CN104361569A - Method and device for splicing images - Google Patents

Abstract

The invention discloses a method and a device for splicing images. The method for splicing the images includes that superposed regions and transformation matrixes are determined by the aid of feature points of the first images and second images; certain pixel points in current rows in the superposed regions are used as splicing points, and transformation errors of the certain pixel points meet preset conditions; the transformation errors are errors between certain pixel point positions of the second images and matched pixel point positions in the second images, and the certain pixel points of the first images are transformed to the certain pixel point positions of the second images via the transformation matrixes; the images are spliced according to the splicing points. The method and the device have the advantage that the problem of splicing gaps when different object distances exist on spliced surfaces can be solved.

Description

Image mosaic method and device

technical field

The present application relates to the field of computer vision and image processing technology, in particular to a method and device for image stitching.

Background technique

Image stitching is an important branch in the field of computer vision and image processing. It seamlessly stitches more than two partially overlapping images to obtain a large-format or wide-view image.

The registration and stitching of adjacent images is the key to panorama generation technology. Currently, the seamless image stitching method based on SIFT feature matching is widely used. This method uses SIFT feature matching, RANSAC algorithm, L-M algorithm, and multi-resolution fusion algorithm. It overcomes the limitations of traditional image stitching technology and realizes multi-view seamless image stitching under the conditions of illumination and scale changes.

This prior art can achieve a better stitching effect when the splicing surfaces are on the same depth-of-field plane, or the distance between the object and the camera is far enough to approximate the same plane. Stitching gaps are perceived when there are image focus blurs, transformation matrix errors, and feature point detection errors. A general and effective method is to use image fusion to decompose the image to be stitched into different frequency bands for processing: taking two frequency bands as an example, the image to be fused is divided into two parts of low-frequency and high-frequency data, and the low-frequency part is processed with a larger weight. Fusion, since the low frequency corresponds to the flat part of the image, there will be no large loss of detail, and the high frequency image is blended with a smaller weight to retain the image details. Finally, restore the processed data of the two frequency bands to the time domain, and add them together to obtain the final fused image.

But the above process can only solve the small splicing gaps, such as 1-2 pixels. In the actual stitching scene, there is often a situation where there are objects with different object distances in the overlapping area of the two cameras. When the depth of field (that is, the object distance) is close and there are different depths of field, the image after stitching Often there will be obvious patchwork.

Contents of the invention

The present application provides an image stitching method and device, which can improve the stitching gap problem when there are different object distances on the stitching surface.

According to a first aspect of an embodiment of the present application, a method for image stitching is provided, the method comprising steps:

Determining an overlapping region and a transformation matrix through the feature points in the first image and the second image;

The pixel points whose transformation errors in the current line in the overlapping area meet the predetermined conditions are used as splicing points; The error between the matching pixel positions in the second image;

Image stitching is performed according to the stitching points.

According to a second aspect of the embodiments of the present application, an image mosaic device is provided, the device comprising:

An image pre-stitching module is used to determine the overlapping area and the transformation matrix through the feature points in the first image and the second image;

The splicing point extraction module is used to use pixels whose transformation errors in the current row in the overlapping area satisfy predetermined conditions as splicing points; the transformation error is the pixel point of the first image transformed into the second image through the transformation matrix an error between a pixel location and a matching pixel location in said second image;

The image splicing module is used for splicing images according to the splicing points.

Before selecting the splicing point, this application first detects whether there are objects with different object distances in the splicing area. If there are objects with different object distances, then select a pixel with a small enough transformation error as the splicing point, and avoid the object distance as much as possible. Different pixels are used as splicing points, so the spliced image will not have obvious splicing, which avoids the splicing problem caused by splicing images with different object distances.

Description of drawings

Fig. 1 is the flowchart of image mosaic method in the embodiment of the present application;

FIG. 2 is a hardware architecture diagram of an image stitching device in an embodiment of the present application;

Fig. 3 is a logic block diagram of the internal structure of the image stitching device in the embodiment of the present application;

Fig. 4 is a processing flowchart of an application example of the present application.

Detailed ways

In the image stitching technology, how to select the stitching points in the stitching area plays a vital role in the effect of image stitching. The existing technology cannot achieve a satisfactory stitching effect when there are multiple object distances in the stitching area. The present application proposes a method for image mosaic to solve this problem. During image mosaic, it is first detected whether there is an obvious difference in distance between objects in the common viewing angle of the camera (that is, in the stitching area). If there are objects with different object distances, Then try to avoid the solution of selecting splicing points for pixel points with different object distances, and use this scheme to select splicing points to realize image splicing. A detailed description will be made below in conjunction with FIG. 1 .

After careful research, the applicant found that during image stitching, when converting the pixels of one of the images to the matching pixel positions in the other image, different transformation matrices need to be used to describe its position under different object distances. Transformation relationship, which is not considered in the way of selecting splicing points in the prior art, because the transformation matrix cannot represent the transformation models of different object distances at the same time, so the spliced image often produces a large splicing seam. The applicant therefore resorts to the transformation matrix as a means of detecting whether the object distances are the same. The transformation matrix refers to the matrix used when calculating the transformation of the pixels in one image to the matching pixels in the other image in the two images to be spliced. For the convenience of description, the process of stitching two images to be stitched is taken as an example, and the stitching of multiple images to be stitched can be decomposed into a pairwise stitching process. In this application, for the convenience of description, one of the two images is called the first image, and the other image is called the second image; the transformation matrix is the transformation of the pixels of the first image to the corresponding positions of the second image The matrix used when .

In order to detect whether there are objects with different object distances in the splicing area, the method first performs feature point detection and matching on the two images to be spliced, and determines the overlapping area and the transformation matrix of the two images to be spliced according to the feature points (step 101), The overlapping area is the overlapping area determined by the feature points. This step can be obtained through existing technical means in the prior art, and the transformation matrix is based on the spatial position relationship of a pair of matching feature points in the two images. It is calculated that, for the convenience of description, in this application, the transformation matrix is the transformation matrix used to transform the feature points of the first image to the matching feature points of the second pixel. As an embodiment, the present application uses the SIFT feature matching algorithm to collect SIFT feature points in two images to be stitched for matching, and find out a pair of feature points belonging to the same physical point in the two images to be stitched, that is, the feature point and the Matching feature points of feature points. If the number of matched feature point pairs exceeds the number of variables in the transformation matrix, all the variables in the transformation matrix can be solved. In this embodiment, the transformation matrix is obtained by using the RANSAC algorithm.

This application uses the obtained transformation matrix as a technical means to check whether there are objects with different object distances. The specific process is as follows:

The overlapping region obtained in step 101 is scanned to judge whether there is a transformation error in the current row (transformation error refers to the pixel point of the first image converted to the pixel point of the second image through the transformation matrix obtained in step 101 and The error between the positions of the matching pixel points in the second image) satisfy the pixel point of the predetermined condition, and then select the pixel point satisfying the predetermined condition in the row as the splicing point (step 102), thereby avoiding objects with different object distances . It is worth pointing out that in this application, the predetermined condition is set to make the transformation error small enough so that each stitching point can be regarded as the same stitching plane during image stitching, so that the same transformation matrix can be used without causing obvious stitching Seams, so how to set it can be different according to different needs of users. In one embodiment, there may be various rules for line scanning in overlapping areas, such as the way of traversing line by line, the way of setting priority scanning areas, etc., and the user can choose flexibly according to requirements. When scanning, it is judged whether the transformation error of the current pixel satisfies the condition. If the error value does not meet the predetermined condition, continue to detect other pixels in the row, and select the pixel in the row whose transformation error meets the predetermined condition as the final splicing point.

Since the transformation error of the selected splicing points is small enough when transforming from the first image to the second image, it can be considered that these splicing points are in the same splicing plane, and the same transformation matrix can be used, so the technical means in the prior art can be used according to Image stitching is performed on these stitching points (step 103), and the stitched image will not produce obvious stitching seams.

For step 102, how to determine whether the transformation error of the pixel point meets the predetermined condition, the application provides multiple implementation methods: as an embodiment, after performing step 101 to obtain the transformation matrix, determine the error of each pixel point in the overlapping area Whether the transformation error meets the predetermined condition, and record the judgment result. The recording method can be freely selected by the user. For example, the pixels whose transformation error meets the predetermined condition can be marked as 0, and the pixels whose transformation error does not meet the predetermined condition can be marked as 0 marked as 1 and so on. After scanning all the pixels in the overlapping area, in order to optimize this scheme and further eliminate the influence of noise, the marking result can be denoised to eliminate misjudgment. The way of denoising can be through various corrosion templates such as 3*3 or 4*4 or 4*3, or it can be processed by low-pass filter and other methods. Since all the pixels in the overlapping area have been marked before step 102 is executed, in step 102 it can be known whether the pixel belongs to a pixel whose transformation error satisfies a predetermined condition according to the marked judgment result. As another embodiment, it is not necessary to mark all the pixels in the overlapping area in advance, but to judge whether the transformation error of the pixel satisfies the predetermined condition when a certain pixel is scanned. If yes, there is no need to continue to scan other pixels in the same row; if not, continue to scan other pixels in the same row.

As an example, the applicant proposed a more specific solution for how to judge whether the transformation error of the pixel points meets the predetermined conditions in the previous paragraph: due to the different positions and angles of different cameras, the position of the pixel points needs to be transferred to The same coordinate plane, therefore, this application uses the transformation matrix in step 102 to perform pixel point mapping (that is, convert the spatial coordinates of the pixels into the same coordinate plane) for each pixel in the overlapping area of the first image, and transform it into the second image , and then by comparing f _SAD with the preset value in the following formula 1, it is judged whether the pixels of the first image in the overlapping area are on the same stitching plane as the second image; as mentioned by the previous applicant, the predetermined condition set is In order to make the transformation error small enough, so that each stitching point can be regarded as the same stitching plane during image stitching, so that the same transformation matrix can be used without causing obvious stitching seams, so the preset value can be changed according to the accuracy required by the user. Different setting methods, for example, the preset value can be set to W*W*0.1*W _MEAN , where W _MEAN is the average value of the image brightness of the pixels of the second image in the window; Considering the extreme case of misjudgment, the preset value can be set to 0 to simplify the solution.

$f_{\mathrm{SAD}}(x,\mathrm{the\; y},d1,d2)=\underset{i\∈W}{\mathrm{\Σ}}\underset{j\∈W}{\mathrm{\Σ}}|T\left(I_1(x+i,\mathrm{the\; y}+j)\right)-I_2(x+i+d1,\mathrm{the\; y}+j+d2)|$ (Formula 1)

Wherein, f _SAD is the pixel in the first image as the sum of the accumulated errors of the matching pixels of the second image transformed from each pixel in the sliding window of the center position; W is the size of the sliding window, and T represents the transformation matrix , d1 refers to the image offset of the image pixel point of I ₁ mapped to the image rear abscissa of I ₂ ; d2 refers to the image offset of the image pixel point of I ₁ mapped to the image rear ordinate of I ₂ ; I ₁ means the first image, I ₂ means the second image. This application uses the sliding window method to detect different object distances in the window to be stitched in consideration of the influence of image noise, lens distortion and other factors. Even pixels in the same transformation matrix have differences within a certain range. As an empirical value obtained by the applicant through trial and error, a preferred value range of W is 3 to 5 pixels. If there is no need to consider the influence of factors such as image noise and lens distortion, the value of W can be set to 1 , which represents the situation where no sliding window is set.

It is worth mentioning that in practice, there may be no pixel points whose transformation error values meet the predetermined conditions in some lines in the overlapping area where the splicing point is located. At this time, in order to avoid objects with different object distances as much as possible, As a further improvement of the above method, the pixel point with the minimum value of f _SAD in the row can be used as a splicing point.

The above is a detailed description of the image stitching method in this application. This application also provides an image stitching device corresponding to the above method, which can be applied to PCs, mobile phones, or camera devices with sufficient processing capabilities. on a hardware device. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking the implementation of software installed on a PC as an example, as a device in a logical sense, it is formed by reading the corresponding computer program instructions in the non-volatile memory into the memory for operation by the processor of the device where it is located. From the perspective of hardware, as shown in Figure 2, it is a hardware structure diagram of the PC where the image stitching device of the present application is located. In addition to the processor, network interface, internal memory and non-volatile memory shown in Figure 2, the implementation The PC where the device in the example is located may generally include other hardware, which is not shown in detail in FIG. 2 .

Fig. 3 is a logical block diagram of the internal structure of the image stitching device. The device includes an image pre-splicing module, a splicing point extraction module and an image splicing module.

The function of the image pre-stitching module is to determine the overlapping area and the transformation matrix of the image to be stitched determined by the feature points. The splicing point extraction module in the device completes the function of judging whether there are pixels with different object distances and how to determine the splicing point. The splicing point extraction module determines whether there are objects with different object distances by judging whether the transformation error of each pixel point in the current line meets the predetermined condition. If the transformation error of the current pixel point in the overlapping area does not meet the predetermined condition, the current Scan the other pixels in the row where the pixel is located, and take the pixel whose transformation error meets the predetermined condition as the splicing point; There are many ways to meet the predetermined conditions. You can first record the transformation error value judgment results of all pixels, and then directly call the judgment results when scanning pixels; you can also scan to a certain pixel and then calculate the transformation of the pixel. The error value is used to judge, and the detailed process can refer to the corresponding part above, so I won’t repeat it here. After the splicing point extraction module finally determines the splicing point, the image splicing module completes the image splicing process according to the splicing point.

Other methods for determining the splicing point by the splicing point extraction module may include: if there is no pixel point whose transformation error value satisfies the predetermined condition in a row, the pixel point with the smallest value of f _SAD in the row is used as the splicing point. Due to the same inventive concept as the above image mosaic method, no further details are given here.

Also based on the same inventive concept, the splicing point extraction module can also judge whether the transformation error value of the pixel meets the predetermined condition by comparing f _SAD in formula 1 with the preset value; for the specific judgment method, please refer to the description of the relevant part above .

An application example of the present application is listed below in conjunction with the flow chart of FIG. 4 , so as to have a clearer understanding of the present application.

Step 401: Perform feature point detection and matching on the two images to be stitched, and calculate a transformation matrix for transforming pixels of the first image to corresponding pixels of the second image, and determine the overlapping area of the two images;

Step 402: According to the overlapping area obtained in step 401, use the obtained transformation matrix to map each pixel in the overlapping area in the first image to the second image.

Step 403: Determine the range W of the sliding window for the pixels in the overlapping area in the mapped image, scan pixel by pixel, and judge the transformation error f _SAD ; Record the results, and in the marking diagram, if f _SAD is less than or equal to the preset value, the pixel belonging to the transformation model is marked as 0, otherwise, it is recorded as 1, indicating that the current pixel is a different object distance.

Step 404: Scan the pixels in the overlapping area in the mapped image one by one, and judge whether the position in the marker map corresponding to the current pixel is 1 according to the marker map obtained in step 403;

Step 405: If the position in the marker map corresponding to the current pixel is 0, use this pixel position as a stitching point.

Step 406: If the position in the marker map corresponding to the current pixel is 1, continue to search for other pixels in the current row, and search for the pixel whose position is 0 in the marker map;

Step 407: Use the searched position of the pixel whose marker image is 0 in the row as the splicing point of the current row.

Step 408: If all the marker images belonging to the overlapping area in a row are 1, then take the pixel position with the smallest f _SAD value as the splicing point.

Other embodiments of the present application will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the application, these modifications, uses or adaptations follow the general principles of the application and include common knowledge or conventional technical means in the technical field not disclosed in the application . The specification and examples are to be considered exemplary only, with a true scope and spirit of the application indicated by the following claims.

It should be understood that the present application is not limited to the precise constructions which have been described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (10)

1. A method for image stitching, characterized in that the method comprises steps: Determining an overlapping region and a transformation matrix through the feature points in the first image and the second image; The pixel points whose transformation errors in the current line in the overlapping area meet the predetermined conditions are used as splicing points; The error between the matching pixel positions in the second image; Image stitching is performed according to the stitching points. 2. The method according to claim 1, characterized in that: the process of using the pixels whose transformation error meets the predetermined condition in the current line in the overlapping area as the splicing point comprises: Judging whether the transformation error of each pixel in the overlapping area satisfies the predetermined condition, and recording the judgment result; According to the recorded judgment results, the pixel points in the current row that satisfy the predetermined condition are used as the splicing points. 3. method according to claim 1, is characterized in that, compares with preset value by f _SAD in the following formula, judges whether described transformation error meets described predetermined condition; ${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; SAD</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$ Wherein, f _SAD is the pixel point in the first image as the sum of error accumulation between the pixel point position in the second image and the matching pixel point position in the second image for each pixel point in the sliding window of the center position; W is the size of the sliding window, T represents the transformation matrix, d1 refers to the image offset of the abscissa after the image pixel of I ₁ is mapped to the image of I ₂ ; d2 refers to the image pixel of I ₁ mapped to I ₂ The image offset of the ordinate after the image; I ₁ refers to the first image, and I ₂ refers to the second image. 4. method according to claim 3, is characterized in that, this method also comprises the step: If there is no pixel point whose transformation error satisfies the predetermined condition in the current row, the pixel point with the smallest value of f _SAD in the row is used as a splicing point. 5. The method according to claim 3, wherein the preset value is W*W*0.1*W _MEAN , wherein W _MEAN is the mean value of the image brightness of the second image in the window, and W is The size of the sliding window. 6. A device for image stitching, characterized in that the device comprises: An image pre-stitching module is used to determine the overlapping area and the transformation matrix through the feature points in the first image and the second image; The splicing point extraction module is used to use the pixels whose transformation errors in the current row in the overlapping area meet the predetermined conditions as splicing points; an error between a pixel location and a matching pixel location in said second image; The image splicing module is used for splicing images according to the splicing points. 7. The device according to claim 6, characterized in that: the splicing point extraction module The process of using the pixels whose transformation error meets the predetermined condition in the current line in the overlapping area as the splicing point includes: Judging whether the transformation error of each pixel in the overlapping area satisfies the predetermined condition, and recording the judgment result; According to the recorded judgment result, the pixel points in the current row that satisfy the predetermined condition are used as the splicing points. 8. The device according to claim 6, wherein the splicing point extraction module judges whether the transformation error satisfies the predetermined condition by comparing f _SAD with a preset value in the following formula; ${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; SAD</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; T</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>$ Wherein, f _SAD is the pixel point in the first image as the sum of error accumulation between the pixel point position in the second image and the matching pixel point position in the second image for each pixel point in the sliding window of the center position; W is the size of the sliding window, T represents the transformation matrix, d1 refers to the image offset of the abscissa after the image pixel of I ₁ is mapped to the image of I ₂ ; d2 refers to the image pixel of I ₁ mapped to I ₂ The image offset of the ordinate after the image; I ₁ refers to the first image, and I ₂ refers to the second image. 9. The device according to claim 8, wherein the splicing point extraction module is also used for: If there is no pixel point whose transformation error satisfies the predetermined condition in the current row, the pixel point with the smallest value of f _SAD in the row is used as a splicing point. 10. The device according to claim 8, wherein the preset value is W*W*0.1*W _MEAN , wherein W _MEAN is the mean value of the image brightness of the second image in the window, and W is The size of the sliding window.