Integrated circuit scanning electron microscope image noise reduction method

Abstract

The invention discloses an integrated circuit scanning electron microscope image noise reduction method, which can avoid photoresist shrinkage and make image imaging clearer compared with the traditional multi-frame superposition technology in the industry. The method specifically comprises the following steps: acquiring a wafer with consistent key feature size on a semiconductor manufacturing production line; obtaining SEM images of the target point in different core particles; searching one image from the SEM images as a reference image, and taking the rest images as to-be-compared images; acquiring all overlapped regions between the reference image and the to-be-compared image; and calculating the maximum intersection area of all the overlapped areas, and carrying out pixel-level averaging on the maximum intersection area to obtain a noise-reduced SEM image. According to the method, the noise of the SEM image can be reduced, the image quality is improved, and a reliable data source is provided for product quality inspection and photoetching model driving in the semiconductor manufacturing process.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor manufacturing and image processing technology, and relates to a method for reducing noise in an integrated circuit scanning electron microscope image, and in particular to CD-SEM image data, which can provide convenient operation for synthesizing CD-SEM measurement data for computational lithography and collecting clear SEM images for a yield analysis unit. Background Art

[0002] In the field of integrated circuit manufacturing, as semiconductor processes develop towards miniaturization and complexity, the importance of scanning electron microscopes (SEM) in measurement and detection has become increasingly prominent. After each process link of the semiconductor production line is completed, it is necessary to use a critical dimension scanning electron microscope (CD-SEM) to sample and inspect the wafer to evaluate whether the integrated circuits after exposure, development, etching, grinding and other processes meet the standards, so as to ensure the smooth progress of the subsequent process flow. If anomalies are found in the early stages, timely measures can be taken to reduce losses. In the process of semiconductor process research and development, the computational lithography and yield improvement teams frequently use CD-SEM to collect data. The computational lithography department collects a large number of CD-SEM images of test patterns to build lithography models and etching models, and conducts model verification and other work. These images provide a solid data foundation for lithography simulation verification. The yield improvement department also widely collects SEM images of hot spots and proposes targeted yield improvement strategies based on them.

[0003] SEM images play a key role in semiconductor manufacturing, but due to the imaging principle of the machine, a lot of noise will appear. In order to effectively reduce the noise generated by CD-SEM equipment and obtain high-quality images, the semiconductor industry generally adopts a technology called multi-frame stacking. This technology involves multiple electron beam scans of the same position to obtain multiple CD-SEM images, which are then stacked and averaged to generate a higher quality composite image. Since the noise distribution in each image is random, through stacking, the noise in different directions can cancel each other out, while the common signal components in the image are retained, thereby forming a clearer image. Although the multi-frame stacking technology can reduce noise to a certain extent, each electron beam scan may cause a slight shrinkage of the photoresist, resulting in a nonlinear deviation between the measurement results and the actual process conditions. In addition, since it takes a certain amount of time to acquire each frame of the image, acquiring multiple frames of images will significantly increase the total image acquisition time. At the same time, a large number of electron beam bombardments may also cause the performance of the material being inspected to deteriorate.

[0004] Semiconductor process R&D and actual production require the collection of a large amount of SEM data, and these data require CD-SEM to work overtime. Due to the high price of the machine and when the imaging quality requirements are high (low noise), the collection of a large amount of data seems to be beyond the ability, which greatly hinders the production and R&D cycle. At present, the application of deep learning in the field of noise reduction also requires a large number of noise-free and noisy paired image labels, which is extremely costly for researchers who have not been deeply involved in the semiconductor manufacturing industry. Summary of the invention

[0005] Based on the above problems, the present invention proposes an integrated circuit scanning electron microscope image noise reduction method to solve the photoresist shrinkage problem caused by the current industry's use of multi-frame image overlay technology, and provide the semiconductor manufacturing industry with clearer SEM image data that is closer to the actual process.

[0006] The objective of the present invention is achieved through the following technical solutions: In a first aspect, the present invention provides a method for reducing noise in an integrated circuit scanning electron microscope image, the method comprising: Get a wafer with consistent critical feature dimensions on a semiconductor manufacturing line; Obtain SEM images of target points in different core particles; Find one image in the above SEM images as a reference image, and the rest of the images as images to be compared; Obtain all overlapping areas between the reference image and the image to be compared; The maximum intersection area of ​​all overlapping areas is calculated, and the maximum intersection area is averaged at the pixel level to obtain the denoised SEM image.

[0007] Preferably, the exposure conditions of each core in the wafer are the same.

[0008] Preferably, the reference image search process is as follows: The amplitude spectrum information of each SEM image was calculated using Fourier analysis method; In the amplitude spectrum, the low-frequency region is intercepted with the origin of the frequency domain as the center; The average amplitude of the low-frequency region of each SEM image was calculated, and the SEM image corresponding to the maximum amplitude average was used as the reference image.

[0009] Preferably, the process of acquiring all overlapping areas between the reference image and the image to be compared is as follows: Finding reference images The image to be compared , i=[1,k] feature points, k represents the number of images to be compared; According to the feature points of the reference image, the feature points of the image to be compared are matched one by one to obtain matching pairs, and then the homography matrix is ​​obtained; The overlapping area between each image to be compared and the reference image is obtained.

[0010] Preferably, the feature point search process is as follows: For each image ,q=[0,k] forms a multi-scale Gaussian pyramid through multi-scale Gaussian filter and image convolution .

[0011] For the same original image, the Gaussian pyramid Subtract adjacent size images to obtain the Gaussian difference scale space .

[0012] Finding the Difference of Gaussian Scale Space The extreme points are taken as the feature points of the current image; Use sub-pixel interpolation to shift each feature point to the correct position to obtain the corrected feature point; The direction and amplitude of the corrected feature points in the Gaussian difference scale space are calculated, and the position and feature vector are used to describe each feature point.

[0013] Preferably, the feature vector calculation process of the feature point is as follows: Take a neighborhood around the corrected feature point position and calculate the gradient size and direction of the neighborhood; Create a direction histogram containing 8 direction columns, count the gradient size into the histogram, and get a direction vector containing different direction strength values.

[0014] Preferably, the implementation process of matching the feature points of the reference image with the feature points of the image to be compared one by one to obtain matching pairs and then obtaining the homography matrix is ​​as follows: 1) Construct a KD tree with spatial geometric features based on the feature vectors describing all feature points of the reference image; 2) For each feature point of the image to be compared, find the feature point closest to it from the KD tree as the reference feature point. The reference feature point and the feature point of the current image to be compared are used as a set of matching pairs, so as to obtain the corresponding relationship between the feature points of the image to be compared and the reference image; 3) Calculate the Euclidean distance between the feature points of the image to be compared and the feature points of the reference image of each matching pair, and remove the feature points of the image to be compared whose distance exceeds the threshold; 4) Solve the following non-sublinear equations to obtain the homography matrix H;

[0015] in The feature vector representing the feature points of the reference image, The feature vector representing the feature points of the image to be compared.

[0016] Preferably, the process of obtaining the overlapping area between each image to be compared and the reference image is as follows: 1) Place the reference image The four edge points are mapped to the image to be compared through the homography matrix H In the projection coordinates , remove the projection coordinates that do not fall within the current image to be compared.

[0017] 2) Map the four edge points in the image to be compared to the reference image through the homography matrix H to obtain the projection coordinates ; 3) , As vertex coordinates, determine an overlapping area in the images to be compared ; 4) Repeat the above steps for all images to be compared to form an overlapping area pool.

[0018] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method described.

[0019] In a third aspect, the present invention provides a computing device, comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the described method is implemented.

[0020] The beneficial effects of the present invention are: The present invention obtains wafers with consistent key feature dimensions in the semiconductor manufacturing production line, obtains target point SEM images from different core particles, uses different core particle images as the data basis, replaces the superposition of multiple frames of images, avoids the problem of photoresist shrinkage from the source, ensures that the measurement results truly reflect the process conditions, and improves measurement accuracy, which is particularly important for advanced node semiconductor manufacturing. The present invention obtains images from different core particles, and a single scan can collect multiple core particle images, reducing the number of electron beam scans, reducing the impact on the performance of the material being tested, improving production efficiency, and alleviating the problems of high data collection costs and long production and research and development cycles in the semiconductor manufacturing industry.

[0021] The present invention forms a Gaussian pyramid by convolution of a multi-scale Gaussian filter and an image, subtracts adjacent size images to obtain a Gaussian difference scale space, finds extreme points as feature points and corrects them, and calculates feature vectors to achieve feature point description. Construct a KD tree to match feature points, remove feature points whose distance exceeds the threshold, solve the non-sublinear equation system to obtain the homography matrix H, thereby obtaining the overlapping area and calculating the maximum intersection area, performing pixel-level averaging to obtain a denoised image, effectively removing noise, improving image clarity, and ensuring that the measurement results are accurate. Perform pixel-level averaging on the maximum intersection area to effectively remove noise in the SEM image, more accurately reveal the process information of the actual semiconductor process, and will not cause drastic shrinkage of the photoresist. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 These are SEM images formed by superimposing multiple frames of traditional electron beam scanning, where (a) the line width of 1 frame is 146.04 nm, (b) the line width of 4 frames is 140.38 nm, (c) the line width of 16 frames is 137.34 nm, (d) the line width of 64 frames is 131.64 nm, (e) the line width of 128 frames is 127.86 nm, and (f) the line width of 256 frames is 122.75 nm.

[0023] Figure 2 It is a schematic diagram of the process of the present invention.

[0024] Figure 3 This is a conceptual diagram of the method of the present invention.

[0025] Figure 4 Schematic diagram of CDU wafer.

[0026] Figure 5 The SEM single-frame images are formed by scanning in different directions, where (a) is the horizontal scanning image of the storage circuit area, (b) is the vertical scanning image of the memory area; (c) is the horizontal scanning image of the logic circuit area, and (d) is the vertical scanning image of the logic circuit area.

[0027] Figure 6 Fourier transform amplitude spectra of SEM images of different core particles, where (a) is the amplitude spectrum distribution of the reference image, (b), (c), (d), (e), and (f) are the amplitude spectra of 5 different images to be compared.

[0028] Figure 7 A schematic diagram of the process of obtaining all overlapping areas between a reference image and an image to be compared.

[0029] Figure 8 The following is an example of the location of feature points extracted from a SEM image.

[0030] Fig. 9 This is the KD tree spatial index matching result adopted by the present invention.

[0031] Fig.10 The feature point matching between the image to be compared (left) and the reference image (right).

[0032] Fig.11 The collected SEM images of different core particles and the noise-reduced SEM images obtained by the present invention, wherein (a) is a single-frame SEM image of horizontal (x-axis) scanning imaging, (b) is a single-frame SEM image of vertical (y-axis) scanning imaging, (c) is a superimposed SEM image of two core particles of horizontal and vertical scanning images, (d) is a superimposed SEM image of 4 core particles, (e) is a superimposed SEM image of 6 core particles, (f) is a superimposed SEM image of 10 core particles, (g) is a superimposed SEM image of 20 core particles, and (h) is a superimposed SEM image of 40 core particles. DETAILED DESCRIPTION

[0033] The specific implementation of the present invention will be described in more detail below in conjunction with the schematic diagram. The advantages and features of the present invention will become clearer based on the following description. It should be noted that the drawings are all in a very simplified form and are not in exact proportions, and are only used to facilitate and clearly assist in explaining the purpose of the embodiments of the present invention.

[0034] CDSEM graphic measurement plays an important role in the fields of computational lithography and yield improvement. The computational lithography department collects a large number of CDSEM images of test graphics to establish lithography models, etching models, model verification, etc. The yield improvement department also collects a large number of SEM images of hot spots for analysis. Semiconductor process research and development and actual production require the collection of a large amount of SEM data, and these data require CD-SEM to work overtime. Due to the high price of the machine and when the imaging quality requirements are high (low noise), the collection of a large amount of data seems to be beyond the capacity, which greatly hinders the production and R&D cycle. At present, the application of deep learning in the field of noise reduction also requires a large number of noise-free and noisy paired image labels, which is extremely costly for researchers who have not been deeply involved in the semiconductor manufacturing industry. The semiconductor manufacturing industry usually uses image multi-frame superposition technology to suppress noise and obtain higher-quality SEM images, that is, multiple electron beam scans are performed on the same position to obtain multiple CD-SEM images, and then these images are superimposed and averaged to obtain a relatively good quality multi-frame image.

[0035] See also Figure 1 Middle (a) to Figure 1In (f), the figure shows SEM images of the same line pattern in different core particles with different scanning frame numbers, where Mean' represents the line width value measured by the mean. Due to the randomness of the noise distribution of each image, the noise in different directions will cancel each other out, and the common signal components will be retained. It can be seen that as the number of scanning frames increases, the image gradually becomes clearer. However, as the number of scanning frames increases, the line width becomes smaller and smaller. This is due to the shrinkage of the photoresist on the wafer caused by multiple reciprocating electron beam scanning. The measured value of the photoresist line width has shrunk from 146.04 nm to 122.75 nm. The SEM image cannot effectively reflect the current actual process information. This defect becomes more obvious at more advanced nodes.

[0036] Inspired by the multi-frame image superposition technology, the present invention proposes a SEM image multi-core particle superposition technology, see Figure 2-Figure 3 , a method for reducing noise in an integrated circuit scanning electron microscope image of this embodiment comprises the following steps: Step S1, obtaining a critical feature dimension uniform wafer (CDU wafer) after exposure on a semiconductor manufacturing production line, wherein the exposure conditions of each die in the wafer are the same.

[0037] See also Figure 4 The exposure energy and focal length of the CDU wafer are fixed at 46 mJ / cm² and 5 The exposure energy and exposure focus of each core on the wafer remain unchanged. It is mainly used to evaluate the consistency of feature size during the lithography process to ensure that a uniform pattern size can be obtained across the entire wafer.

[0038] Step S2, obtaining SEM images of target points in different core particles: According to the coordinate information of the target point, SEM images of the target point are collected at different core positions of the wafer to obtain multiple SEM images. Parts with significant defects in the image are removed from the above SEM images, such as overexposure of the image, inconsistency between the developed image and the target image, photoresist collapse, photoresist stickiness, etc., and SEM images with normal images and no defects are screened out. If the number of defect-free images is too small, additional images should be collected. The present invention recommends not less than 10 defect-free images. In particular, in order to ensure the clarity of the final imaging, the present invention recommends scanning and acquiring SEM images from multiple angles. See Figure 5 In (a) to (d) in FIG5 , the present invention performs horizontal and vertical scanning respectively in the actual process. The first column is a horizontal scan, and it can be seen that the longitudinal lines are relatively clear; the second column is a vertical scan, and it can be seen that the horizontal lines are relatively clear. It can be observed that the vertical contour of the horizontal scan image is relatively clear, while the horizontal contour of the longitudinal scan image is relatively clear. Therefore, by superimposing the horizontal and longitudinal SEM images, an image with a clearer contour can be obtained.

[0039] Step S3: Finding a reference image: S3-1 Since there is a certain degree of displacement between each SEM image, the traditional spatial distance cannot effectively measure the similarity between images, so as to find the optimal reference image. Therefore, the present invention adopts the Fourier analysis method to select the optimal reference image. SEM images often have a lot of noise pollution. In the spectrum analysis, the low frequency represents the general outline of the image, and the high frequency is the image details and noise part.

[0040] Specifically, the amplitude spectrum information of each SEM image screened in step S2 is calculated using the Fourier analysis method;

[0041] in is the frequency domain coordinate The amplitude value at , Re is the amplitude function extracted after Fourier transform, Coordinates on the image The pixel value of the image, H and W are the image height and width respectively.

[0042] See also Figure 6 Middle (a) to Figure 6 In (f), the amplitude spectrum of different SEM images is shown. The middle area and the horizontal and vertical center axis represent low-frequency information. Since each image has different noise pollution, the signal intensity in the low-frequency part will also be different.

[0043] S3-2 In the amplitude spectrum, the low-frequency region is intercepted with the origin of the frequency domain as the center; the size of the low-frequency region is preferably one tenth of the image pixel; S3-3 calculates the amplitude average value of the low-frequency region of each SEM image, takes the SEM image corresponding to the maximum amplitude average value as the reference image, and the remaining SEM images as the images to be compared.

[0044] In this embodiment, the SEM image corresponding to the first spectrum information is selected as the reference image. It can be seen from the image that its low-frequency information is more obvious, so the noise pollution level is lower.

[0045] Step S4: Obtain all overlapping areas between the reference image and the image to be compared, see Appendix Figure 7 : S41 Find reference image The image to be compared , feature points between i=[1,k], the main steps are as follows: 1) For each image ,q=[0,k] forms a multi-scale Gaussian pyramid through multi-scale Gaussian filter and image convolution ; Where G is a multi-scale Gaussian filter function, and its variance You can choose according to the actual situation. For example, taking values ​​of 3, 4, 7, and 9 respectively can generate 4 Gaussian convolution kernels with different blur scales, and then complete the convolution to generate a layer of scale space.

[0046] 2) For the same original image, the Gaussian pyramid Subtract adjacent size images to obtain the Gaussian difference scale space (DOG). .

[0047] 3) Finding the Gaussian Difference Scale Space The extreme points are taken as the feature points of the current image, and the above steps are repeated to obtain the feature point set T of all images.

[0048] 4) Use sub-pixel interpolation to offset each feature point to the correct position. The offset is , and obtain the corrected feature points.

[0049] 5) Calculate the direction and amplitude of the corrected feature points in the Gaussian difference scale space, and use position and feature vector to describe each feature point; The feature vector calculation process of the feature point: Take a neighborhood around the corrected feature point position and calculate the gradient size and direction of the neighborhood.

[0050] Create a direction histogram containing 8 direction columns, each direction column represents 45 degrees, and count the gradient size into the histogram to obtain a direction vector containing different direction strength values.

[0051] See also Figure 8 Through the above steps, the feature points in an image can be obtained. The circles of different colors in the figure mark the location of each feature point. When storing, a set of mathematical vectors (position, scale, direction, etc.) are used to describe the feature points to ensure the complete information of the feature points.

[0052] S42 matches the feature points of the reference image with the feature points of the image to be compared one by one, obtains matching pairs, and then obtains the homography matrix. Homography refers to the transformation from a point in one image to a corresponding point in another image. Specifically: 1) Based on the feature vectors describing all feature points of the reference image, a KD tree with spatial geometric features is constructed.

[0053] 2) For each feature point of the image to be compared, find the feature point closest to it from the KD tree as the reference feature point. The reference feature point and the feature point of the current image to be compared are used as a set of matching pairs to obtain the corresponding relationship between the feature points of the image to be compared and the reference image.

[0054] 3) Calculate the Euclidean distance between the feature points of the image to be compared and the feature points of the reference image of each matching pair, and remove the feature points of the image to be compared whose distance exceeds the threshold; The Euclidean distance is calculated as follows:

[0055] Among them, n represents the spatial dimension of the feature vector describing the feature point, represents the feature vector of the reference image feature point in the i-th matching pair, The feature vector representing the feature points of the images to be compared in the i-th matching pair; The feature vector representing the feature points of the reference image, A feature vector representing a feature point of the image to be compared; See also Fig. 9 , a KD tree is constructed using the feature point information of the reference image. This tree has spatial geometric information. When matching any feature point of the image to be compared, the nearest neighbor matching point can be quickly found based on the spatial tree structure, which speeds up the matching process. The blue dots in the figure are the position distribution of the feature points in the multidimensional space. The horizontal and vertical lines constitute different spatial dimensions, and the intersection position is the intersection between the spatial dimensions. Fig.10 , showing the matching of two different core particle SEM images. The process described in the present invention can effectively pair the feature points.

[0056] 4) Solve the following non-sublinear equations to obtain the homography matrix H;

[0057] S43 obtains the overlapping area between each to-be-compared image and the reference image, and the specific method is as follows.

[0058] 1) Place the reference image The four edge points are mapped to the image to be compared through the homography matrix H In the projection coordinates ,in ,r=1, 2, 3, 4, which are the image edge points (0, 0), (0, H), (W, 0), (HW) respectively. For reference image The image to be compared with The intersection between them retains the projection coordinates inside the current image to be compared, that is, .

[0059] 2) Map the four edge points in the image to be compared to the reference image through the homography matrix H to obtain the projection coordinates ; 3) As vertex coordinates, determine an overlapping area in the images to be compared .

[0060] 4) Repeat the above steps for all images to be compared to form an overlapping area pool.

[0061] Step S5, calculate the maximum intersection area of ​​all overlapping areas in the overlapping area pool, and perform pixel-level averaging on the maximum intersection area, and output a clear, low-noise SEM image. Fig.11 Middle (a) to Fig.11 As shown in (h), the number of SEM images collected in the vertical scanning direction and the number of SEM images collected in the horizontal scanning direction in this embodiment are each half, and the yellow frame in the figure represents the overlapping area of ​​the two original single-frame SEM images.

[0062] The calculation formula for pixel-level averaging of the maximum intersection area is as follows:

[0063] This operation cleverly utilizes the information of pixels at the same position in multiple images. Through averaging, it effectively reduces the noise interference in a single image while retaining the common signal components in the image. Finally, the image after pixel-level averaging is output to obtain a clear, low-noise SEM image. This image is not only visually clearer, but also has a significant improvement in data accuracy. It effectively removes noise in the SEM image and can more accurately reveal the process information of the actual semiconductor process without causing drastic shrinkage of the photoresist.

[0064] Through the above series of implementation steps, the present invention successfully realizes an effective integrated circuit scanning electron microscope image noise reduction method. From the careful preparation of the wafer and image acquisition, to the scientific selection of the reference image and the precise matching of the feature points, to the precise determination of the overlapping area and the superposition processing of the image, this method not only solves the photoresist shrinkage problem caused by the traditional multi-frame superposition technology, but also significantly improves the image clarity and the accuracy of the measurement results, providing a reliable image data source for product quality inspection and lithography model driving in the semiconductor manufacturing process.

Claims

1. A method for reducing noise in an integrated circuit scanning electron microscope image, characterized in that: The method comprises: Get a wafer with consistent critical feature dimensions on a semiconductor manufacturing line; Obtain SEM images of target points in different core particles; Find one image in the above SEM images as a reference image, and the rest of the images as images to be compared; Obtain all overlapping areas between the reference image and the image to be compared; The maximum intersection area of ​​all overlapping areas is calculated, and the maximum intersection area is averaged at the pixel level to obtain the denoised SEM image.

2. The method according to claim 1, characterized in that: The exposure conditions of each die in the wafer are the same.

3. The method according to claim 1, characterized in that: The reference image search process is as follows: The amplitude spectrum information of each SEM image was calculated using Fourier analysis method; In the amplitude spectrum, the low-frequency region is intercepted with the origin of the frequency domain as the center; The average amplitude of the low-frequency region of each SEM image was calculated, and the SEM image corresponding to the maximum amplitude average was used as the reference image.

4. The method according to claim 1, characterized in that: The process of obtaining all overlapping areas between the reference image and the image to be compared is as follows: Finding reference images The image to be compared , i=[1,k] feature points, k represents the number of images to be compared; According to the feature points of the reference image, the feature points of the image to be compared are matched one by one to obtain matching pairs, and then the homography matrix is ​​obtained; The overlapping area between each image to be compared and the reference image is obtained.

5. The method according to claim 4, characterized in that: The feature point search process is as follows: For each image ,q=[0,k] forms a multi-scale Gaussian pyramid through multi-scale Gaussian filter and image convolution ; For the same original image, the Gaussian pyramid Subtract adjacent size images to obtain the Gaussian difference scale space ; Finding the Difference of Gaussian Scale Space The extreme points are taken as the feature points of the current image; Use sub-pixel interpolation to shift each feature point to the correct position to obtain the corrected feature point; The direction and amplitude of the corrected feature points in the Gaussian difference scale space are calculated, and the position and feature vector are used to describe each feature point.

6. The method according to claim 5, characterized in that: The feature vector calculation process of the feature point is as follows: Take a neighborhood around the corrected feature point position and calculate the gradient size and direction of the neighborhood; Create a direction histogram containing 8 direction columns, count the gradient size into the histogram, and get a direction vector containing different direction strength values.

7. The method according to claim 4, characterized in that: The implementation process of matching the feature points of the reference image with the feature points of the image to be compared one by one, obtaining matching pairs, and then obtaining the homography matrix is ​​as follows: 1) Construct a KD tree with spatial geometric features based on the feature vectors describing all feature points of the reference image; 2) For each feature point of the image to be compared, find the feature point closest to it from the KD tree as the reference feature point. The reference feature point and the feature point of the current image to be compared are used as a set of matching pairs, so as to obtain the corresponding relationship between the feature points of the image to be compared and the reference image; 3) Calculate the Euclidean distance between the feature points of the image to be compared and the feature points of the reference image of each matching pair, and remove the feature points of the image to be compared whose distance exceeds the threshold; 4) Solve the following non-sublinear equations to obtain the homography matrix H; ； in The feature vector representing the feature points of the reference image, The feature vector representing the feature points of the image to be compared.

8. The method according to claim 4, characterized in that: The implementation process of obtaining the overlapping area between each image to be compared and the reference image is as follows: 1) Place the reference image The four edge points are mapped to the image to be compared through the homography matrix H In the projection coordinates , remove the projection coordinates that do not fall within the current image to be compared; 2) Map the four edge points in the image to be compared to the reference image through the homography matrix H to obtain the projection coordinates ; 3) , As vertex coordinates, determine an overlapping area in the images to be compared ; 4) Repeat the above steps for all images to be compared to form an overlapping area pool.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 8.

10. A computing device, comprising a memory and a processor, wherein the memory stores executable codes, and when the processor executes the executable codes, the method according to any one of claims 1 to 8 is implemented.

Images