Feature descriptor reconstruction method and device based on multi-center sampling equalization filtering, medium and terminal

Abstract

This invention discloses a feature descriptor reconstruction method, apparatus, medium, and terminal based on multi-center sampling equalization filtering. The method includes: extracting key points from the original image, determining the key points as a central descriptor centroid, and determining six sampling points evenly distributed around the key points as sampling descriptor centroids; performing fixed sampling on the original image based on the central descriptor centroid and the sampling descriptor centroids to obtain grayscale change information of the descriptor centroid and grayscale change information of the sampling points around the descriptor centroid; filtering the grayscale change information to obtain information of several precise sampling points; encoding the descriptor centroid and surrounding sampling points based on the information of the precise sampling points to obtain a reconstructed binary feature descriptor. This invention improves the robustness of the feature descriptor and increases the extraction quantity and matching accuracy under harsh conditions such as illumination changes.

Description

Technical Field

[0001] This invention relates to the field of computer vision, and in particular to a method, apparatus, medium, and terminal for feature descriptor reconstruction based on multi-center sampling equalization filtering. Background Technology

[0002] Local image feature description and matching are core steps in computer vision technology, widely used in many computer vision applications, including 3D reconstruction, object recognition, and image registration. The purpose of local image feature description is to represent an input local image using a vector; this reconstructed representation vector is typically called a local image feature descriptor. By comparing the similarity / distance between the feature descriptors of different local images, local image matching can be performed to obtain matching relationships between local images corresponding to the same physical scene.

[0003] Feature points in an image consist of two parts: keypoints and feature descriptors. Feature descriptors are used to encode and describe the neighboring image patches of the keypoints. Reconstruction methods have evolved from directly using neighboring pixel values ​​to HOG histograms, and then to binary encoding. Binary descriptors such as BRIEF, ORB, and BRISK are fast to construct and have the advantages of fast matching and low memory consumption. However, because they do not consider the grayscale distribution characteristics of keypoints, the feature descriptors lack robustness in the matching process, resulting in fewer extracted features and lower matching accuracy under harsh conditions such as illumination changes. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this application is to provide a feature descriptor reconstruction method, device, medium and terminal based on multi-center sampling equalization filtering, which aims to solve the problem of how to enhance the robustness of feature descriptors and improve matching accuracy.

[0005] To address the aforementioned technical problems, the first aspect of this application provides a feature descriptor reconstruction method based on multi-center sampling equalization filtering, the method comprising:

[0006] Key points are extracted from the original image, and the key points are determined as a central descriptive centroid. Six sampling points centered on the key points and evenly distributed around the circumference are respectively determined as sampling descriptive centroids.

[0007] Based on the central descriptive centroid and the sampled descriptive centroid, the original image is sampled in a fixed manner to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid.

[0008] The grayscale change information is filtered to obtain information from several precise sampling points;

[0009] Based on the information from the precise sampling points, the descriptive centroid and surrounding sampling points are encoded to obtain a reconstructed binary feature descriptor.

[0010] As a further improved technical solution, the step of extracting key points from the original image, determining the key points as a central descriptive centroid, and determining six sampling points evenly distributed around the key points as sampling descriptive centroids includes:

[0011] Key points are extracted from the original image, and the initial feature descriptor of the original image is sampled in a circular distribution based on the key points to obtain key point information and sampling point information around the key points;

[0012] Based on the key point information and the sampling point information around the key point, the key points in the original image are extracted, and the key point is determined as a central descriptive centroid. Six sampling points centered on the key point and evenly distributed around the circumference are respectively determined as sampling descriptive centroids.

[0013] As a further improved technical solution, the fixed sampling of the original image based on the described centroid point and the sampled described centroid points to obtain the grayscale change information of the described centroid point and the grayscale change information of the sampled points around the described centroid point includes:

[0014] The original image is sampled using the description centroid based on the center and the sampled description centroid.

[0015] By comparing the pixel values ​​of the described centroid point with those of the sampling points around the centroid point, grayscale change information of the described centroid point and grayscale change information of the sampling points around the centroid point are obtained.

[0016] As a further improved technical solution, the filtering process of the grayscale change information to obtain information from several precise sampling points includes:

[0017] Based on the grayscale change information, the sampling points are weighted and summed using a weighted summation method to obtain the coordinate information and grayscale values ​​of several precise sampling points.

[0018] As a further improved technical solution, the step of encoding the descriptive centroid and surrounding sampling points based on the information of the precise sampling points to obtain the reconstructed binary feature descriptor includes:

[0019] The gray values ​​of the central descriptive centroid and the precise sampling points are compared to obtain a 64-dimensional binary string;

[0020] Draw a circle with each sampled centroid as its center, and use the circle as the image patch for each sampled centroid. The distance from the sampled centroid to the central centroid is used as the radius of the circle.

[0021] Each image block is sequentially sampled circumferentially to obtain grayscale values ​​of 32 pixels;

[0022] The description centroid of each sample is compared with the gray value of the corresponding 32 pixels in turn to obtain a 6×32-dimensional binary string;

[0023] The 6×32-dimensional binary string is concatenated with the 64-dimensional binary string to obtain the reconstructed 256-dimensional binary feature descriptor.

[0024] As a further improvement to the technical solution, the first formula used in the weighted summation method is as follows:

[0025] p1(x1,x2)=(A×I(x1,x2)+B×(I(x1-1,x2-1)+

[0026] I(x1-1,x2)+I(x1-1,x2+1)+

[0027] I(x1,x2-1)+I(x1,x2+1)+

[0028] I(x1+1,x2-1)+I(x1+1,x2)+

[0029] I(x1+1,x2+1))) / (A+8B)

[0030] The second formula used in the weighted summation method is as follows:

[0031] p2(x1,x2)=(A×I(x1,x2)+B×(I(x1+1,x2-1)+

[0032] I(x1+1,x2)+I(x1,x2-1))) / (A+4B)

[0033] In the second formula and the second formula, p1 and p2 are gray values, (x1, x2) are the coordinates of the sampling point, I is the gray value, and A and B are the assigned weights.

[0034] As a further improvement to the technical solution, the sampling formula used for the fixed sampling is as follows:

[0035] (xa) 2 +(yb) 2 =R,R=2.9f,4.9f,7.4f,10.8f

[0036] num∈[0,Nums i ), i∈[1,4]

[0037] P x =R×cosα

[0038] P y =R×sinα

[0039] Wherein, P in the sampling formula x P represents the x-coordinate of the circularly distributed sampling points. y Let R be the ordinate of the circularly distributed sampling points, f = 1.39, (a, b) be the coordinates of the keypoints, α be the angle of the sampling point, and num be the num-th sampling point of each circle. i It is the number of sampling points on the i-th circle, and its value increases linearly, Nums i =4i+6, i∈[1,4].

[0040] A second aspect of this application provides a feature descriptor matching device based on multi-center sampling equalization filtering, comprising:

[0041] The description center point selection module extracts key points from the original image, determines the key points as a central description center point, and determines six sampling points evenly distributed around the key points as sampling description center points.

[0042] The fixed sampling module performs fixed sampling on the original image based on the central descriptive centroid and the sampled descriptive centroid, to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid.

[0043] The filtering module filters the grayscale change information to obtain information from several precise sampling points;

[0044] The encoding module encodes the descriptive centroid and surrounding sampling points based on the information of the precise sampling points to obtain the reconstructed binary feature descriptor.

[0045] A third aspect of this application provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described above.

[0046] A fourth aspect of this application provides a terminal device, which includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor;

[0047] The communication bus enables communication between the processor and the memory;

[0048] When the processor executes the computer-readable program, it implements the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described above.

[0049] Beneficial Effects: Compared with existing technologies, the feature descriptor reconstruction method based on multi-center sampling equalization filtering of the present invention includes: extracting key points in the original image; determining the key points as a central descriptor centroid; determining six sampling points evenly distributed around the key points as sampling descriptor centroids; performing fixed sampling on the original image based on the central descriptor centroid and the sampling descriptor centroids to obtain grayscale change information of the descriptor centroid and grayscale change information of the sampling points around the descriptor centroid; filtering the grayscale change information to obtain information of several precise sampling points; encoding the descriptor centroid and the surrounding sampling points based on the information of the precise sampling points to obtain a reconstructed binary feature descriptor; the present invention improves the robustness of the feature descriptor and increases the extraction quantity and matching accuracy under harsh conditions such as illumination changes. Attached Figure Description

[0050] Figure 1 This is a flowchart of the feature descriptor reconstruction method based on multi-center sampling equalization filtering according to the present invention.

[0051] Figure 2 This is a structural schematic diagram of the terminal device provided by the present invention.

[0052] Figure 3 This is a structural block diagram of the device provided by the present invention.

[0053] Figure 4 This is a schematic diagram illustrating the seven descriptive center points of this invention.

[0054] Figure 5 This is a schematic diagram of the multi-center circular coding of the present invention.

[0055] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0056] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application.

[0058] The inventors discovered through research that the existing technology has the following problems:

[0059] (1) Feature points in an image consist of two parts: key points and feature descriptors. The feature descriptors are used to encode and describe the neighboring image blocks of the key points. Reconstruction methods have evolved from directly using the values ​​of neighboring pixels to HOG histograms, and then to binary encoding. Among them, binary descriptors such as BRIEF, ORB, and BRISK are fast to construct and have the advantages of fast matching and low storage consumption. However, because they do not consider the gray distribution characteristics of key points, the feature descriptors are not robust enough in the matching process. Under harsh conditions such as illumination changes, the number of extracted features is small and the matching accuracy is low.

[0060] like Figure 1 As shown in the figure, the feature descriptor reconstruction method based on multi-center sampling equalization filtering provided in this application includes the following steps:

[0061] S1, extract key points from the original image, determine the key points as a central descriptive centroid, and determine the six sampling points that are evenly distributed around the key points as sampling descriptive centroids.

[0062] Specifically, the coordinate information of key points in the original image is extracted, and the key point is determined as a central descriptive centroid. Six sampling points centered on the key point and evenly distributed around the circumference are determined as sampling descriptive centroids, resulting in a total of six sampling descriptive centroids. These six sampling descriptive centroids, plus the central descriptive centroid, make a total of seven descriptive centroids.

[0063] The step of extracting key points from the original image, determining the key points as a central descriptive centroid, and determining six sampling points evenly distributed around the key points as sampling descriptive centroids includes the following steps:

[0064] S101, extract key points from the original image, and perform circumferential distribution sampling on the initial feature descriptor of the original image based on the key points to obtain key point information and sampling point information around the key points;

[0065] S102, based on the key point information and the sampling point information around the key point, extract the key points in the original image, determine the key point as a central descriptive centroid, and determine the six sampling points evenly distributed around the key point as sampling descriptive centroids.

[0066] Specifically, circular sampling is sparser and more efficient than uniform sampling, and it can better reflect the relationship between the centroid and neighboring points. Circular sampling can obtain gray-level change information of key points and sampling points around the key points. The gray-level information of the neighborhood of key points is related. Traditional circular sampling algorithms compare the pixel values ​​of the center point with those of the surrounding sampling points pairwise, and concentrate the information of the key points on the center point. This is easily affected by the environment and has low robustness. In order to improve the robustness of feature descriptors, multiple centroids are added to the traditional sampling distribution, thereby reconstructing the single centroid of the traditional descriptor into a multi-centroid. Compared with the traditional sampling distribution, the fixed sampling increases the number of centroids from one to seven.

[0067] like Figure 4 As shown, firstly, key points are extracted from the original image. Based on these key points, the initial feature descriptor of the original image is sampled in a circular distribution to obtain key point information and sampling point information around the key points. Based on the key point information and the sampling point information around the key points, key points in the original image are extracted. The key point is determined as a central descriptor centroid. Six sampling points evenly distributed in a circle around the key point are determined as the sampling descriptor centroids, as shown below. Figure 4 As shown, the process of determining the six sampling descriptive centroids is as follows: among the several sampling points obtained by circumferential sampling, six sampling points are evenly distributed on the third circle formed by connecting the sampling points from the inside out with the key point as the center. The six sampling points are respectively determined as the sampling descriptive centroids to describe the information of the neighboring image blocks other than the central descriptive centroid.

[0068] S2, based on the central descriptive centroid and the sampled descriptive centroid, the original image is sampled in a fixed manner to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid.

[0069] Specifically, after determining a central descriptive centroid and six sampled descriptive centroids, the original image needs to be resampled to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid. In other words, the original image is resampled based on the seven descriptive centroids using a fixed sampling method.

[0070] The step of performing fixed sampling on the original image based on the centroid of the center and the sampled centroid to obtain the grayscale change information of the centroid and the grayscale change information of the sampling points around the centroid includes the following steps:

[0071] S201, the original image is fixedly sampled based on the description centroid point of the center and the sampled description centroid point;

[0072] S202, compare the pixel values ​​of the described centroid point with those of the sampling points around the centroid point to obtain the grayscale change information of the described centroid point and the grayscale change information of the sampling points around the centroid point.

[0073] The sampling formula used for the fixed sampling is as follows:

[0074] (xa) 2 +(yb) 2 =R,R=2.9f,4.9f,7.4f,10.8f

[0075] num∈[0,Nums i ), i∈[1,4]

[0076] P x =R×cosα

[0077] P y =R×sinα

[0078] Wherein, P in the sampling formula x P represents the x-coordinate of the circularly distributed sampling points. y Let R be the ordinate of the circularly distributed sampling points, f = 1.39, (a, b) be the coordinates of the keypoints, α be the angle of the sampling point, and num be the num-th sampling point of each circle. i It is the number of sampling points on the i-th circle, and its value increases linearly, Nums i =4i+6, i∈[1,4].

[0079] S3, The grayscale change information is filtered to obtain information from several precise sampling points;

[0080] To filter out noise points generated by fixed sampling, further filtering of the grayscale change information of the sampling points obtained by fixed sampling is required. The grayscale value after processing the grayscale change information of the sampling points is obtained by weighted summation. Weighted summation of the sampling points can not only solve the noise problem, but also add more neighborhood information in the grayscale value comparison process, thereby increasing the reliability of the descriptor.

[0081] The filtering process of the grayscale change information to obtain information from several precise sampling points includes:

[0082] Based on the grayscale change information, the sampling points are weighted and summed using a weighted summation method to obtain the coordinate information and grayscale values ​​of several precise sampling points.

[0083] The first formula used in the weighted summation method is as follows:

[0084] p1(x1,x2)=(A×I(x1,x2)+B×(I(x1-1,x2-1)+

[0085] I(x1-1,x2)+I(x1-1,x2+1)+

[0086] I(x1,x2-1)+I(x1,x2+1)+

[0087] I(x1+1,x2-1)+I(x1+1,x2)+

[0088] I(x1+1,x2+1))) / (A+8B)

[0089] The second formula used in the weighted summation method is as follows:

[0090] p2(x1,x2)=(A×I(x1,x2)+B×(I(x1+1,x2-1)+

[0091] I(x1+1,x2)+I(x1,x2-1))) / (A+4B)

[0092] In the second formula and the second formula, p1 and p2 are gray values, (x1, x2) are the coordinates of the sampling point, I is the gray value, and A and B are the assigned weights.

[0093] Using two weighted summation formulas simultaneously is to avoid sampling a single pixel multiple times in the neighborhood sampling weighting. If the sampling point during weighted summation is the centroid of the description, or the distance from the sampling point to the corresponding centroid of the description is greater than the set threshold of 8, then the first formula is used; otherwise, the second formula is used.

[0094] S4. Based on the information of the precise sampling points, the description centroid and surrounding sampling points are encoded to obtain the reconstructed binary feature descriptor.

[0095] like Figure 5As shown, after confirming the location of the centroid, it is necessary to perform multi-center circular encoding on the centroid. This means adding a circular distribution sampling for each centroid. The purpose is to obtain more information about the surrounding sampling points and the differences between the centroid and the centroid. This allows for better identification of the same feature point in different images. At the same time, it also obtains information about the centroid's neighborhood, which is the difference between the gray values ​​of the neighborhood and the gray values ​​of the centroid.

[0096] The step of encoding the centroid and surrounding sampling points based on the information of the precise sampling points to obtain the reconstructed binary feature descriptor includes the following steps:

[0097] S401, compare the gray values ​​of the central descriptive centroid with the precise sampling points to obtain a 64-dimensional binary string;

[0098] S402, draw a circle with each sampled description centroid as the center, and use the circle as the image block of each sampled description centroid, wherein the distance from the sampled description centroid to the central description centroid is used as the radius of the circle;

[0099] S403, perform circumferential sampling on each of the image blocks in sequence to obtain grayscale values ​​of 32 pixels;

[0100] S404, compare the description centroid of each sample with the gray value of the corresponding 32 pixels in sequence to obtain a 6×32-dimensional binary string;

[0101] S405, the 6×32-dimensional binary string is concatenated with the 64-dimensional binary string to obtain the reconstructed 256-dimensional binary feature descriptor.

[0102] The advantages are:

[0103] 1. By incorporating the concept of centroid into traditional feature descriptors, the grayscale information of the key point neighborhood is increased, making the key point feature descriptors less sensitive to environmental interference factors such as lighting and rotation, thereby enhancing the robustness of the feature descriptors.

[0104] 2. A multi-center circular coding step has been added to further obtain information describing the neighborhood of the center;

[0105] 3. A weighted summation method has been added to effectively prevent some sampling points from being prominent noise points or producing aliasing effects when sampling at a fixed time.

[0106] Based on the above-mentioned feature descriptor reconstruction method based on multi-center sampling equalization filtering, this embodiment provides a feature descriptor matching device based on multi-center sampling equalization filtering, including:

[0107] The description center point selection module 1 extracts key points from the original image, determines the key points as a central description center point, and determines six sampling points evenly distributed around the key points as sampling description center points.

[0108] Fixed sampling module 2 performs fixed sampling on the original image based on the central descriptive centroid and the sampled descriptive centroid, to obtain grayscale change information of the descriptive centroid and grayscale change information of the sampling points around the descriptive centroid.

[0109] Filtering module 3 filters the grayscale change information to obtain information from several precise sampling points;

[0110] Encoding module 4 encodes the descriptive centroid and surrounding sampling points based on the information of the precise sampling points to obtain a reconstructed binary feature descriptor.

[0111] Furthermore, it is worth noting that the working process of the feature descriptor matching device based on multi-center sampling equalization filtering provided in this embodiment is the same as the working process of the feature descriptor reconstruction method based on multi-center sampling equalization filtering described above. For details, please refer to the working process of the feature descriptor reconstruction method based on multi-center sampling equalization filtering, which will not be repeated here.

[0112] Based on the above-described feature descriptor reconstruction method based on multi-center sampling equalization filtering, this embodiment provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described in the above embodiment.

[0113] like Figure 2 As shown, based on the feature descriptor reconstruction method based on multi-center sampling equalization filtering described above, this application also provides a terminal device, which includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communication interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communication interface 23 can communicate with each other via the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setup mode. The communication interface 23 can transmit information. The processor 20 can call logical instructions in the memory 22 to execute the methods in the above embodiments.

[0114] Furthermore, the logical instructions in the aforementioned memory 22 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0115] The memory 22, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of this disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions, or modules stored in the memory 22, thereby implementing the methods in the above embodiments.

[0116] The memory 22 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory. Examples include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, as well as transient storage media.

[0117] Compared to existing technologies, the feature descriptor reconstruction method based on multi-center sampling equalization filtering of the present invention includes: extracting key points in the original image; determining the key points as a central descriptor centroid; and determining six sampling points evenly distributed around the key points as sampling descriptor centroids; performing fixed sampling on the original image based on the central descriptor centroid and the sampling descriptor centroids to obtain grayscale change information of the descriptor centroid and grayscale change information of the sampling points around the descriptor centroid; filtering the grayscale change information to obtain information of several precise sampling points; and encoding the descriptor centroid and the surrounding sampling points based on the information of the precise sampling points to obtain a reconstructed binary feature descriptor. The present invention improves the robustness of the feature descriptor and increases the extraction quantity and matching accuracy under harsh conditions such as illumination changes.

[0118] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A feature descriptor reconstruction method based on multi-center sampling equalization filtering, characterized in that, include: Key points are extracted from the original image, and the key points are determined as a central descriptive centroid. Six sampling points centered on the key points and evenly distributed around the circumference are respectively determined as sampling descriptive centroids. Based on the central descriptive centroid and the sampled descriptive centroid, the original image is sampled in a fixed manner to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid. The grayscale change information is filtered to obtain information from several precise sampling points; Based on the information of the precise sampling points, the descriptive centroid and surrounding sampling points are encoded to obtain the reconstructed binary feature descriptor; The process of encoding the centroid and surrounding sampling points based on the information from the precise sampling points to obtain the reconstructed binary feature descriptor includes: The gray values ​​of the central descriptive centroid and the precise sampling points are compared to obtain a 64-dimensional binary string; Draw a circle with each sampled centroid as its center, and use the circle as the image patch for each sampled centroid, wherein the distance from the sampled centroid to the central centroid is used as the radius of the circle. Each image block is sequentially sampled circumferentially to obtain grayscale values ​​of 32 pixels; The description centroid of each sample is compared sequentially with the gray values ​​of the corresponding 32 pixels to obtain a 6×32-dimensional binary string; The 6×32-dimensional binary string is concatenated with the 64-dimensional binary string to obtain the reconstructed 256-dimensional binary feature descriptor.

2. The feature descriptor reconstruction method based on multi-center sampling equalization filtering according to claim 1, characterized in that, The process of extracting key points from the original image, determining the key points as a central descriptive centroid, and determining six sampling points evenly distributed around the key points as sampling descriptive centroids includes: Key points are extracted from the original image, and the initial feature descriptor of the original image is sampled in a circular distribution based on the key points to obtain key point information and sampling point information around the key points; Based on the key point information and the sampling point information around the key point, the key points in the original image are extracted, and the key point is determined as a central descriptive centroid. Six sampling points centered on the key point and evenly distributed around the circumference are respectively determined as sampling descriptive centroids.

3. The feature descriptor reconstruction method based on multi-center sampling equalization filtering according to claim 2, characterized in that, The process of performing fixed sampling on the original image based on the centroid of the center and the sampled centroids to obtain grayscale change information of the centroid and grayscale change information of the sampling points around the centroid includes: The original image is sampled using the description centroid based on the center and the sampled description centroid. By comparing the pixel values ​​of the described centroid point with those of the sampling points around the centroid point, grayscale change information of the described centroid point and grayscale change information of the sampling points around the centroid point are obtained.

4. The feature descriptor reconstruction method based on multi-center sampling equalization filtering according to claim 3, characterized in that, The filtering process of the grayscale change information to obtain information from several precise sampling points includes: Based on the grayscale change information, the sampling points are weighted and summed using a weighted summation method to obtain the coordinate information and grayscale values ​​of several precise sampling points.

5. The feature descriptor reconstruction method based on multi-center sampling equalization filtering according to claim 4, characterized in that, The step of filtering the grayscale change information to obtain information from several precise sampling points also includes: If the sampling point during weighted summation is the centroid of the description, or if the distance from the sampling point to the corresponding centroid is greater than a preset threshold, then the first formula is selected; otherwise, the second formula is selected. The first formula used in the weighted summation method is as follows: The second formula used in the weighted summation method is as follows: Among them, in the first formula and the second formula Grayscale value The coordinates of the sampling point, Grayscale value , The weights assigned.

6. The feature descriptor reconstruction method based on multi-center sampling equalization filtering according to claim 5, characterized in that, The sampling formula used for the fixed sampling is as follows: Wherein, in the sampling formula The x-coordinate of the sampling points distributed in a circle. The vertical coordinates of the sampling points distributed in a circle are: For radius, =1.39, ( , () represents the coordinates of the key point. For the angle of the sampling point, For each circle's circumference... One sampling point, It is the first The number of sampling points on each circle increases linearly. .

7. A feature descriptor matching device based on multi-center sampling equalization filtering, used to implement the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described in any one of claims 1-6, characterized in that, include: The description center point selection module extracts key points from the original image, determines the key points as a central description center point, and determines six sampling points evenly distributed around the key points as sampling description center points. The fixed sampling module performs fixed sampling on the original image based on the central descriptive centroid and the sampled descriptive centroid, to obtain the grayscale change information of the descriptive centroid and the grayscale change information of the sampling points around the descriptive centroid. The filtering module filters the grayscale change information to obtain information from several precise sampling points; The encoding module encodes the descriptive centroid and surrounding sampling points based on the information of the precise sampling points to obtain the reconstructed binary feature descriptor.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described in any one of claims 1-6.

9. A terminal device, characterized in that, include: Processor, memory, and communication bus; the memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it implements the steps in the feature descriptor reconstruction method based on multi-center sampling equalization filtering as described in any one of claims 1-6.

Images