature point x i ; represents the weight vector of the b-th neighborhood point and the feature point y j ;

[0032] Step 4.2. Establish a similarity matrix i h j according to the expected similarity between the neighborhood points of the feature point x h a,b is expressed as:

[0033]

[0034] where represents the matrix space, with 2K rows and 2K columns;

[0035] Step 4.3. Obtain the quadratic signature type distance according to the connection weight vector and the similarity matrix composed of the first weight vector and the second weight vector. The quadratic signature type distance is expressed as:

[0036]

[0037] where T represents the transpose.

[0038] In an embodiment of the present invention, Step 5 includes:

[0039] Step 5.1. Obtain a consistency result based on the consistency of the neighborhood topological structure in terms of vector angle and length. The calculation formula for the consistency result is:

[0040]

[0041] where f(n i , n j ) ∈ [-1, 1], n i represents the vector of the feature point pair, n j represents the vector of the corresponding neighborhood point pair of the feature point pair, and (·, ·) represents the inner product;

[0042] Step 5.2. Obtain a first quantization distance based on the relationship between the consistency result and the first threshold. The first quantization distance is expressed as:

[0043]

[0044] where d(n i , n j ) represents the first quantization distance, n i represents the vector of the feature point pair, n j represents the vector of the corresponding neighborhood point pair of the feature point pair, and τ1 represents the first threshold;

[0045] Step 5.3. Obtain a neighborhood topological structure consistency result based on the consistency measurement of the quadratic signature - type distance for the structure. The calculation formula for the neighborhood topological structure consistency result is:

[0046]

[0047] where represents the quadratic signature - type distance;

[0048] Step 5.4. Obtain a second quantization distance based on the relationship between the neighborhood topological structure consistency result and the second threshold. The second quantization distance is expressed as:

[0049]

[0050] where d Si (x i , y i ) represents the second quantization distance, and τ2 represents the second threshold.

[0051] In an embodiment of the present invention, the third cost function is expressed as:

[0052]

[0053] In an embodiment of the present invention, the said step 7 includes:

[0054] Step 7.1. Convert the third cost function into a fourth cost function in a multi-scale neighborhood. The fourth cost function is expressed as:

[0055]

[0056] where K l represents the number of neighborhood points, and L represents the number of different numbers of selected neighborhood points;

[0057] Step 7.2. Combine the terms containing p i in the fourth cost function to obtain a simplified fourth cost function. The simplified fourth cost function is expressed as:

[0058]

[0059] where:

[0060]

[0061] Step 7.3. Minimize the simplified fourth cost function to obtain an optimal correspondence set. The optimal solution of p i is:

[0062]

[0063] The optimal correspondence set is:

[0064] Ι * ={i|p i =1, i = 1, …, N}

[0065] where Ι * represents the optimal correspondence set.

[0066] Advantages of the present invention:

[0067] The image matching method based on the quadratic signature type proposed by the present invention, while solving the problems of outliers and noise in the prior art, maintains the topological structure of the point set and is not affected by the positional relationship between the two point sets. Compared with the existing methods, this method can more accurately measure the similarity of local structures, improve the ability to remove incorrect correspondences, and retain more reliable correspondences. Brief description of the drawings

[0068] Figure 1 is a schematic flowchart of an image matching method based on the quadratic signature type provided by an embodiment of the present invention;

[0069] Figure 2 is a schematic flowchart of another image matching method based on the quadratic signature type provided by an embodiment of the present invention. Detailed implementation manners

[0070] The present invention will be further described in detail below in conjunction with specific embodiments, but the implementation manners of the present invention are not limited thereto.

[0071] Embodiment 1

[0072] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of an image matching method based on secondary signature provided by an embodiment of the present invention, Figure 2 and which is a schematic flowchart of another image matching method based on secondary signature provided by an embodiment of the present invention. The present invention provides an image matching method based on secondary signature, which is used to solve the problem in the prior art that due to the influence of outliers and noise, the similarity of two local structures cannot be accurately described. The image matching method based on secondary signature proposed by the present invention includes steps 1 - 7, where:

[0073] Step 1: Respectively extract the first feature point set X p×2 =(x1,...,x p ) T and the second feature point set Y q×2 =(y1,...,y q ) T from the first image and the second image, and remove the mismatched point pairs in the first feature point set and the second feature point set to obtain a correspondence set, where the correspondence set is composed of the mutually matching feature point pairs (x i ,y i ) in the first feature point set and the second feature point set. Among them, x i is a feature point in the first feature point set, yi is a feature point in the second feature point set, p is the number of feature points in the first feature point set, q is the number of feature points in the second feature point set, and T represents transpose.

[0074] Specifically, first, feature points in the first image and the second image are extracted respectively by the Scale-invariant feature transform (SIFT) method. All the feature points extracted from the first image form the first feature point set, and all the feature points extracted from the second image form the second feature point set. After extracting feature points using SIFT, the Euclidean distance of the key point feature vectors is used as the similarity determination metric for feature points in the two images. The similarity determination method is, for example: take a certain feature point in the first image, and find the two feature points in the second image that are closest to the feature point in the first image by traversing. Among these two feature points, if the closest distance divided by the second-closest distance is less than a preset threshold c1, then it is determined that the feature point in the first image and the feature point in the second image that is closest to the feature point in the first image are a pair of matching points (i.e., a feature point pair). Therefore, the non-matching point pairs between the first feature point set and the second feature point set are removed, and the initial matching correspondence set is determined. Where N is the number of matching feature points in the correspondence set.

[0075] Where the first image and the second image are two similar images or images from the same scene.

[0076] Step 2: Determine the neighborhood points of each feature point in the first relationship set and the second feature point set, so as to obtain a first cost function based on the feature points and the neighborhood points of the feature points in the first relationship set and the second feature point set.

[0077] Specifically, first, a threshold c2 is set, the distances between a certain feature point in the first relationship set and other feature points are calculated, and the feature points with distances less than the threshold c2 are used as the neighborhood points of this feature point. Similarly, the neighborhood points of each feature point in the second feature point set can be obtained.

[0078] Therefore, after determining the neighborhood points of each feature point, a first cost function that maintains the consistency of the local spatial structure is constructed. The first cost function is expressed as:

[0079]

[0080] Where d(x i ,x j ) represents the Euclidean distance between feature point xi and feature point x j , d(y i ,y j ) represents the Euclidean distance between feature point y i and feature point y j , y j represents the neighborhood point of feature point yi, represents the set of neighborhood points of feature point x i . Denote the set of neighborhood points of the feature point yi, Ι denote the set of unknown corresponding relationships, |·| denote the number of elements in the set, λ denote the coefficient for balancing the two terms and λ > 0, and K denote the number of neighborhood points.

[0081] Step 3: Based on the distance between feature point pairs under the local structure consistency constraint and the association between the feature point pairs and the binary vector P, convert the first cost function into a simplified second cost function.

[0082] Step 3.1: Quantify the distance between a feature point and its neighborhood points based on the local structure consistency constraint. The quantified distance is expressed as:

[0083]

[0084]

[0085] Step 3.2: Associate the corresponding relationship of the feature point pair (xi, yi) with the binary vector P. Among them, the element pi in the binary vector P ∈ {0, 1} represents the matching correctness of the i-th feature point pair (xi, yi), and p i = 1 indicates a correct match (inlier), and p i = 0 indicates an incorrect match (outlier).

[0086] Step 3.3: Based on the quantified distance and the feature point pair (x i , y i ) associated with the binary vector P, convert the first cost function into a second cost function. The second cost function is expressed as:

[0087]

[0088] Where:

[0089]

[0090] Among them, count(·) represents calculating the number of elements in the set.

[0091] Similarly, it can be obtained that:

[0092]

[0093] Where, n i represents the number of common elements in the neighborhood point set and the neighborhood point set in total.

[0094] Step 3.4: Convert the second cost function obtained in Step 3.3 into a simplified second cost function. The simplified second cost function is expressed as:

[0095]

[0096] Among them, N represents the number of matching feature points in the correspondence set.

[0097] Step 4: Establish a similarity matrix according to the expected similarity between two neighborhood points, and obtain the quadratic signature distance according to the weight vector between the feature point and the neighborhood point and the similarity matrix.

[0098] Step 4.1: Determine the first weight vector and the second weight vector based on the inverse ratio of the distance from the neighborhood point to the corresponding feature point. Among them, the first weight vector is The second weight vector is indicating the weight vector of the a-th neighborhood point and the feature point x i and indicating the weight vector of the b-th neighborhood point and the feature point y j and

[0099] Specifically, given the neighborhood point set i of the feature point x and the neighborhood point i of the feature point y where x a and y b are respectively the a-th neighborhood point of the feature point x i and the b-th neighborhood point of the feature point y i , the Gaussian similarity function is where d(x a , y b ) represents the Gaussian distance between the neighborhood point x a and the neighborhood point y b .

[0100] Determine the first weight vector and the second weight vector which consists of the inverse ratio of the distance from the neighborhood point to the corresponding feature point, is the inverse ratio of the distance between the feature point x i and the a-th neighborhood point, is the inverse ratio of the distance between the feature point y j and the b-th neighborhood point, 0 ≤ a, b ≤ K, and the connection weight vector is

[0101] Step 4.2: Establish a similarity matrix i h j according to the expected similarity between the neighborhood points of the feature point x h a,b is expressed as:

[0102]

[0103] Among them, represents a matrix space with 2K rows and 2K columns.

[0104] Step 4.3: Obtain the quadratic signature form distance according to the connection weight vector and the similarity matrix composed of the first weight vector and the second weight vector. The quadratic signature form distance is expressed as:

[0105]

[0106] where T represents the transpose.

[0107] Step 5: Based on the neighborhood topological structure consistency constraint, obtain the first quantization distance according to the consistency of the neighborhood topological structure of the vector angle and length, and obtain the second quantization distance according to the consistency measurement of the quadratic signature form distance for the structure.

[0108] Step 5.1: Obtain the consistency result based on the consistency of the neighborhood topological structure of the vector angle and length. The calculation formula for the consistency result is:

[0109]

[0110] where f(n i , n j ) ∈ [-1, 1]. The larger the value of f(n i , n j ), the higher the consistency of the two neighborhood structures. n i represents the vector of the feature point pair, n j represents the vector of the corresponding neighborhood point pair of the feature point pair, (·, ·) represents the inner product, and the cosine similarity represents the consistency of the vector angle.

[0111] Step 5.2: Considering the error interference between images, set the first threshold. Then, obtain the first quantization distance according to the relationship between the consistency result and the first threshold. The first quantization distance is expressed as:

[0112]

[0113] where d(n i , n j ) represents the first quantization distance, and τ1 represents the first threshold;

[0114] Step 5.3: Obtain the neighborhood topological structure consistency result according to the consistency measurement of the quadratic signature form distance for the structure. The calculation formula for the neighborhood topological structure consistency result is:

[0115]

[0116] where g(x i , y i)∈(0,1),g(x i ,y i )The larger the value, the higher the consistency of the two neighborhood structures. represents the quadratic signed distance.

[0117] Step 5.4: Set a second threshold, and obtain a second quantized distance based on the relationship between the neighborhood topological structure consistency result and the second threshold. The second quantized distance is expressed as:

[0118]

[0119] in, represents the second quantization distance, and τ2 represents the second threshold.

[0120] Step 6: Based on the first quantization distance and the second quantization distance, the simplified second cost function is converted into a third cost function. The third cost function is expressed as:

[0121]

[0122] Step 7: Convert the third cost function into a fourth cost function under a multi-scale neighborhood, and minimize the simplified fourth cost function to obtain an optimal set of corresponding relationships.

[0123] Step 7.1: To solve the problem that the optimal value of K nearest neighbors is not fixed, a multi-scale neighborhood representation is introduced. set up is the feature point x i K in Euclidean distance l The third cost function is converted into the fourth cost function under multi-scale neighborhood. The fourth cost function is expressed as:

[0124]

[0125] in, K controls the size of the multi-scale neighborhood, K l Represents the number of neighborhood points, and L represents the number of different numbers of selected neighborhood points.

[0126] Step 7.2: Replace the fourth cost function with p i The terms of are combined to obtain the simplified fourth cost function, which is expressed as:

[0127]

[0128] in:

[0129]

[0130] Step 7.3: Minimize the simplified fourth cost function to obtain the optimal correspondence set, thereby achieving the final image registration.

[0131] Specifically, for a given set of corresponding relations, its neighborhood relations are fixed and known, so t can be calculated. i The value of the i-th corresponding relationship (x i ,y i ) satisfies the consistency of its neighborhood topological structure constraints. If the corresponding relationship t i If the value is less than λ, the cost function decreases, otherwise the cost function increases. A correct match will make the cost function close to or equal to 0, while an incorrect match will make the cost function close to or equal to 1. Therefore, minimizing the simplified fourth cost function can remove incorrect matches and retain the correct correspondence. The optimal solution of P is set as:

[0132]

[0133] The optimal interior point set is:

[0134] Ι * ={i|p i =1,i=1,,N}

[0135] Among them, * represents the optimal set of correspondences.

[0136] Image registration is a bottleneck and difficult technology in the field of image processing and computer vision. The present invention effectively avoids the problem of the existing technology being affected by outliers and noise, can more accurately measure the similarity of local structures, improve the ability to remove erroneous correspondences, and retain more reliable correspondences. The present invention can be widely used in the fields of target recognition and classification, remote sensing image processing, medical image processing motion estimation, robot simultaneous positioning and reconstruction, etc.

[0137] The present invention uses quadratic signature distance to describe the consistency of neighborhood topological structure, which solves the problem that the vector angle and length used in the prior art are affected by noise and outliers and cannot accurately describe the similarity of local structures. It measures the similarity of local structures more accurately and effectively improves the ability to remove erroneous correspondences.

[0138] The quadratic signature type distance used in the present invention is insensitive to position information. Regardless of whether the centroids of two structures coincide, the quadratic signature type distance of correct structural matching is always the smallest, thereby improving the accuracy of matching.

[0139] The present invention simultaneously introduces vector-based angles and lengths and quadratic signature distance-based descriptions of neighborhood topological structure consistency constraints, further improving the ability to remove erroneous correspondences.

[0140] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc., mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification.

[0141] Although the present application has been described herein in connection with various embodiments, however, in the process of implementing the claimed present application, those skilled in the art can understand and achieve other variations of the disclosed embodiments by viewing the accompanying drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality. A single processor or other unit can implement several functions recited in the claims. Certain measures are recited in mutually different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0142] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. An image matching method based on quadratic signature type, characterized in that, The described image matching method includes: Step 1: Extract a first set of feature points and a second set of feature points from the first image and the second image respectively, and remove the mismatched point pairs in the first set of feature points and the second set of feature points to obtain a correspondence set. The correspondence set consists of the feature point pairs (x i , y i ) that match each other in the first set of feature points and the second set of feature points. Among them, x i is a feature point in the first set of feature points, and y i is a feature point in the second set of feature points; Step 2: Determine the neighborhood points of each feature point in the first relationship set and the second feature point set, and based on the feature points and the neighborhood points of the feature points in the first relationship set and the second feature point set, obtain the first cost function; Step 3: Based on the distance between feature point pairs under local structure consistency constraints and the association between the feature point pairs and the binary vector P, convert the first cost function into a simplified second cost function; Step 4: Establish a similarity matrix according to the expected similarity between two neighborhood points, and obtain the quadratic signature distance according to the weight vector between the feature point and the neighborhood point and the similarity matrix; Step 5: Based on the neighborhood topological structure consistency constraint, obtain the first quantization distance according to the consistency of the neighborhood topological structure of the vector angle and length, and obtain the second quantization distance according to the consistency measure of the structure by the quadratic signature distance; Step 6: Based on the first quantization distance and the second quantization distance, convert the simplified second cost function into a third cost function; Step 7: Convert the third cost function into a fourth cost function in a multi-scale neighborhood, and minimize the simplified fourth cost function to obtain the optimal corresponding relationship set.

2. The image matching method based on the secondary signature type according to claim 1, wherein, The first cost function is expressed as: where, d(x i , x j ) represents the Euclidean distance between feature point x i and feature point x j , d(y i , y j ) represents the Euclidean distance between feature point y i and feature point y j , y j represents the neighborhood point of feature point y i , represents the set of neighborhood points of feature point x i , represents the set of neighborhood points of feature point y i , Ι represents the set of unknown corresponding relationships, |·| represents the number of elements in the set, λ represents the coefficient for balancing the two terms and λ > 0, K represents the number of neighborhood points, and N represents the number of matching feature points in the set of corresponding relationships.

3. The image matching method based on the secondary signature type according to claim 2, wherein The described Step 3 includes: Step 3.1: Quantify the distance between a feature point and its neighborhood points based on local structure consistency constraints, and the quantified distance is expressed as: Step 3.2, associate the correspondence of the feature point pairs (x i , y i ) with the binary vector P, where the element p i ∈ {0, 1} in the binary vector P represents the matching correctness of the i-th feature point pair (x i , y i ). Step 3.

3. Based on the quantized distance sum and the feature point pairs (x i , y i ) associated with the binary vector P, convert the first cost function into a second cost function, and the second cost function is expressed as: Where: where n i represents the number of elements common to the neighborhood point set and the neighborhood point set; Step 3.4: Convert the second cost function obtained in Step 3.3 into a simplified second cost function, and the simplified second cost function is expressed as: Where N represents the number of matching feature points in the corresponding relationship set.

4. The image matching method based on the secondary signature type according to claim 3, characterized in that The described Step 4 includes: Step 4.1: Determine the first weight vector and the second weight vector based on the inverse ratio of the distance from the neighborhood points to the corresponding feature points, where the first weight vector is The second weight vector is denotes the weight vector of the a-th neighborhood point and the feature point x i ; denotes the weight vector of the b-th neighborhood point and the feature point y j ; Step 4.2: Establish a similarity matrix i based on the expected similarity between the neighborhood points of feature point x j and the neighborhood points of feature point y h a,b which is expressed as: Among them, represents a matrix space with 2K rows and 2K columns; Step 4.3: Obtain the quadratic signature distance according to the connection weight vector composed of the first weight vector and the second weight vector and the similarity matrix, and the quadratic signature distance is expressed as: Where T represents the transpose.

5. The image matching method based on the secondary signature type according to claim 4, wherein The described Step 5 includes: Step 5.1: Obtain the consistency result according to the consistency of the neighborhood topological structure of the vector angle and length, and the calculation formula of the consistency result is: Among them, f(n i ,n j ) ∈ [-1, 1], where n i represents the vector of the feature point pair, and n j represents the vector of the neighborhood point pair corresponding to the feature point pair, and (·, ·) represents the inner product; Step 5.2: Obtain the first quantization distance according to the relationship between the consistency result and the first threshold, and the first quantization distance is expressed as: where d(n i , n j ) represents the first quantization distance, n i represents the vector of the feature point pair, n j represents the vector of the neighborhood point pair corresponding to the feature point pair, and τ1 represents the first threshold; Step 5.3: Obtain the neighborhood topological structure consistency result according to the consistency measure of the structure by the quadratic signature distance, and the calculation formula of the neighborhood topological structure consistency result is: where g(x i , y i ) ∈ (0, 1), represents the quadratic signature type distance; Step 5.4: Obtain the second quantization distance according to the relationship between the neighborhood topological structure consistency result and the second threshold, and the second quantization distance is expressed as: Among them, represents the second quantization distance, and τ2 represents the second threshold.

6. The image matching method based on the secondary signature type according to claim 5, wherein The third cost function is expressed as:

7. The image matching method based on the secondary signature type according to claim 6, wherein The described Step 7 includes: Step 7.1: Convert the third cost function into a fourth cost function in a multi-scale neighborhood, and the fourth cost function is expressed as: Among them, K l represents the number of neighborhood points, and L represents the number of different selected neighborhood points; Step 7.2: Combine the terms containing p in the fourth cost function i to obtain the simplified fourth cost function, which is expressed as: Where: Step 7.3, minimize the simplified fourth cost function to obtain the optimal set of correspondence relations, p i The optimal solution of The optimal corresponding relationship set is: Ι * = {i | p i = 1, i = 1, …, N} Among them, Ι * represents the optimal set of corresponding relationships.

Images