Image splicing method for improving image registration and combining splicing seam fusion

Abstract

The invention discloses an image splicing method for improving image registration and combining splicing seam fusion. The image splicing method comprises the following steps: carrying out local registration on a source picture and a target picture; global matching is carried out on the source picture and the target picture, and an optimal splicing seam is found in the two registered pictures; partition fusion is used to eliminate the splicing seam influence of the spliced picture. According to the method, the splicing problems of ghosting of an image overlapping part, ghosting of a moving object, exposure difference and the like in parallax image splicing are solved. According to the method, the quality of picture splicing can be improved, ghosting of the overlapped part in the parallax image is effectively eliminated, the problem of small exposure difference is solved, and meanwhile, the integrity of important moving objects of the overlapped part is greatly reserved.

Description

Technical Field

[0001] The present invention relates to the field of image and video technology, and in particular to an image stitching method combining stitching seam fusion for improving image registration. Background Art

[0002] The process of processing and screening multi-source data information by computer, integrating, analyzing and synthesizing the data according to certain criteria, and finally completing data decision and estimation is called information fusion. Image stitching technology generally includes three parts: image feature extraction, image registration and image fusion. It mainly stitches the overlapping parts of two or more images seamlessly to obtain a wide-angle image. Image fusion is a kind of information fusion and has been widely used in image processing fields such as remote sensing detection and computer vision. . Traditional image stitching methods have the problems of slow speed and low precision when processing large amounts of data images, especially at the seams of image stitching, obvious connection marks are likely to appear. Summary of the invention

[0003] In view of the above problems, the present invention is proposed to provide an image stitching method combining stitching seam fusion with improved image registration, which overcomes the above problems or at least partially solves the above problems.

[0004] In order to solve the above technical problems, the embodiments of the present application disclose the following technical solutions:

[0005] In a first aspect, an embodiment of the present invention discloses an image stitching method combining stitching seam fusion for improving image registration, comprising:

[0006] S100. Perform local registration on the source image and the target image;

[0007] S200. Globally align the source image and the target image

[0008] S300. Find the best stitching seam between two registered images in S100-S200;

[0009] S400. Use partition fusion to eliminate the influence of the stitching seams of the stitched images.

[0010] Furthermore, in S100, the source image and the target image are locally aligned, and the specific method includes: detecting key positions and extracting feature points of the two images, locating key points from key positions, extracting descriptors from key points, and matching feature vectors in the model and image; after the two images are determined to match pairs using the SURF algorithm or the SIFT algorithm, the RANSAC algorithm is used to filter out incorrect matching pairs, and the correct matching pairs are retained, and the homography matrix is ​​calculated using the retained matching pairs. The matching points of the overlapping parts of the source image and the target image need to be represented by three-dimensional coordinates. The homography matrix can seamlessly align points in the same plane, but cannot overlap corresponding points in other planes.

[0011] Furthermore, the three-dimensional coordinates of the matching points of the overlapping parts are expressed as:

[0012]

[0013] Among them, (x2, y2, 1) is the coordinate after mapping, (x1, y1, 1) is the coordinate before mapping, h i,j is the parameter of the transformation matrix. Using the four pairs of matching points in the two images, we can solve a homography matrix, a homography matrix H * Can seamlessly align points in the same plane.

[0014] Furthermore, the homography matrix H * The solution method includes: introducing a local homography matrix to solve the homography matrix H * , calculated by the following formula:

[0015] x * =H*x * '

[0016] where x * is the coordinate before transformation, x′ * is the transformed coordinate, and the transformation matrix parameter h is obtained by calculating the following weighted formula * :

[0017]

[0018] Weight matrix W * The composition is as follows:

[0019]

[0020] Use the global homography transformation to calculate the size of the final canvas, divide the canvas into N×N grids, calculate the vertex coordinates of each grid, calculate the Euclidean distance and Gaussian weight of each point in the left image and the center point of the grid obtained by the RANSAC algorithm, and the weight component The calculation formula is as follows, where σ is a scalar parameter and γ is a 0-1 parameter:

[0021]

[0022] Further, in S200, the source image and the target image are globally registered, and the specific method includes: for the projection distortion problem of the non-overlapping part, the problem is solved by combining the image with the global similarity transformation matrix, the global similarity transformation matrix is ​​calculated by the inlier matching pair obtained by the RANSAC algorithm, and the inlier matching needs to be screened after it is obtained, and a preset threshold is set to extract the matching pairs less than the preset threshold. The preset threshold is set to be lower than the value of the homography matrix, and the operation is stopped when the number of extracted matching pairs reaches the set number; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the transformation function H of the target image is obtained by weighting with a sine function. t Next, the target image is smoothly interpolated, and the transformation function H of the source image is calculated by the target transformation function r .

[0023] Furthermore, the transformation function of the target image and the source image is calculated by the global similarity matrix and the local homography matrix, which is:

[0024]

[0025] Among them, S i is the global similarity matrix, a ii is the global homography matrix parameter. The similarity matrix is ​​calculated for each of the 50 matching pairs. The pair with the smallest similarity transformation angle is taken as the global similarity matrix S. * , the overlapping part is locally homographed; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the sine function is used for weighting to obtain the transformation function H of the target image t , smoothly insert the target image, and calculate the transformation function H of the source image through the target transformation function r .

[0026] Furthermore, the transformation function H of the target graph t And the transformation function H of the source image r The calculation formula is:

[0027] H t =(a-0.2×sin(πa))×H * +(1-a+0.2×sin(πa))×S *

[0028]

[0029] Among them, the coefficient a is obtained by calculating the weight through the integral of the projection point of the target graph.

[0030] Further, in S300, the two images registered in S100-S200 are searched for the best stitching seam, and the specific method includes: using a search path with the smallest difference in the overlapping parts of the two images as the stitching seam of the images for stitching, setting the overlapping parts of the two images to be I1 and I2 respectively, obtaining the grayscale and gradient difference images of the two images, performing threshold processing on the two images to obtain a binary image, and filtering out the area I with a pixel of 1 d1 and I d2 , and then use the dynamic programming algorithm to get the corresponding pixels with the smallest color difference and the most similar structure. Then, trace back each row of pixels to get the best stitching seam.

[0031] Furthermore, the energy function criterion of the improved optimal seam is as follows:

[0032]

[0033] a, b are 0.2 and 0.8 respectively, where E color is the energy value of the image color difference, and the specific formula is:

[0034]

[0035] E geometry is the energy value of the image structure difference, and the specific formula is:

[0036] E geometry (x,y)=ω2×(S x ×(I1(x,y)-I2(x,y))) 2

[0037] +ω2×(S y ×(I1(x,y)-I2(x,y))) 2

[0038] Define ω1 and ω2 as the weighting coefficients of the pixels in the color and gradient threshold regions, respectively. x , S y And use the following operator mask:

[0039]

[0040] Furthermore, in S400, the influence of the stitching seams of the stitched images is eliminated by using partition fusion, and the specific method includes:

[0041] S401. Cut out the image of the left area from the target image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the first image to be processed;

[0042] S402. Cut out the image of the right area from the source image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the second image to be processed;

[0043] S403. Splicing the first image to be processed and the second image to be processed according to the optimal splicing seam;

[0044] S404. Perform fade-in and fade-out fusion within 10 pixels of the best stitching seam with the path with the maximum energy value as the boundary.

[0045] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0046] The present invention discloses an image stitching method combined with seam fusion for improving image registration, including: locally registering a source image and a target image; globally registering the source image and the target image to find the best seam between two registered images; and using partition fusion to eliminate the seam influence of the stitched image. The present invention solves the splicing problems such as ghosting of overlapping parts of images, ghosting of moving objects, and exposure differences in parallax image stitching. The present invention can improve the quality of image stitching, effectively eliminate ghosting of overlapping parts of parallax images and eliminate small exposure differences, while greatly retaining the integrity of important moving objects in the overlapping parts.

[0047] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0049] Figure 1 This is a flow chart of an image stitching method combining stitching seam fusion with improved image registration in Embodiment 1 of the present invention;

[0050] Figure 2 The source image and the target image of scene 1 in embodiment 1 of the present invention;

[0051] Figure 3 This is a comparison diagram of the fusion bands of three algorithms in scene 1 in embodiment 1 of the present invention;

[0052] Figure 4 This is a diagram of the fusion effect of the algorithm of the present invention in scene 1 in embodiment 1 of the present invention. DETAILED DESCRIPTION

[0053] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0054] In order to solve the problems existing in the prior art, an embodiment of the present invention provides an image stitching method combined with stitching seam fusion for improving image registration.

[0055] Example 1

[0056] The present invention discloses an image stitching method combining stitching seam fusion with improved image registration, which is characterized by comprising:

[0057] S100. Perform local registration on the source image and the target image; local registration technology uses automatic or semi-automatic computer algorithms to correct the differences between different data sets and achieve alignment between them. In the field of computer vision, this usually involves accurately matching and aligning image data from different perspectives, different times, or different imaging conditions to the same coordinate system.

[0058] In S100 of the present embodiment, the source image and the target image are locally registered, and the specific method includes: detecting key positions and extracting feature points of the two images, locating key points from key positions, extracting descriptors from key points, and matching feature vectors in the model and image; after the two images are determined to match pairs using the SURF algorithm or the SIFT algorithm, the RANSAC algorithm is used to filter out mismatched pairs and retain correct matching pairs, and the retained matching pairs are used to calculate the homography matrix, and the matching points of the overlapping parts of the source image and the target image need to be represented by three-dimensional coordinates. The homography matrix can seamlessly align points in the same plane, but cannot overlap corresponding points in other planes.

[0059] In some preferred embodiments, the three-dimensional coordinates of the matching points of the overlapping parts are expressed as:

[0060]

[0061] Among them, (x2, y2, 1) is the coordinate after mapping, (x1, y1, 1) is the coordinate before mapping, h is the parameter of the transformation matrix, and four pairs of matching points in the two images are used to solve a homography matrix, a homography matrix H * Can seamlessly align points in the same plane.

[0062] In some preferred embodiments, a homography matrix H *Points in the same plane can be seamlessly aligned. Here, a local homography matrix is ​​introduced to solve this problem. The homography matrix H * The solution method includes: introducing a local homography matrix to solve the homography matrix H * , calculated by the following formula:

[0063] x * =H * x * '

[0064] where x * is the coordinate before transformation, x * ′ is the transformed coordinate, and the transformation matrix parameter h is obtained by calculating the following weighted formula * :

[0065]

[0066] Weight matrix W * The composition is as follows:

[0067]

[0068] Use the global homography transformation to calculate the size of the final canvas, divide the canvas into N×N grids, calculate the vertex coordinates of each grid, calculate the Euclidean distance and Gaussian weight of each point in the left image and the center point of the grid obtained by the RANSAC algorithm, and the weight component The calculation formula is as follows, where σ is a scalar parameter and γ is a 0-1 parameter:

[0069]

[0070] S200. Global registration of the source image and the target image; global registration is a technology widely used in the fields of point cloud processing, computer vision and medical imaging, and aims to accurately align two or more data sets (such as point clouds, images or medical images) in a global scope. In S200 of this embodiment, the source image and the target image are globally registered, and the specific method includes: for the projection distortion problem of the non-overlapping part, the problem is solved by combining the image with the global similarity transformation matrix, the global similarity transformation matrix is ​​calculated by the inlier matching pair obtained by the RANSAC algorithm, and the inlier matching is required to be screened, and a preset threshold is set to extract matching pairs less than the preset threshold. The preset threshold is set to be lower than the value of the homography matrix. When the number of extracted matching pairs reaches the set number, the operation is stopped; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the transformation function H of the target image is obtained by weighting with a sine function t Next, the target image is smoothly interpolated, and the transformation function H of the source image is calculated by the target transformation functionr .

[0071] In some preferred embodiments, the transformation function of the target image and the source image is calculated by the global similarity matrix and the local homography matrix, specifically:

[0072]

[0073] Among them, S i is the global similarity matrix, a ii is the global homography matrix parameter. The similarity matrix is ​​calculated for each of the 50 matching pairs. The pair with the smallest similarity transformation angle is taken as the global similarity matrix S. * , the overlapping part is locally homographed; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the sine function is used for weighting to obtain the transformation function H of the target image t , smoothly insert the target image, and calculate the transformation function H of the source image through the target transformation function r .

[0074] In some preferred embodiments, the transformation function H of the target graph is t And the transformation function H of the source image r The calculation formula is:

[0075] H t =(a-0.2×sin(πa))×H * +(1-a+0.2×sin(πa))×S *

[0076]

[0077] The coefficient a is obtained by calculating the weight through the integral of the projection points of the target graph.

[0078] S300. In S100-S200, the two images that have been registered are searched for the best stitching seam; specifically, if there is a search path with the smallest difference in the overlapping parts of the two images, this embodiment uses this path as the stitching seam of the images for stitching. The traditional stitching line considers the smallest difference, which means that the color difference and structure difference of the overlapping part are the smallest. Let the overlapping parts of the two images be I1 and I2 respectively, and obtain the grayscale and gradient difference images of the two. This paper performs threshold processing on the two images to obtain a binary image, and screens out the area I with a pixel of 1 d1 and I d2 Then, a dynamic programming algorithm is used to obtain the corresponding pixels with the smallest color difference and the most similar structure. The optimal stitching seam can be obtained by backtracking each row of pixels.

[0079] In some preferred embodiments, the energy function criterion of the improved optimal seam is as follows:

[0080]

[0081] a, b are 0.2 and 0.8 respectively, where E color is the energy value of the image color difference, and the specific formula is:

[0082]

[0083] E geometry is the energy value of the image structure difference, and the specific formula is:

[0084] E geometry (x,y)=ω2×(S x ×(I1(x,y)-I2(x,y))) 2

[0085] +ω2×(S y ×(I1(x,y)-I2(x,y))) 2

[0086] Define ω1 and ω2 as the weighting coefficients of the pixels in the color and gradient threshold regions, respectively. x , S y And use the following operator mask:

[0087]

[0088] S301. First, calibrate the motion and important objects that you want to retain in the grayscale difference map, and record the horizontal coordinate of the central pixel.

[0089] S302. Taking the pixels in the first row of the overlapping part as the initial point, the energy of each pixel is calculated by considering the boundary factor, and each pixel is numbered in sequence. N pixels are N different paths.

[0090] S303. Traverse downward from the first row to the last row, where one pixel in the previous row is considered to correspond to the three adjacent pixels in the next row as extension options, and the energy value is calculated according to the formula. The pixel with the smallest energy among the three pixels is used as the extension point of this row, and the path energy values ​​are accumulated. Specifically, in the operation, the point with the smallest energy value in this row is considered to be connected with the three paths with adjacent labels in the previous row, and the label of the path with the lowest energy among the three paths is assigned to this pixel point, and this is updated from the first row to the last row.

[0091] S304. When the last row is traversed, the energy of each path reaches the maximum. At this time, the energy value of each path is compared, and the path with the smallest energy value is selected. The entire path is traced back to obtain the optimal splicing seam.

[0092] S400. Use partition fusion to eliminate the effect of the stitching seam of the stitched image. Specifically, after the images are stitched according to the best stitching seam, the pixels around the line will have exposure differences and other problems, so Poisson fusion can be used to process the divided fusion zone to make the seam between the two images smoother and more natural.

[0093] In S400 of this embodiment, the influence of the stitching seams of the stitched images is eliminated by using partition fusion. The specific method includes:

[0094] S401. Cut out the image of the left area from the target image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the first image to be processed;

[0095] S402. Cut out the image of the right area from the source image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the second image to be processed;

[0096] S403. Splicing the first image to be processed and the second image to be processed according to the optimal splicing seam;

[0097] S404. Perform fade-in and fade-out fusion within 10 pixels of the best stitching seam with the path with the maximum energy value as the boundary.

[0098] In order to verify the actual effect of the disclosed method, the traditional optimal stitching seam combined with the fade-in and fade-out fusion algorithm is used as algorithm 1, and the traditional optimal stitching seam combined with Poisson fusion is used as algorithm 2. The fusion length of both is a fixed value of 1 / 5 of the length of the overlapping part. Figure 2 The two algorithms are compared with the patent algorithm in the experiment. The comparison chart is as follows: Figure 3 shown.

[0099] 3. Analysis of simulation results:

[0100] from Figure 4 It can be seen that the method proposed in this patent realizes splicing and fusion, the fusion effect is relatively natural and effectively eliminates the splicing seams and reduces the exposure differences.

[0101] In scene 1, the two fusion effects are compared. It can be seen that the traditional stitching algorithm cuts off the high-rise building in the distance, resulting in poor stitching and fusion effects. The fusion process of algorithm 1 also causes the moving object, the cyclist, to have a ghosting phenomenon, resulting in information loss. In algorithm 2, due to the use of a fixed region length, the girl on the road is not completely preserved. The method proposed in this paper can effectively make the transition of the fusion zone more natural, avoid cutting important objects, and at the same time, the length division more completely preserves the cyclist and the girl, that is, the moving object.

[0102] The simulation results demonstrate the effectiveness of the image stitching method proposed in this patent that combines improved image registration with stitching seam fusion.

[0103] This embodiment discloses an image stitching method combined with seam fusion for improved image registration, including: locally registering a source image and a target image; globally registering the source image and the target image to find the best seam between the two registered images; and using partition fusion to eliminate the influence of the seam of the stitched image. The present invention solves the stitching problems such as ghosting of overlapping parts of images, ghosting of moving objects, and exposure differences in parallax image stitching. The present invention can improve the quality of image stitching, effectively eliminate ghosting of overlapping parts of parallax images and eliminate small exposure differences, while greatly retaining the integrity of important moving objects in the overlapping parts.

[0104] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0105] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0106] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.

[0107] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0108] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0109] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. An image stitching method combining stitching seam fusion for improving image registration, characterized in that: include: S100. Perform local registration on the source image and the target image; S200. Perform global registration on the source image and the target image; S300. Find the best stitching seam between two registered images in S100-S200; S400. Use partition fusion to eliminate the influence of the stitching seams of the stitched images.

2. The image stitching method combined with stitching seam fusion for improving image registration according to claim 1, characterized in that: In S100, the source image and the target image are locally registered, and the specific method includes: detecting key positions and extracting feature points of the two images, locating key points from key positions, extracting descriptors from key points, and matching feature vectors in the model and image; after the two images are determined to match pairs using the SURF algorithm or the SIFT algorithm, the RANSAC algorithm is used to filter out mismatched pairs and retain correct matching pairs, and the retained matching pairs are used to calculate the homography matrix, and the matching points of the overlapping parts of the source image and the target image need to be represented by three-dimensional coordinates. The homography matrix can seamlessly align points in the same plane, but cannot overlap corresponding points in other planes.

3. The image stitching method combined with stitching seam fusion for improving image registration as claimed in claim 2, characterized in that: The three-dimensional coordinates of the matching points of the overlapping part are expressed as: Among them, (x2, y2, 1) is the coordinate after mapping, (x1, y1, 1) is the coordinate before mapping, h is the parameter of the transformation matrix, and four pairs of matching points in the two images are used to solve a homography matrix, a homography matrix H * Can seamlessly align points in the same plane.

4. The image stitching method combined with stitching seam fusion for improving image registration according to claim 2, characterized in that: The homography matrix H * The solution method includes: introducing a local homography matrix to solve the homography matrix H * , calculated by the following formula: x * =H * x * ′ Among them, x * is the coordinate before transformation, x * ′ is the transformed coordinate, and the transformation matrix parameter h* is obtained by calculating the following weighted formula: Weight matrix W * The composition is as follows: Use the global homography transformation to calculate the size of the final canvas, divide the canvas into N×N grids, calculate the vertex coordinates of each grid, calculate the Euclidean distance and Gaussian weight of each point in the left image and the center point of the grid obtained by the RANSAC algorithm, and the weight component The calculation formula is as follows, where σ is a scalar parameter and γ is a 0-1 parameter:

5. The image stitching method combined with stitching seam fusion for improving image registration according to claim 2, characterized in that: In S200, the source image and the target image are globally registered. The specific method includes: for the projection distortion problem of the non-overlapping part, the problem is solved by combining the image with the global similarity transformation matrix, the global similarity transformation matrix is ​​calculated by the inlier matching pair obtained by the RANSAC algorithm, and the inlier matching is screened after it is obtained, and a preset threshold is set to extract the matching pairs less than the preset threshold. The preset threshold is set to be lower than the value of the homography matrix, and the operation is stopped when the number of extracted matching pairs reaches the set number; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the transformation function H of the target image is obtained by weighting with a sine function t Next, the target image is smoothly interpolated, and the transformation function H of the source image is calculated by the target transformation function r .

6. The image stitching method combined with stitching seam fusion for improving image registration as claimed in claim 5, characterized in that: The transformation function of the target image and the source image is calculated by the global similarity matrix and the local homography matrix, specifically: Among them, S i is the global similarity matrix, a ii is the global homography matrix parameter. The similarity matrix is ​​calculated for each of the 50 matching pairs. The pair with the smallest similarity transformation angle is taken as the global similarity matrix S. * , the overlapping part is locally homographed; the global similarity matrix is ​​combined with the local homography matrix for nonlinear weighting, and the sine function is used for weighting to obtain the transformation function H of the target image t , smoothly insert the target image, and calculate the transformation function H of the source image through the target transformation function r .

7. The image stitching method combined with stitching seam fusion for improving image registration according to claim 6, characterized in that: The transformation function H of the target graph t And the transformation function H of the source image r The calculation formula is: H t =(a-0.2×sin(πa))×H * +(1-a+0.2×sin(πa))×S * The coefficient a is obtained by calculating the weight through the integral of the projection points of the target graph.

8. The image stitching method combined with stitching seam fusion for improving image registration according to claim 1, characterized in that: In S300, the two images registered in S100-S200 are searched for the best stitching seam, and the specific method includes: using a search path with the smallest difference in the overlapping parts of the two images as the stitching seam of the images for stitching, setting the overlapping parts of the two images to be I1 and I2 respectively, obtaining the grayscale and gradient difference images of the two images, performing threshold processing on the two images to obtain a binary image, and screening out the area I with a pixel of 1 d1 and I d2 , and then use the dynamic programming algorithm to get the corresponding pixels with the smallest color difference and the most similar structure. Then, trace back each row of pixels to get the best stitching seam.

9. The image stitching method combined with stitching seam fusion for improving image registration as claimed in claim 8, characterized in that: The energy function criterion of the improved optimal seam is as follows: Among them, a and b are 0.2 and 0.8 respectively, E color is the energy value of the image color difference, and the specific formula is: E geometry is the energy value of the image structure difference, and the specific formula is: E geometry (x,y)=ω2×(S x ×(I1(x,y)-I2(x,y))) 2 +ω2×(S y ×(I1(x,y)-I2(x,y))) 2 Define ω1 and ω2 as the weighting coefficients of the pixels in the color and gradient threshold regions, respectively. x , S y And use the following operator mask:

10. The image stitching method combined with stitching seam fusion for improving image registration according to claim 1, characterized in that: In S400, partition fusion is used to eliminate the influence of the stitching seams of the stitched images. The specific method includes: S401. Cut out the image of the left area from the target image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the first image to be processed; S402. Cut out the image of the right area from the source image according to the fusion range algorithm, put it into the spliced ​​image for Poisson interpolation, complete the new area fusion, and make the processed image the second image to be processed; S403. Splicing the first image to be processed and the second image to be processed according to the optimal splicing seam; S404. Perform fade-in and fade-out fusion within 10 pixels of the best stitching seam with the path with the maximum energy value as the boundary.

Images