Image splicing method and system, and use thereof

Abstract

An image splicing method, comprising: (1) extracting any two images, having an overlapping area, to a canvas; (2) respectively identifying feature points in the two images extracted in step (1) and the features of the feature points; (3) on the basis of the similarity of the feature of each feature point, carrying out matching on the feature points in the two images to obtain all feature point pairs that are successfully matched, and extracting the feature point pairs; (4) on the basis of the feature point pairs extracted in step (3), acquiring a matching relationship between the two images, and on the basis of the matching relationship, splicing the two images; and (5) repeating steps (1)-(4), and extracting two images different from the previous two images each time for processing until all the images are spliced.

Description

Image stitching method, system and application thereof Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an image stitching method, system and application thereof. Background Art

[0002] Depending on the medical scenario, some clinical judgments need to be based on images of larger areas of human tissues, organs, or bones. For example, the observation of scoliosis needs to be based on medical images of the entire spine. DR (Digital Radiography) is a digital X-ray imaging technology that can be used to obtain images of the entire spine. During the DR imaging process, the subject needs to stand in front of the DR machine and maintain a specific posture while the DR device takes pictures. However, DR is not suitable for target image acquisition in surgical situations or situations where timely verification is required, because the operation and image acquisition of the target in these situations require the target position and posture to remain unchanged. O-type X-ray machines can capture large-scale target images, but due to the large space and high price of O-type X-ray machines, they are not suitable for all medical scenarios.

[0003] Mobile C-arm X-ray machines are a common medical imaging device, offering low cost, compact size, portability, and low radiation exposure. They are suitable for capturing target images in these situations. However, due to limitations in their field of view (FOV), C-arm X-ray machines cannot capture images of large areas of tissue, such as the entire spine.

[0004] This problem can be solved by image stitching. Image stitching generally starts by obtaining matching feature point pairs between two images, then obtains the matching relationship between the two images based on the feature point pairs, and finally stitches the images based on the matching relationship. However, there are the following problems in the image stitching process: (1) In the matching process of feature point pairs, there are many mismatching points, resulting in low accuracy of image matching and stitching; (2) In the acquisition of image matching relationships, there may be mismatches due to inaccurate matching of feature point pairs themselves, or there may be positional relationships of multiple pairs of feature point pairs that are not completely consistent, and there are uncertainty errors in the positional relationship calculation process; (3) In the image stitching process, when there are a large number of images, the accumulated errors will lead to poor stitching results, which is mainly reflected in the fact that the stitching of the later images is more abnormal, and there may even be a "stitching fly" situation.

[0005] Summary of the Invention

[0006] In response to the above-mentioned shortcomings, the present invention provides an image stitching method, system and application thereof, which improve the accuracy of feature point matching, the fault tolerance of erroneous feature point pairing, and reduce the errors caused by uncertain factors in positional relationship calculations, thereby quickly and accurately completing automatic image stitching during the image stitching process.

[0007] Technical solution:

[0008] A method for extracting feature point pairs, comprising the steps of:

[0009] Identify feature points in the first image and the second image respectively, and obtain the position and features of each feature point;

[0010] Matching the feature points in the first image with the feature points in the second image based on the similarity of the features of each feature point, and marking the feature point pairs that have successfully matched;

[0011] reducing noise on the successfully matched feature point pairs according to a motion range of an image acquisition device and / or a distribution of the feature point pairs;

[0012] The feature point pairs after denoising are marked as extracted feature point pairs.

[0013] Specifically, according to the motion range of the image acquisition device, denoising the successfully matched feature point pair includes:

[0014] According to the movement specification of the image acquisition device and the initial positioning of the two images, a constraint model is constructed to filter the feature point pairs. The constraint model is: j -y j ′|<t1*r+t2*r (1) |x j -x j ′|<t3*c+t4*c (2)

[0015] Among them, (x j ,y j ) and (x j ′,y j ′) are the coordinates of the two feature points of the successfully matched feature point pair, r is the height of the image, and c is the width of the image; t1 and t2 are the restriction coefficients related to the y-direction translation distance and translation stability of the image acquisition device, and t3 and t4 are the coefficients related to the x-direction translation distance and translation stability of the image acquisition device; the x-direction and y-direction of the image acquisition device correspond to the x-direction and y-direction of the image, respectively.

[0016] More specifically, where 0≤t1<1, 0≤t3<1, 0 <t2<0.5,0<t4<0.5。

[0017] Specifically, denoising the successfully matched feature point pairs according to the distribution of the feature point pairs includes:

[0018] Obtain the positional relationship between the two corresponding feature points in each successfully matched feature point pair;

[0019] According to the positional relationship, each feature point is converted into a point to obtain a transformation point;

[0020] Cluster the transformation points, identify and retain the feature point pairs corresponding to the category with the largest number of transformation points.

[0021] More specifically, clustering the conversion points is as follows:

[0022] 1) drawing a histogram of the transformation points;

[0023] 2) Randomly select num points from the transformation points in the histogram as the initial centers of each category;

[0024] 3) According to the distance from each transformation point to the current center of each category, each transformation point is classified into the center with the smallest distance L, and num initial clusters are obtained, and the initial clustering factor T of each category is obtained.

[0025] 4) Update the current center of each category according to the average position of the transformation points in each category of the current num clusters, and obtain the current clustering factor T' corresponding to the current center of each category;

[0026] 5) Calculate the difference ΔT between the initial clustering factor T and the current clustering factor T', and determine whether the corresponding difference ΔT of each category is less than the set threshold. Otherwise, return to step 2); if so, end clustering.

[0027] A method for obtaining an image matching relationship, comprising:

[0028] S1, extracting feature point pairs from two images with overlapping areas;

[0029] S2, building a matching relationship model between the two images;

[0030] S3. Select different feature point pairs from the feature point pairs obtained in S1, calculate the image matching relationships corresponding to the different feature point pairs based on the selected feature point pairs and the matching relationship model obtained in S2, count the number of feature point pairs that are suitable for each image matching relationship, and identify the optimal image matching relationship based on the number.

[0031] Specifically, identifying the optimal image matching relationship according to the quantity includes:

[0032] Comparing the numbers of feature point pairs that are suitable for the matching relationships of the respective images, and identifying the optimal image matching relationship based on the comparison results; or,

[0033] The number of feature point pairs that fit the matching relationship of each image is compared with the threshold, and the optimal image matching relationship is identified based on the comparison result.

[0034] More specifically, the numbers of feature point pairs that are suitable for each image matching relationship are compared with each other, and the image matching relationship with the largest number of feature point pairs that are suitable for each image matching relationship is determined as the optimal image matching relationship.

[0035] Specifically, the matching relationship model between the two images is:

[0036] Among them, r1, r2, r3, and r4 are model parameters related to rotation, and tx and ty are model parameters related to translation.

[0037] More specifically, calculating the image matching relationships corresponding to the selected different feature point pairs includes:

[0038] At least three pairs of feature point pairs from the feature point pairs obtained in S1 are randomly selected to form a group of feature point pairs, and restrictive conditions are constructed with the matching relationship model between the two images obtained in S2, and the model parameters are calculated to obtain the matching relationship between the two images.

[0039] More specifically, the restrictive conditions also include the following constraints:

[0040] where θ∈[3°,5°].

[0041] Specifically, the number of feature point pairs adapted to the matching relationship of each image is statistically determined as follows:

[0042] Construct the energy equation:

[0043] Among them, (xj, yj) and (xj′, yj′) are the coordinates of any feature point pair other than the selected feature point, and H is the matching relationship between the two images;

[0044] Substitute all feature point pairs except the selected ones into the energy equation to obtain energy, and record the number of feature point pairs whose energy values ​​are less than the set threshold, which is the number of point pairs that meet the matching relationship.

[0045] An image stitching method, comprising:

[0046] (1) Extract any two images with overlapping areas to the canvas;

[0047] (2) identifying the feature points and features in the two images extracted in step (1);

[0048] (3) Match the feature points in the two images according to the similarity of the features of each feature point, obtain all the successfully matched feature point pairs, and extract the feature point pairs;

[0049] (4) obtaining a matching relationship between the two images based on the feature point pairs extracted in step (3), and splicing the two images based on the matching relationship;

[0050] (5) Repeat steps (1) to (4), extracting two images different from the previous ones each time for processing, until all images are stitched together.

[0051] Specifically, it also includes fine-tuning any one of the images stitched in step (5), wherein the fine-tuning includes translation and / or rotation step fine-tuning, and the step direction is divided into forward and reverse.

[0052] More specifically, the fine-tuning is:

[0053] The positional relationship before and after image fine-tuning is: pimg = pscreen*Fn1*(F-1)n2*Tm1*(T-1)m2, pimg is the position after image fine-tuning, pscreen is the position before image fine-tuning; n1 is the cumulative number of positive translation steps of the monitored image, n2 is the cumulative number of negative translation steps of the monitored image, m1 is the cumulative number of positive rotation steps monitored, and m2 is the cumulative number of negative rotation steps monitored; F is the forward translation step relationship, F-1 is the reverse translation step relationship, T is the forward rotation step relationship, and T-1 is the reverse rotation step relationship.

[0054] More specifically, after obtaining the panoramic stitching image, the fusion step is also included:

[0055] For any point in the overlapping area after stitching, its pixel is recalculated and updated. The pixel calculation formula is: pix=k1*pix1+k2*pix2;

[0056] Among them, pix1 is the pixel value of the corresponding point in the first image, pix2 is the pixel value of the corresponding point in the second image, k1 is the first weight coefficient representing the proximity of the overlapping boundary between point W and the first image, and k2 is the second weight coefficient representing the proximity of the overlapping boundary between point W and the second image.

[0057] An image stitching unit, comprising:

[0058] The feature extraction module is used to extract feature points on two images with overlapping areas and transmit the extracted feature point information to the feature matching module;

[0059] A feature matching module is used to match the feature points of the two images extracted by the feature extraction module, obtain feature point pairs, extract feature point pairs, and transmit the extracted feature point pair information to the image matching module;

[0060] The image matching module calculates the matching relationship between the two images based on the feature point pairs, and accordingly shifts and splices the subsequent time sequence image onto the previous time sequence image to obtain a panoramic stitching image.

[0061] Specifically, the feature extraction module extracts feature points from two images with overlapping areas using a SURF algorithm or a SIFT algorithm.

[0062] More specifically, the feature matching module matches the feature points of the two images extracted by the feature extraction module to obtain feature point pairs:

[0063] Calculate the similarity of the feature vectors between the feature points of one image and the feature points of the other image in the two images, and compare the calculation result with the similarity threshold. According to the comparison result, determine whether the two feature points can be paired, and obtain the feature point pairs that can be paired.

[0064] An image stitching system, comprising:

[0065] An image acquisition unit, which acquires an image of a target in a directional translation state;

[0066] The image stitching unit, as described above, extracts the images acquired by the image acquisition unit and automatically stitches the images to obtain a stitched image;

[0067] An image display unit displays the spliced ​​image.

[0068] Specifically, it also includes an image pre-processing unit, which pre-processes the image before extracting it to the image stitching unit. The pre-processing includes at least one of window position and width adjustment, image rotation, image sorting, image selection, and image deletion.

[0069] Specifically, an indication unit is further included, and the indication unit includes a direction indication unit indicating the directional translation direction of the image acquisition unit and / or a distance indication unit indicating the translation distance of the image acquisition unit.

[0070] Specifically, the distance indicating unit includes a height detector for detecting the height of the bed where the target is located and a prompter for calculating the critical distance of the directional translation of the image acquisition unit according to the height of the bed.

[0071] More specifically, the prompter calculates the critical distance of the directional translation of the image acquisition unit according to the bed height detected by the height detector, and the calculation formula is: d = F[g(h)]

[0072] Where d is the critical distance of the directional translation of the image acquisition unit, beyond which the two images will have no overlapping area; function g is the mapping relationship between the bed height h and the image range displayed on the imaging detector of the image acquisition unit; and F is the mapping relationship between the image displayed on the imaging detector and the actual imaging distance.

[0073] A computer storage medium stores a computer program, which implements the steps of the above method when executed.

[0074] A C-arm machine comprises the above-mentioned system.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] 1. In the present invention, the user only needs to drag the image acquisition unit a few times according to the operating specifications to capture images until the image covers all the areas to be captured of the target. Multiple images can be automatically and quickly stitched into a panoramic image, which can overcome the problem of limited field of view of the image acquisition device.

[0077] 2. The present invention obtains feature point pairs by extracting and matching feature points on two images, and corrects the feature point pairs multiple times with the aid of factors such as image acquisition shift and feature point pair distribution. It can accurately extract feature point pairs in the two images, and then realize rapid splicing of images with overlapping areas.

[0078] 3. The present invention calculates the optimal matching relationship between two images through multiple iterations, which can eliminate outlier point pairs in the feature point pairs and reduce the interference of outlier points on the image matching relationship, thereby improving the tolerance to incorrect pairing of feature point pairs and reducing the errors caused by uncertain factors in positional relationship calculations, thereby quickly and accurately completing automatic image stitching during the image stitching process.

[0079] 4. After the image stitching unit of the present invention adds a fine-tuning module and fine-tuning steps, it can fine-tune any of the stitched images, including translation fine-tuning and rotation fine-tuning, to update the positional relationship between the two images, thereby greatly improving the operating experience and stitching accuracy, and effectively avoiding the occurrence of the "stitching flight" phenomenon.

[0080] 5. After the stitching is completed, the present invention introduces a fusion module and a fusion method that integrates the two factors of pixels and positions, so that the overlapping areas transition naturally, making the stitching visual effect clearer and softer. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] FIG1 is a structural diagram of an image stitching unit of the present invention;

[0082] FIG2 is a diagram illustrating the architecture of an image stitching system according to the present invention;

[0083] FIG3 is a flow chart of extracting feature point pairs according to the present invention;

[0084] FIG4 is a flow chart of obtaining a matching relationship between images according to the present invention;

[0085] FIG5 is a flow chart of the image stitching method of the present invention;

[0086] FIG6 is a schematic diagram of image fusion in the present invention. DETAILED DESCRIPTION

[0087] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0088] The present invention provides an embodiment of an image stitching unit for automatically stitching two images with overlapping areas, as shown in FIG1 , comprising:

[0089] A feature extraction module is used to extract feature points from two images with overlapping areas, preferably from two adjacent images, and transmit the extracted feature point information to a feature matching module;

[0090] A feature matching module is used to match the feature points of the two images extracted by the feature extraction module, obtain feature point pairs, and transmit the feature point pair information to the image matching module;

[0091] The image matching module calculates the matching relationship between the two images based on the feature point pairs, and accordingly shifts and splices the subsequent time sequence image onto the previous time sequence image to obtain a panoramic stitching image.

[0092] In an exemplary embodiment, the image stitching unit further includes a fine-tuning module to facilitate the user to fine-tune the position of each image in the stitched panoramic stitched image, including horizontal fine-tuning and rotation fine-tuning.

[0093] In an exemplary embodiment, the image stitching unit further includes an image fusion module for fusing overlapping areas of the panoramic stitching image output by the image matching module or the panoramic stitching image fine-tuned by the fine-tuning module to remove overlapping images in the overlapping areas.

[0094] In this embodiment, the feature extraction module uses the SURF (Speeded Up Robust Features) algorithm or the SIFT algorithm to extract feature points from two images with overlapping regions. The present invention may also use other feature point recognition algorithms, not limited to the two algorithms described above. A feature point recognition algorithm provides a feature vector for each feature point to represent its characteristics.

[0095] In this embodiment, the feature matching module matches the feature points of the two images extracted by the feature extraction module according to the similarity of the features of each feature point, and marks all successfully matched feature point pairs.

[0096] In an exemplary embodiment, the similarity of the feature vectors of each feature point is calculated, and the calculation result is compared with a similarity threshold. Based on the comparison result, it is determined whether each feature point can be paired, and the image stitching unit records the coordinates of the two feature points that can be paired.

[0097] The present invention further provides an embodiment of an image stitching system, as shown in FIG2 , comprising:

[0098] An image acquisition unit, used for acquiring an image of a target in a directional translation state;

[0099] In this embodiment, the image acquisition unit is specifically a mobile C-arm X-ray machine having a transmitter and a flat-panel detector; the target may be a human body. Thus, illustratively, the mobile C-arm X-ray machine performs directional translation along a direction perpendicular to the plane of the C-arm to acquire an image of the target.

[0100] In this embodiment, an image storage unit may also be included, for storing the images captured by the image capturing unit.

[0101] The aforementioned image stitching unit;

[0102] In one exemplary embodiment, an image pre-processing unit is provided between the image storage unit and the image stitching unit. This pre-processing unit is configured to perform preliminary processing on the images stored in the image storage unit. This processing includes adjusting window levels and widths, image rotation and / or image sorting, image selection, image deletion, and the like. The processed images are then transmitted to the image stitching unit. Specifically, the window levels and widths are adjusted to achieve an ideal single image display effect; image rotation and / or image sorting are used to eliminate image deflection or disorder, thereby increasing fault tolerance for operational errors; and image selection and image deletion are used to select the optimal image for stitching.

[0103] The direction indicating unit may be a laser indicating unit, provided on the transmitter or the flat probe, and used to form a mark on the target, preferably a cross mark to indicate the directional translation direction of the image acquisition unit;

[0104] In an exemplary embodiment, a distance indicating unit is further included for detecting the height of the target bed in real time and indicating the directional translation distance of the image acquisition unit accordingly. The image acquisition unit can perform directional translation operations according to the directional translation direction indicated by the direction indicating unit and the directional translation distance indicated by the distance indicating unit.

[0105] Specifically, the distance indicating unit includes a depth detector and a prompter. The depth detector can be a potentiometer or a depth camera, etc.; wherein the potentiometer or depth camera is used to detect the height of the bed where the target is located, and the prompter calculates and indicates the directional translation distance of the image acquisition unit based on the detected bed height. Among them, the image stitching of the present invention is based on the premise that the images have overlapping areas. Due to the scaling characteristics of the vertebral beam, the closer the target to be imaged, such as a human body, is to the transmitter, the smaller the imaging range, and vice versa. If the same distance is moved each time, then targets of different shapes will obtain images with different overlapping ranges. Therefore, the critical step of the directional translation of the image acquisition unit can be calculated based on the bed height, specifically: d = F[g(h)]

[0106] Where d is the critical distance of the directional translation of the image acquisition unit, beyond which the two images will have no overlapping area; the function g is the mapping relationship between the bed height h and the image range displayed on the imaging detector of the image acquisition unit, such as the flat-panel detector of a mobile C-arm X-ray machine; and F is the mapping relationship between the image displayed on the flat-panel detector and the actual imaging distance; the bed height h can be obtained in various ways, such as the aforementioned potentiometer measurement or depth camera detection.

[0107] The image display unit is used to display the panoramic image stitched by the image stitching unit.

[0108] In an exemplary embodiment, a motion control unit is further included, which is interconnected with the direction indication unit and / or the distance indication unit, and is used to control the directional translation of the image acquisition unit according to the indication or the user's setting based on the indication, such as setting the step size and setting the direction to control the directional translation of the image acquisition unit.

[0109] In one exemplary embodiment, a motion control unit is not required; the user can simply manually drag the image acquisition unit for directional translation according to the direction indication unit. Alternatively, the image acquisition unit can be directional translated in conjunction with the directional translation distance indicated by the distance indication unit. Specifically, the user controls the image acquisition unit to translate in the target direction and uses a bedside scale (not shown) and a laser indicator unit to determine whether the unit has been translated into position. The specific movement and imaging are shown in the image acquisition unit in FIG2 . As can be seen in the figure, the two vertebral beams have appropriate overlap to ensure that the two images also have appropriate overlap.

[0110] The present invention also provides an embodiment of a method for extracting feature point pairs, as shown in FIG3 , comprising the steps of:

[0111] A1. Extract any two images with overlapping areas (hereinafter referred to as a first image P1 and a second image P2), identify feature points in the first image and the second image respectively, and mark the feature points.

[0112] In an exemplary embodiment, the feature points in the two images are identified by using a SURF (Speeded Up Robust Features) algorithm or a SIFT algorithm. The above algorithm provides a feature vector for each feature point to represent its features.

[0113] A2. Match the feature points in the first image with the feature points in the second image according to the similarity of the feature vectors, mark the successfully matched feature point pairs, and obtain a first feature point pair set Q1.

[0114] Specifically include:

[0115] A21. Calculate the difference coefficient dij of the feature vectors of each feature point pi in the first image and each feature point pj in the second image. The calculation formula is:

[0116] Among them, b iw and b jw The w-th dimension of a feature point in the first feature point set P1 and a feature point in the second feature point set P2 is represented respectively, where W represents the dimension of the feature vector of the feature point. The smaller the difference coefficient, the higher the similarity between the two feature points.

[0117] A22, the above difference coefficient d ij Compare with the threshold and determine whether the match is successful based on the comparison result. If it is less than the threshold, the match is successful. Otherwise, the feature point pair is eliminated, the two feature points that are successfully matched are marked, and their positional relationship is obtained.

[0118] In an exemplary embodiment, obtaining the positional relationship U of two successfully matched feature points includes: obtaining the coordinate difference Δx and Δy of the two successfully matched feature points; more specifically, connecting the two successfully matched feature points and calculating the x- and y-direction components Δx and Δy of the connecting line. The two successfully matched feature points constitute a feature point pair, and all successfully matched feature point pairs constitute a first feature point pair set Q. Each feature point pair U in the first feature point pair set Q k Use U k (Δx k ,Δy k ) as an identity marker.

[0119] In an exemplary embodiment, the threshold is obtained by the following method: sorting all difference coefficients dij calculated by A21 from small to large to obtain the maximum difference coefficient D; taking eD as the threshold, e as the proportional coefficient, and further, e<0.6.

[0120] A3. According to the motion range of the image acquisition device, denoise the feature point pairs in the first feature point pair set Q1 to obtain a second feature point pair set Q2.

[0121] Specifically include:

[0122] A31. Construct a constraint model based on the motion range of the image acquisition device. The constraint model is: |y j -y j ′|<t1*r+t2*r (1) |x j -x j ′|<t3*c+t4*c (2)

[0123] Wherein, Δy=yj-yj′, Δx=xj-xj′, (xj, yj) and (xj′, yj′) are the coordinates of the two feature points of the successfully matched feature point pair, r is the height of the image, and c is the width of the image; t1 and t2 are the restriction coefficients related to the y-direction translation distance and translation stability of the image acquisition device, and t3 and t4 are the coefficients related to the x-direction translation distance and translation stability of the image acquisition device; wherein the x-direction and y-direction of the image acquisition device correspond to the x-direction and y-direction of the image, respectively, wherein 0≤t1<1, 0≤t3<1, 0 <t2<0.5,0<t4<0.5。

[0124] In an exemplary embodiment, the image acquisition device moves forward along the y-axis to acquire an image, and the positional relationship of the feature point pair should satisfy the above-mentioned models (1) and (2) and Δy>0, wherein t1 takes a value of 0.5 to 0.8, t3 takes a value of 0, and t2 and t4 take values ​​of 0.1 to 0.3. Furthermore, the image acquisition device moves forward along the y-axis with a step length of 0.8r to acquire an image, and the movement range of the image field in the y-axis direction is equal to 0.8r, and the movement accuracy in the y-axis direction has a deviation within 0.1r. The x-axis does not move, and the x-axis is maintained within a deviation within 0.1c, wherein t1=0.8, t2=0.1, t3=0, and t4=0.1.

[0125] In another exemplary embodiment, the image acquisition device moves in the negative direction along the image x direction to acquire the image, the feature point correspondence satisfies the above-mentioned models (1), (2) and Δx<0, t1 takes a value of 0, t3 takes a value of 0.5~0.8, and t2 and t4 take values ​​of 0.1~0.3.

[0126] A32. Determine whether all feature point pairs in the first feature point pair set Q1 satisfy the constraint conditions of step A31, and obtain the second feature point pair set Q2 based on the judgment result, that is, mark the feature point pairs Uk' that satisfy the constraint conditions to form the second feature point pair set Q2, and the feature points that do not satisfy the constraint conditions are considered to be unqualified feature point pairs and are eliminated.

[0127] In an exemplary embodiment, the feature point pair Uk' in the second feature point pair set Q2 is marked as the final feature point pair.

[0128] In an exemplary embodiment, the method further includes:

[0129] A4. Cluster all feature point pairs Uk' in the second feature point pair set Q2, and mark the feature point pair Uk" corresponding to the category with the largest number of feature point pairs to obtain the third feature point pair set Q3.

[0130] Specifically include:

[0131] A41. Transform each feature point pair Uk' in the second feature point pair set Q2 based on the positional relationship between the two feature points to obtain a transformation point Zk'. Specifically, based on the feature point pair identity label Uk(Δxk, Δyk), extract the positional relationship (Δxk', Δyk') of each feature point pair Uk' in the second feature point pair set Q2 as the coordinates of the transformation point Zk'.

[0132] A42. Draw a histogram of the transformation points Zk'. Specifically, place all transformation points Zk' in the second feature point pair set Q2 into a rectangular coordinate system. This creates a distribution pattern, with point pairs with similar transformation points Zk' clustering in the same area.

[0133] A43. Randomly select num points from the transformation points Zk' in the plane rectangular coordinate system of the above histogram as the initial centers of each category.

[0134] Further, Wherein N is the number of feature point pairs Uk' in the second feature point pair set Q2, that is, the number of transformation points Zk'. When this value is selected, grouping can be performed effectively, and the accuracy of classification of feature point pairs Uk' is high.

[0135] A44. Classify each transformation point Zk' according to the distance from each transformation point Zk' to the current center of each class, obtain num initial clusters, and obtain the initial clustering factor T of each class.

[0136] The classification of each transformation point Zk' is to calculate the distance between the transformation point Zk' and each current center, and assign the transformation point Zk' to the center with the smallest distance L, until all transformation points Zk' are assigned to obtain the initial clustering, and calculate the initial clustering factor T for each category, which is calculated as follows:

[0137] Among them, Lm is the distance from the mth transformation point to the corresponding center in the class, and n is the number of transformation points Zk' in the class.

[0138] A45. Update the current centers of each class according to the positions of the transformation points Zk' in each of the current num clusters, and obtain the current clustering factors T' corresponding to the current centers of each class (i.e., the centers of each class after this step of updating).

[0139] Here, updating the current class center is to determine the current class center based on the positions of all transformation points Zk' in the current class. More specifically, the average coordinate value of all transformation points Zk' in the current class is calculated, and the current class center position is updated based on the average coordinate value.

[0140] A46. Calculate the difference ΔT between the initial clustering factor T and the current clustering factor T', and determine whether the corresponding difference ΔT of each category is less than the set threshold. Otherwise, return to A44; if so, end clustering.

[0141] A47. Count the number of each type of transformation points Zk' in the current clustering result. The feature point pairs Uk" corresponding to the class with the largest number of transformation points constitute the third feature point pair set Q3.

[0142] A58. Mark the feature point pairs in the third feature point pair set Q3 as the final recognized feature point pairs.

[0143] The present invention also provides an embodiment of a method for obtaining an image matching relationship, as shown in FIG4 , comprising:

[0144] B1. Extract feature point pairs corresponding to two images with overlapping areas, and identify the positions of two feature points in each feature point pair.

[0145] In this embodiment, the feature point pair extraction method of the prior art may be used, or the feature point pair extraction method described above may be used.

[0146] B2. Establish image matching relationship model;

[0147] Specifically, the position of the feature points is expressed in the matrix Represented by the homography matrix is the image matching relationship model, where r1, r2, r3, and r4 are the model parameters related to rotation, and t x With t y is the model parameter related to translation;

[0148] Theoretically, the coordinates of the two points in each feature point pair should satisfy:

[0149] B3. Construct restrictive conditions of the image matching relationship model based on the two-point position matrix of each feature point pair in B1 and the image matching relationship model in B2, calculate the model parameters based on the restrictive conditions, and obtain the image matching relationship.

[0150] Specifically, among the feature point pairs described in B1, M pairs of feature point pairs are randomly selected, 3≤M≤N; restrictive conditions are constructed based on the M pairs of feature point pairs and the above-mentioned image matching relationship model, and the model parameters are calculated based on the restrictive conditions to obtain the corresponding image matching relationship H1.

[0151] In an exemplary embodiment, the restrictive conditions of the image matching relationship model are:

[0152] The above equations are solved by SVD algorithm (when M>3, it is an overdetermined equations), and the model parameters r1, r2, r3, r4, t are calculated. x , t y The determined value of , and then the image matching relationship H1 is obtained.

[0153] In the present invention, in addition to the above restrictive conditions, in order to further overcome the errors caused by uncertain factors in the SVD algorithm solution and other operations, the following constraints are also included:

[0154] Among them, r1 2 +r2 2 = 1 ensures the orthogonality of the rotation matrix, and arccos(r1) < θ / (180 / π) ensures that the image does not exhibit significant rotation. Considering that during the movement of the C-arm, the machine only translates in two directions, with minimal rotation, appearing in the image between 3° and 5°, the rotation can be limited to θ. In the present invention, θ∈[3°,5°], with 3° being preferred in this embodiment.

[0155] B4. Determine whether the M pairs of feature points are suitable for the image matching relationship.

[0156] The judgment is specifically as follows:

[0157] Construct the energy equation:

[0158] (x j ,y j ) and (x j ′,y j ′) are the coordinates of any feature point pair between two feature points of M pairs, and H is the image matching relationship;

[0159] For this image matching relationship H = H1, substitute all the M pairs of feature points into the energy equation to calculate the energy. If the resulting energy value is less than a set threshold E, the feature point pair is considered an adaptive point pair and suitable for this image matching relationship H1. If it is greater than E, the feature point pair is considered a non-adaptive point pair and unsuitable for this image matching relationship H1. The number of adaptive point pairs (SUM1) is recorded. Furthermore, the threshold E is preferably 0.1.

[0160] B5. Repeat B3-B4 for a set number of times P, where the value of P is affected by M and N, P = f(M, N); randomly select M pairs of feature points that are different from the previous ones in each repetition, and obtain the image matching relationship Hp and the corresponding number of adaptive point pairs SUMp. In the present invention, 3≤M≤N-1, because when M=N, there is only one set of M pairs of feature points, and there are no different M pairs of feature points; when M=N-1, there will be N groups of M different pairs of feature points according to the permutations and combinations; when M=N-2, there will be N pairs of feature points according to the permutations and combinations. 2 -N groups of M different pairs of feature points; and so on. 3 ≤ M is the basic requirement for the restrictive conditions of the image matching relationship model to be solvable. Theoretically, the smaller M, the more accurate the comparison results. However, the smaller M, the more repetitions and the greater the computational effort. Therefore, in this embodiment, N / 2 ≤ M ≤ N-2 is preferred, as it is a trade-off between accuracy and computational effort.

[0161] B6. Based on the image matching relationships Hp obtained in B5 and the corresponding number of adaptive point pairs SUMp, the matching relationship in which the number of adaptive point pairs meets the set conditions is selected as the optimal image matching relationship.

[0162] In an exemplary embodiment, the number of feature point pairs that adapt to each image matching relationship is compared, and the optimal image matching relationship is identified based on the comparison result. Preferably, the image matching relationship with the largest number of feature point pairs is selected as the optimal image matching relationship.

[0163] In another exemplary embodiment, the number of adaptive point pairs SUMp each time is compared with the set amount K. When the number of adaptive point pairs is less than the set amount K, the repetition continues; when the number of adaptive point pairs is greater than the set amount K, the repetition stops. Specifically, the set amount can be selected as K=0.9N, and the current matching relationship is selected as the optimal image matching relationship.

[0164] Through the above steps, while calculating the feature point pair matching relationship, i.e., the image matching relationship, outlier pairs can be eliminated from the feature point pairs, reducing the interference of outliers on the image matching relationship. Even for a feature point pair set containing 80% noise, the optimal image matching relationship obtained is still accurate and valid.

[0165] The present invention also provides an image stitching method, as shown in FIG5 , comprising the steps of:

[0166] C1. The image acquisition device acquires images.

[0167] In an exemplary embodiment, the image acquisition device may be a mobile C-arm X-ray machine. The user turns on the laser pointer on the C-arm and adjusts the position of the C-arm X-ray machine so that the intersection of the two laser lines is located in the middle of the desired acquisition position and one of the laser lines substantially coincides with the human body's force line, completing the initial positioning.

[0168] The image acquisition device captures and stores the first image in the initial position and records its timing; then the image acquisition device moves in a directionally step-by-step manner, capturing and storing an image after each step, while recording its timing, until the translation of the image acquisition device covers the acquisition target. During this process, there is an overlapping area between two adjacent images; in this embodiment, taking the panoramic stitching of the human spine in the frontal position as an example, the acquisition target is the entire spine, and each image is an image of a different segment of the spine.

[0169] C2. Extract any two images with overlapping areas (hereinafter referred to as the first image and the second image) from the images captured by the image acquisition device to the canvas, and obtain the current positions of the two images;

[0170] In this embodiment, two adjacent images are preferred.

[0171] C3. Respectively identify the feature points in the two images extracted in step C2, and obtain the position and features of each feature point.

[0172] In this embodiment, the feature points extracted from the two images may be extracted using a SURF (Speeded Up Robust Features) algorithm or a SIFT algorithm.

[0173] C4. Match the feature points in the first image with the feature points in the second image based on the similarity of the features of each feature point, and obtain all successfully matched feature point pairs.

[0174] In this embodiment, all successfully matched feature point pairs can be obtained using the feature point pair extraction method described above.

[0175] C5. Obtain a matching relationship H between the first image and the second image based on all successfully matched feature point pairs obtained in step C4;

[0176] In this embodiment, the matching relationship between the first image and the second image may be acquired by using the aforementioned image matching method.

[0177] C6. Determine the target position A2' of the second image in the canvas based on the matching relationship H between the second image and the first image obtained in step C5 and the current position of the second image in the canvas. Specifically, A2'=A2*H, and move the second image A2 to the target position A2' to splice the second image onto the first image.

[0178] C7. Repeat steps C1 to C6 above, extracting two images different from the previous ones each time for processing, until all images are stitched together.

[0179] In the present invention, during the image stitching process, there are often more than two images. At this time, it is only necessary to calculate the positional relationship between the adjacent images and use chain operations to obtain the positional relationship between the current image and the first image, that is, the positional transformation relationship between the k+1th image and the first image is: T=H1*H2*...*Hk, Hk is the image matching relationship between the k+1th image and the kth image; then, the target positions of all images can be obtained to complete the stitching of all images.

[0180] In an exemplary embodiment, after step C7, the method further includes the following steps:

[0181] C8. The user fine-tunes any image in the images stitched together in step C7. The fine-tuning includes translation fine-tuning and / or rotation fine-tuning.

[0182] In an exemplary embodiment, the user fine-tunes any picture using the mouse. Specifically:

[0183] The user clicks the mouse on the image or the button related to the image on the display screen to make fine adjustments. The click has a corresponding translation and / or rotation step. Generally, the step direction is divided into forward and reverse. In terms of translation and rotation, it includes the forward translation step relationship F, the reverse translation step relationship F-1, the forward rotation step relationship T, and the reverse rotation step relationship T-1, each with a corresponding trigger mark. Based on this, by monitoring the movement of the mouse on the display screen, the image is rotated and translated for fine adjustments. Before and after each fine adjustment, the position transformation relationship of the finely adjusted image before and after adjustment is: H 微 =F n1 *(F -1 ) n2 *T m1 *(T -1 ) m2 , n1 is the cumulative number of positive translation steps of the monitored image, n2 is the cumulative number of negative translation steps of the monitored image, m1 is the cumulative number of positive rotation steps of the monitored image, and m2 is the cumulative number of negative rotation steps of the monitored image. The position of the image before and after fine-tuning satisfies p img =p screen *F n1 *(F -1 ) n2 *T m1 *(T -1 ) m2 , p img is the position after fine-tuning, p screen To fine-tune the front position.

[0184] In another exemplary embodiment, the user fine-tunes any image by clicking on the touch screen, and the corresponding fine-tuning is to rotate and translate the image by monitoring the movement of the cursor on the display screen.

[0185] The fine-tuning control method of the present invention is not limited to the above two methods, and any method that can form a trigger relationship with the step fine-tuning is acceptable.

[0186] Further optionally, after step C8, the method further includes:

[0187] C9. Based on the position of any fine-tuned image, the target positions of all images in the time sequence after the image are automatically updated, and all images after the image are automatically and adaptively fine-tuned.

[0188] This step avoids the need to manually adjust subsequent images after fine-tuning an intermediate image, thus improving the efficiency and simplicity of user fine-tuning.

[0189] In actual use, due to the different shooting positions, the quality and clarity of the images are different, which will affect the extraction and pairing of feature points, resulting in inaccurate image matching, and then leading to unsatisfactory stitching effects. In addition, because the image matching relationship is a chain calculation, the calculation error will be accumulated and amplified step by step, resulting in a large stitching error in the tail picture, and even "stitching fly". After the image stitching unit of the present invention adds a fine-tuning module and fine-tuning steps, it can fine-tune any of the spliced ​​images, including translation fine-tuning and rotation fine-tuning, to update the positional relationship between the two images, thereby greatly improving the operating experience and stitching accuracy, and effectively avoiding the occurrence of the "stitching fly" phenomenon.

[0190] In an exemplary embodiment, after obtaining the panoramic stitched image, the method further includes performing a fusion process on the overlapping areas of the stitched image, specifically:

[0191] 6 , for any point W in the overlapping area after stitching, the image fusion module recalculates its pixel. The pixel calculation formula is: pix=k1*pix1+k2*pix2

[0192] Among them, pix1 is the pixel value of point W in the first image P1, pix2 is the pixel value of point W in the second image P2, k1 is the first weight coefficient representing the proximity of point W to the overlapping boundary of the first image, and k2 is the second weight coefficient representing the proximity of point W to the overlapping boundary of the second image.

[0193] Furthermore, k1 = d2 / (d1+d2); k2 = d1 / (d1+d2)

[0194] Where d1 is the distance from point W to the overlapping boundary L2 of the second image. The overlapping boundary L1 of the first image P1 is determined based on the farthest point of the first image P1 in the stitching axis, that is, in the Y direction of the canvas coordinate system, and is a straight line passing through the farthest point and parallel to the X direction of the canvas coordinate system;

[0195] d2 is the distance from point W to the overlapping boundary L1 of the first image. The overlapping boundary L2 of the second image P2 is determined based on the closest point of the second image P2 along the stitching axis, that is, in the Y direction of the canvas coordinate system, and is a straight line passing through the closest point and parallel to the X direction of the canvas coordinate system;

[0196] As shown in Figure 6, the canvas coordinate system O is constructed with the image acquisition unit's directional translation direction as the y-axis and the closest point of the first image P1 in the image acquisition unit's directional translation direction as the origin. The interval between the first image overlap boundary L1 and the second image overlap boundary L2 constitutes the stitching interval. The position of point W within the stitching interval and its degree of deviation from the two boundaries of the stitching interval determine the first and second weight coefficients.

[0197] The image fusion module updates the pixels of the above points according to the calculation results.

[0198] This fusion process re-assigns the pixels in the overlapping area by combining the two factors of pixel and position, which can remove the overlapping images in the overlapping area, make the overlapping area transition naturally, and make the splicing visual effect clearer and softer.

[0199] C10. Output the automatically stitched image, print or display the current stitched image.

[0200] The present invention is not limited to anteroposterior panoramic stitching, but can also be lateral panoramic stitching or anteroposterior and lateral panoramic stitching.

[0201] The present invention extracts and matches feature points from two images to obtain feature point pairs, and then corrects the feature point pairs multiple times based on factors such as image acquisition shift and feature point pair distribution. This allows accurate extraction of feature point pairs from the two images, enabling rapid stitching of images with overlapping areas, overcoming the limited field of view of the image acquisition device. Furthermore, by repeatedly iteratively calculating the optimal matching relationship between the two images, the present invention eliminates outlier pairs from the feature point pairs, reducing the interference of outliers on the image matching relationship. This improves the tolerance for incorrect pairing of feature point pairs and reduces errors caused by uncertainties in positional relationship calculations, allowing for rapid and accurate automatic image stitching during the image stitching process.

[0202] Furthermore, existing registration methods suffer from poor stitching results when the number of images is large. This is primarily manifested in the increasingly irregular stitching of later images, with some even experiencing "stitching errors." This invention effectively avoids this problem by introducing a fine-tuning module and method. Furthermore, after image stitching, overlapping areas can experience unnatural and blurry transitions. This invention introduces a fusion module and method that integrates pixel and position factors to achieve a natural transition between overlapping areas, resulting in a clearer and smoother stitching effect.

[0203] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations (such as quantity, shape, position, etc.) can be made to the technical solution of the present invention, and these equivalent transformations all fall within the scope of protection of the present invention.

Claims

1. A method for extracting feature point pairs, characterized in that: Includes steps: Respectively identify feature points in the first image and the second image, and obtain the position and features of each feature point; Matching the feature points in the first image with the feature points in the second image according to the similarity of the features of each feature point, and marking the feature point pairs that have successfully matched; Denoising the successfully matched feature point pairs according to the motion range of the image acquisition device and / or the distribution of the feature point pairs; The feature point pairs after denoising are marked as extracted feature point pairs.

2. The feature point pair extraction method according to claim 1, characterized in that: According to the motion range of the image acquisition device, denoising the successfully matched feature point pair includes: According to the movement specification of the image acquisition device and the initial positioning of the two images, a constraint model is constructed to screen the feature point pairs. The constraint model is: |y j -y j ′|<t1*r+t2*r (1) |x j -x j ′|<t3*c+t4*c (2) Among them, (x j ,y j ) and (x j ′,y j ′) are the coordinates of the two feature points of the successfully matched feature point pair, r is the height of the image, c is the width of the image; t1 and t2 are the restriction coefficients related to the y-direction translation distance and translation stability of the image acquisition device, t3 and t4 are the coefficients related to the x-direction translation distance and translation stability of the image acquisition device; the x-direction and y-direction of the image acquisition device correspond to the x-direction and y-direction of the image, respectively.

3. The method for extracting feature point pairs according to claim 2, characterized in that: in, 0≤t1<1,0≤t3<1,0 <t2<0.5,0<t4<0.5。 4. The method for extracting feature point pairs according to claim 1, characterized in that: Denoising the successfully matched feature point pairs according to the distribution of the feature point pairs includes: Obtain the positional relationship between the two corresponding feature points in each successfully matched feature point pair; According to the positional relationship, each feature point is converted into a point to obtain a conversion point; Cluster the transformation points, identify and retain the feature point pairs corresponding to the category with the largest number of transformation points.

5. The method for extracting feature point pairs according to claim 4, characterized in that: The specific clustering of transformation points is as follows: 1) drawing a histogram according to the transformation points; 2) Randomly select num points from the transformation points in the histogram as the initial centers of each category; 3) According to the distance from each transformation point to the current center of each category, each transformation point is classified to the center with the smallest distance L, num initial clusters are obtained, and the initial clustering factor T of each category is obtained. 4) Update the current center of each category according to the average position of the transformation points in each category of the current num clusters, and obtain the current clustering factor T' corresponding to the current center of each category; 5) Calculate the difference ΔT between the initial clustering factor T and the current clustering factor T', and determine whether the corresponding difference ΔT of each category is less than the set threshold, otherwise return to step 2); if so, end clustering.

6. A method for obtaining image matching relationships, characterized in that: include: S1, extracting feature point pairs from two images with overlapping areas; S2, constructing a matching relationship model between two images; S3. Select different feature point pairs from the feature point pairs obtained in S1, calculate the image matching relationships corresponding to the different feature point pairs based on the selected feature point pairs and the matching relationship model obtained in S2, count the number of feature point pairs that are suitable for each image matching relationship, and identify the optimal image matching relationship based on the number.

7. The image matching relationship acquisition method according to claim 6, characterized in that: Identifying the optimal image matching relationship according to the quantity includes: Compare the numbers of feature point pairs that are suitable for the matching relationships of the respective images, and identify the optimal image matching relationship based on the comparison results; or, The number of feature point pairs that are suitable for each image matching relationship is compared with a threshold, and the optimal image matching relationship is identified based on the comparison result.

8. The image matching relationship acquisition method according to claim 7, characterized in that: The numbers of feature point pairs that are suitable for each image matching relationship are compared with each other, and the image matching relationship with the largest number of feature point pairs that are suitable for each image matching relationship is taken as the optimal image matching relationship.

9. The image matching relationship acquisition method according to claim 6, characterized in that: The matching relationship model between the two images is: Among them, r1, r2, r3, and r4 are model parameters related to rotation, and tx and ty are model parameters related to translation.

10. The image matching relationship acquisition method according to claim 9, characterized in that: The image matching relationships corresponding to the selected different feature point pairs are calculated separately, including: At least three pairs of feature point pairs from the feature point pairs obtained in S1 are randomly selected to form a group of feature point pairs, and restrictive conditions are constructed with the matching relationship model between the two images obtained in S2, and the model parameters are calculated to obtain the matching relationship between the two images.

11. The image matching relationship acquisition method according to claim 10, characterized in that: The restrictive conditions also include the following constraints: where θ∈[3°,5°].

12. The image matching relationship acquisition method according to claim 6, characterized in that: The number of feature point pairs of the statistical adaptation relationship of each image matching relationship includes: Construct the energy equation: Among them, (x j ,y j ) and (x j ′,y j ′) are the coordinates of any feature point other than the selected feature point, and H is the matching relationship between the two images; Substitute all feature point pairs except the selected ones into the energy equation to obtain energy, and record the number of feature point pairs whose energy values ​​are less than the set threshold, which is the number of point pairs that meet the matching relationship.

13. An image stitching method, characterized in that: include: (1) Extract any two images with overlapping areas to the canvas; (2) respectively identifying the feature points and features in the two images extracted in step (1); (3) Matching the feature points in the two images according to the similarity of the features of each feature point, obtaining all successfully matched feature point pairs, and extracting the feature point pairs; (4) obtaining a matching relationship between the two images based on the feature point pairs extracted in step (3), and splicing the two images based on the matching relationship; (5) Repeat steps (1) to (4), extracting two images different from the previous ones each time for processing, until all images are stitched together.

14. The image stitching method according to claim 13, characterized in that: The method further includes fine-tuning any one of the images stitched in step (5), wherein the fine-tuning includes translation and / or rotation step fine-tuning, and the stepping direction is divided into forward and reverse directions.

15. The image stitching method according to claim 14, characterized in that: The fine-tuning is: The position relationship of the image before and after fine-tuning is: img =p screen *F n1 *(F -1 ) n2 *T m1 *(T -1 ) m2 , p img is the position of the image after fine-tuning, p screen is the position of the image before fine-tuning; n1 is the cumulative number of positive translation steps of the monitored image, n2 is the cumulative number of negative translation steps of the monitored image, m1 is the cumulative number of positive rotation steps of the monitored image, and m2 is the cumulative number of negative rotation steps of the monitored image; F is the relationship between positive translation steps and F -1 is the reverse translation stepping relationship, T is the forward rotation stepping relationship, T -1 It is a reverse rotation step relationship.

16. The image stitching method according to claim 13 or 14, characterized in that: After obtaining the panoramic stitching image, the fusion step is also included: For any point in the overlapping area after stitching, its pixel is recalculated and updated. The pixel calculation formula is: pix=k1*pix1+k2*pix2; Among them, pix1 is the pixel value of the corresponding point in the first image, pix2 is the pixel value of the corresponding point in the second image, k1 is the first weight coefficient representing the proximity of the overlapping boundary between point W and the first image, and k2 is the second weight coefficient representing the proximity of the overlapping boundary between point W and the second image.

17. An image stitching unit, characterized in that: include: A feature extraction module is used to extract feature points on two images with overlapping areas, and transmit the extracted feature point information to a feature matching module; A feature matching module is used to match the feature points of the two images extracted by the feature extraction module, obtain feature point pairs, extract feature point pairs, and transmit the extracted feature point pair information to the image matching module; The image matching module calculates the matching relationship between the two images based on the feature point pairs, and accordingly shifts and splices the subsequent time-series image onto the previous time-series image to obtain a panoramic stitching image.

18. The image stitching unit according to claim 17, characterized in that: The feature extraction module extracts feature points on two images with overlapping areas by using a SURF algorithm or a SIFT algorithm.

19. The image stitching unit according to claim 18, characterized in that: The feature matching module matches the feature points of the two images extracted by the feature extraction module to obtain a pair of feature points: Calculate the similarity of the feature vectors between the feature points of one image and the feature points of the other image, and compare the calculation result with the similarity threshold. According to the comparison result, determine whether the two feature points can be paired, and obtain the feature point pairs that can be paired.

20. An image stitching system, characterized in that: include: An image acquisition unit, which acquires an image of a target in a directional translation state; The image stitching unit according to any one of claims 17 to 19, extracting the images acquired by the image acquisition unit, and automatically stitching the images to obtain a stitched image; An image display unit displays the stitched image.

21. The image stitching system according to claim 20, characterized in that: It also includes an image pre-processing unit, which pre-processes the image before the image is extracted to the image stitching unit. The pre-processing includes at least one of window position and window width adjustment, image rotation, image sorting, image selection, and image deletion.

22. The image stitching system according to claim 20, characterized in that: The device further comprises an indicating unit, wherein the indicating unit comprises a direction indicating unit indicating a directional translation direction of the image acquisition unit and / or a distance indicating unit indicating a translation distance of the image acquisition unit.

23. The image stitching system according to claim 22, characterized in that: The distance indicating unit comprises a height detector for detecting the height of a bed where the target is located and a prompter for calculating a critical distance of directional translation of the image acquisition unit according to the height of the bed.

24. The image stitching system according to claim 23, characterized in that: The prompter calculates the critical distance of the directional translation of the image acquisition unit according to the bed height detected by the height detector, and the calculation formula is: d = F[g(h)] Among them, d is the critical distance of the directional translation of the image acquisition unit, beyond which the two images will have no overlapping area; function g is the mapping relationship between the bed height h and the image range displayed on the imaging detector of the image acquisition unit, and F is the mapping relationship between the image displayed on the imaging detector and the actual imaging distance.

25. A computer storage medium, characterized in that The computer storage medium stores a computer program, which, when executed, implements the steps of the method according to any one of claims 13 to 16.

26. A C-arm machine, characterized in that: The C-arm machine comprises the system according to any one of claims 19 to 24.

Images