Image registration method and device based on spatial composite descriptor

Abstract

The invention discloses an image registration method and device based on a spatial composite descriptor. The method comprises the steps of obtaining a matching result between a first image and a second image; according to the similarity index of the image, screening from the plurality of matching point pairs to obtain a first anchor point pair corresponding to the maximum similarity index; determining an anchor point corresponding to each target image based on the first anchor point pair; obtaining a first composite feature point of each target image based on the feature point of each target image and the corresponding anchor point; based on the first composite feature point of each target image, determining a spatial composite descriptor of the feature point of each target image; matching the feature points of each target image according to the spatial composite descriptors of the feature points to obtain calculation point pairs; and determining a corresponding transformation matrix according to the calculation point pairs. According to the method, the feature point matching accuracy of the first image and the second image can be improved, so that the image registration accuracy and registration efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular to an image registration method and apparatus based on spatial composite descriptors. Background Technology

[0002] In the field of medical image diagnosis, fundus images obtained through scanning are often used to assist in eye diagnosis. In actual diagnostic procedures, after obtaining multiple fundus images, it is necessary to register these images to facilitate subsequent image fusion and obtain a suitable medical image for diagnosis. Existing image registration methods mainly include feature-based methods, which match images by extracting feature points (or feature regions) from the fundus images. However, the blood vessels in fundus images are complex and have many similarities; using individual feature point pairs for matching can easily lead to matching errors. Summary of the Invention

[0003] This application provides an image registration method and apparatus based on spatial composite descriptors, with the aim of improving the accuracy of fundus image registration.

[0004] To achieve the above objectives, this application provides the following technical solution:

[0005] An image registration method based on spatial composite descriptors includes:

[0006] Obtain a matching result between a first image and a second image; the first image and the second image are both fundus images of the same retina; the matching result includes multiple matching point pairs; each matching point pair includes two feature points that have a matching relationship; one of the two feature points belongs to the first image and the other belongs to the second image;

[0007] Using the similarity transformation matrix corresponding to each matching point pair, the similarity index of the image is calculated, and the matching point pair corresponding to the maximum similarity index is selected and used as the first anchor point pair.

[0008] Based on the first anchor point pair, anchor points corresponding to each target image are determined; each target image includes the first image and the third image; the third image is obtained by transforming the second image.

[0009] Based on the feature points of each target image and the corresponding anchor points, the first composite feature points of each target image are obtained.

[0010] Based on the first composite feature points of each of the target images, a spatial composite descriptor of the feature points of each of the target images is determined;

[0011] Based on the spatial composite descriptor of the feature points, the feature points of each target image are matched to obtain computed point pairs;

[0012] Based on the calculated point pairs, a corresponding first transformation matrix is ​​determined; the first transformation matrix is ​​used to transform the second image to obtain a registered second image.

[0013] An image registration device based on spatial composite descriptors includes:

[0014] A matching unit is used to obtain a matching result between a first image and a second image; the first image and the second image are both fundus images of the same retina; the matching result includes multiple matching point pairs; each matching point pair includes two feature points that have a matching relationship; one of the two feature points belongs to the first image and the other belongs to the second image;

[0015] Anchor point pair filtering unit is used to calculate the similarity index of the image using the similarity transformation matrix corresponding to each matching point pair, and to filter out the matching point pair corresponding to the maximum similarity index, and use the matching point pair as the first anchor point pair.

[0016] An anchor point determination unit is used to determine anchor points corresponding to each target image based on the first anchor point pair; each target image includes the first image and a third image; the third image is obtained by transforming the second image;

[0017] A composite feature point determination unit is used to obtain the first composite feature point of each target image based on the feature points of each target image and the corresponding anchor points;

[0018] A spatial composite descriptor determining unit is used to determine the spatial composite descriptor of the feature points of each of the target images based on the first composite feature points of each of the target images;

[0019] The feature point matching unit is used to match the feature points of each target image according to the spatial composite descriptor of the feature points to obtain computed point pairs;

[0020] The matrix determination unit is used to determine the corresponding first transformation matrix based on the calculated point pair; the first transformation matrix is ​​used to transform the second image to obtain the registered second image.

[0021] The technical solution provided in this application obtains the matching result between a first image and a second image. Based on the image similarity index, a first anchor point pair corresponding to the maximum similarity index is selected from multiple matching point pairs. Based on the first anchor point pair, anchor points corresponding to each target image are determined. Based on the feature points of each target image and their corresponding anchor points, first composite feature points of each target image are obtained. Based on the first composite feature points of each target image, a spatial composite descriptor of the feature points of each target image is determined. Based on the spatial composite descriptor of the feature points, the feature points of each target image are matched to obtain calculated point pairs. Based on the calculated point pairs, the corresponding transformation matrix is ​​determined. This application, based on the anchor points and feature points of each target image, determines the spatial composite descriptor of the feature points, and based on the spatial composite descriptor of the feature points, matches the feature points of each target image, improving the feature point matching accuracy of the first image and the second image, thereby improving the accuracy and efficiency of image registration. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating an image registration method based on spatial composite descriptors provided in this application embodiment;

[0024] Figure 2 A schematic diagram of the architecture of an image registration device based on spatial composite descriptors provided in this application embodiment;

[0025] Figure 3 A schematic diagram of the first composite feature point provided in the embodiments of this application;

[0026] Figure 4 A schematic diagram of the second composite feature point provided in the embodiments of this application;

[0027] Figure 5 This is a schematic diagram of feature point matching provided in an embodiment of this application;

[0028] Figure 6 This is a schematic diagram of image registration provided in an embodiment of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0031] like Figure 1 The diagram shown is a flowchart of an image registration method based on spatial composite descriptors provided in an embodiment of this application, including the following steps.

[0032] S101: Obtain the matching result between the first image and the second image.

[0033] In this embodiment, both the first and second images are fundus images of the same retina. The matching result includes multiple matching point pairs. Each matching point pair consists of two feature points that have a matching relationship. One of the two feature points belongs to the first image, and the other belongs to the second image. In this embodiment, the first image is used as the reference image, and the second image is used as the image to be transformed.

[0034] Optionally, the process of obtaining the matching result between the first image and the second image can be found in steps A1-A5 below.

[0035] Step A1: Obtain the first image and the second image.

[0036] Step A2: Extract feature points from the first image to obtain the feature points of the first image.

[0037] In some examples, the number of feature points in the first image can be multiple.

[0038] Step A3: Extract feature points from the second image to obtain the feature points of the second image.

[0039] In some examples, the number of feature points in the second image can be multiple.

[0040] It should be noted that the feature points of the first and second images contain information including scale and angle.

[0041] Step A4: Calculate the descriptors of the feature points in the first image and the descriptors of the feature points in the second image.

[0042] In some examples, the descriptor can be a binary descriptor, specifically a brief descriptor. The process of calculating the brief descriptor of feature points is a conventional technique and will not be elaborated here.

[0043] Step A5: Based on the descriptors of the feature points of the first image and the second image respectively, match the feature points of the first image and the second image respectively to obtain multiple matching point pairs.

[0044] It should be noted that the matching method used is a two-dimensional feature point matching method, which can extract two feature descriptor sets (i.e., descriptors of multiple feature points) from the first image and the second image. For each descriptor in the first feature descriptor set, the descriptor that matches it is found in the second feature descriptor set to obtain the corresponding matching point pair.

[0045] It should be emphasized that this embodiment uses a brute-force matching method based on Hamming distance to achieve feature point matching.

[0046] S102: Calculate the similarity index of the image using the similarity transformation matrix corresponding to each matching point pair, and select the matching point pair corresponding to the maximum similarity index, and use the matching point pair as the first anchor point pair.

[0047] Optionally, the process of calculating the image similarity index using the similarity transformation matrix corresponding to each matching point pair can be found in steps B1-B4 below.

[0048] Step B1: For each pair of matching points, calculate the transformation scale and transformation angle of the similarity transformation.

[0049] In some examples, the transformation scale is the ratio between a first scale and a second scale, where the first scale represents the scale of the feature points in the first image of the matching point pair, and the second scale represents the scale of the feature points in the second image of the matching point pair.

[0050] In a possible implementation, the transformation scale = first scale / second scale.

[0051] In some examples, the transformation angle is the difference between a first angle and a second angle, where the first angle represents the angle of the feature points in the first image of the matching point pair, and the second angle represents the angle of the feature points in the second image of the matching point pair.

[0052] In a possible implementation, the transformation angle = first angle - second scale.

[0053] Step B2: Based on the transformation scale and transformation angle of the similarity transformation, and combined with the coordinates of the two feature points in the matching point pair, calculate the similarity transformation matrix corresponding to the matching point pair.

[0054] In some examples, the similarity transformation matrix calculated using the transformation scale, transformation angle, and the coordinates of the two feature points can be seen in Equation (1).

[0055]

[0056] In formula (1), Scale represents the transformation scale and Angle represents the transformation angle. The calculation process of dx and dy can be found in formula (2).

[0057]

[0058] In formula (2), baseP represents the coordinates of the feature points of the first image, planP represents the coordinates of the feature points of the second image, h[0,0] represents the value at coordinate (0,0) in the similarity transformation matrix h, h[0,1] represents the value at coordinate (0,1) in the similarity transformation matrix h, h[1,0] represents the value at coordinate (1,0) in the similarity transformation matrix h, and h[1,1] represents the value at coordinate (1,1) in the similarity transformation matrix h.

[0059] Step B3: Transform the second image using a similarity transformation matrix to obtain the corresponding transformed image.

[0060] It is important to note that the image size of the transformed image is the same as that of the reference image.

[0061] Step B4: Calculate the similarity between the transformed image and the reference image.

[0062] In some examples, a similarity algorithm (such as the NCC algorithm) can be used to calculate the similarity between the transformed image and the reference image, obtain the similarity transformation matrix and the matching point pair corresponding to the maximum similarity index, and use the matching point pair as the first anchor point pair.

[0063] S103: Based on the first anchor point pair, determine the anchor points corresponding to each target image.

[0064] Each target image includes a first image and a third image, with the third image obtained by transforming the second image.

[0065] It should be noted that after determining the first anchor point pair, the anchor points corresponding to the first image and the third image can be determined based on the first anchor point pair. In a possible implementation, the first anchor point pair includes point A and point B, where point A belongs to the feature point of the first image and point B belongs to the feature point of the second image. Thus, point A can be determined as the anchor point corresponding to the first image, and point B as the anchor point corresponding to the second image. Then, point B is transformed to obtain the corresponding anchor point C in the third image.

[0066] It should be emphasized that the first transformation matrix M1 can be used to transform the second image to obtain the third image. The purpose of the transformation is to further align the first and second images, making the subsequent calculation of descriptors more accurate and improving the matching success rate.

[0067] In some examples, the first transformation matrix M1 can be the similarity transformation matrix corresponding to the first anchor point pair shown in steps B1-B5, that is, the similarity transformation matrix corresponding to the maximum similarity index.

[0068] S104: Based on the feature points of each target image and the corresponding anchor points, obtain the first composite feature points of each target image.

[0069] In this process, feature points can be extracted from each target image to obtain the feature points of each target image, and all feature points of each target image can be combined with the corresponding anchor points to obtain multiple first composite feature points of each target image.

[0070] See in some examples Figure 3 As shown, the feature points of the target image are A2 and A3, and the corresponding anchor point is A1. The first composite feature point obtained by combining them can be (A1, A2) and (A1, A3).

[0071] S105: Based on the first composite feature points of each target image, determine the spatial composite descriptor of the feature points of each target image.

[0072] The process of determining the spatial composite descriptor of the feature points of each target image based on the first composite feature points of each target image can be found in steps C1-C5 below.

[0073] Step C1: For each first composite feature point, determine the target vector and the corresponding vector angle based on the feature points and anchor points shown in the first composite feature point.

[0074] In some examples, the line segment pointing from the anchor point to the feature point is considered the target vector.

[0075] Step C2: Update the angles of the feature points to vector angles, and update the angles of the anchor points to target angles.

[0076] In some examples, the target angle includes, but is not limited to, the negative of the vector angle, the sum of the vector angle and 90°, or the difference between 180° and the vector angle.

[0077] Step C3: Update the scale of the anchor point to the scale of the feature point, or update the scale of the feature point to the scale of the anchor point.

[0078] Step C4: Calculate the updated descriptors for the feature points and the updated descriptors for the anchor points.

[0079] Step C5: Generate a spatial composite descriptor for the feature points based on the updated descriptors of the feature points and anchor points.

[0080] In some examples, the descriptors for both the updated feature points and anchor points are binary strings. The spatial composite descriptor of a feature point is equal to the binary string of the updated feature point plus the binary string of the updated anchor point. Specifically, assuming the binary string has 512 characters, the spatial composite descriptor has 1024 characters.

[0081] S106: Match the feature points of each target image based on the spatial composite descriptor of the feature points to obtain the calculated point pairs.

[0082] Specifically, by using a spatial composite descriptor of feature points to match feature points in the first and third images, and then transforming the matched points in the third image into matching points in the second image to obtain calculated point pairs, the probability of incorrect matching at similar locations in the second and first images can be reduced, thus improving the accuracy of feature point matching. Furthermore, transforming the matched points in the third image into matching points in the second image can be achieved by performing an inverse transformation of the matched points in the third image using the first transformation matrix M1, or by calculating the transformation matrix between the matched point pairs obtained from the first and third images and then multiplying it by the first transformation matrix M1.

[0083] Optionally, the process of matching feature points of each target image according to the spatial composite descriptor of feature points to obtain the calculation point pair can be referred to the steps shown in the following embodiments S201-206.

[0084] S107: Determine the corresponding transformation matrix based on the calculated point pairs.

[0085] The transformation matrix is ​​used to transform the second image to obtain the registered second image.

[0086] In some examples, the transformation matrix types include homography transformation matrices and binary quadratic transformation matrices. The calculated point pairs include the target matching point pairs, valid matching point pairs, and available matching point pairs for each image region. Therefore, the transformation matrix can be determined based on the mapping situation of each image region. Specifically, if the feature point mapping situation exceeds a set condition (e.g., the number of feature points in more than half of the image regions exceeds a set threshold), the corresponding binary quadratic transformation matrix is ​​determined to obtain the best detail alignment effect; otherwise, the corresponding homography transformation matrix is ​​determined.

[0087] In some examples, a transformation matrix is ​​used to transform the second image, and the resulting registered second image can be found in [reference needed]. Figure 6 As shown in the figure, the matching results are highly accurate.

[0088] The process described in S101-S107 above determines the spatial composite descriptor of the feature points based on the anchor points and feature points of each target image. Based on the spatial composite descriptor of the feature points, the feature points of each target image are matched, which improves the feature point matching accuracy of the first image and the second image, thereby improving the accuracy and efficiency of image registration.

[0089] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0090] S201: Obtain multiple first points and multiple second points.

[0091] Among them, the first point and the second point belong to feature points of different target images.

[0092] S202: For each first point, determine the corresponding restricted area based on the coordinates of the first point.

[0093] The defined region includes, in the target image to which the second point belongs, a circular region centered on the coordinates of the first point with a radius equal to a preset calibration value, or a rectangular region with a length and width equal to a preset calibration value, etc.

[0094] In a possible implementation, taking point A in the first image as an example, in the third image, a circular area with the coordinates of point A as the center and a radius equal to a preset calibration value is defined as the limited area corresponding to point A.

[0095] S203: Based on the defined region, determine at least one valid second point from a plurality of second points.

[0096] Among them, the coordinates of the valid second point are located within the defined area.

[0097] It is understandable that after determining the restricted area corresponding to the first point, the second point whose coordinates are located within the restricted area is determined as the valid second point corresponding to the first point.

[0098] S204: Determine the matching value between the first point and the valid second point based on their respective spatial composite descriptors.

[0099] In this process, after determining at least one valid second point corresponding to the first point, a matching algorithm is used to match each valid second point with the first point to obtain the matching value between each valid second point and the first point.

[0100] S205: Based on the second valid point with the largest matching value, determine the matching point as the first point.

[0101] In this process, after determining the matching value corresponding to each valid second point, the valid second point with the largest matching value is selected and determined as the matching point of the first point.

[0102] S206: Based on each first point and its corresponding matching point, determine the calculation point pair.

[0103] Optionally, the process of determining the calculation point pair based on each first point and its corresponding matching point can be found in the steps shown in S301-S305 of the following embodiments.

[0104] It should be noted that when determining the point pair based on the first point and the matching point, since the matching point belongs to the feature point of the third image, it is necessary to use the first transformation matrix M1 to perform an inverse transformation on the matching point, and finally determine the point pair based on the first point and the inversely transformed matching point.

[0105] The processes shown in S201-S206 above can use the coordinates of each first point to determine at least one valid second point corresponding to each first point, and use the spatial composite descriptors of the first point and the valid second point to determine the matching point corresponding to each first point from at least one valid second point.

[0106] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0107] S301: According to the first preset rule, the matching point of each first point is corrected so that the matching points of each first point are different from each other.

[0108] The first preset rule is as follows: for multiple first points with the same matching point, the matching point of the first point with the largest matching value is determined to remain unchanged, and other valid second points with the largest matching value outside the same matching point within the limited area corresponding to other first points are determined as the matching points of other first points.

[0109] In some examples, the first preset rule can also be understood as: traversing the matching points of each first point in sequence, if the best match corresponding to two first points points to the same matching point, then comparing their matching values, and changing the matching point of the first point with the lower matching value to: the second largest valid second point among all valid second points that match the first point with the lower matching value (subsequently the third largest matching value, the fourth largest matching value, and so on).

[0110] S302: According to the second preset rule, eliminate part of the first point, and determine the remaining first point and the corresponding matching point as the target matching point pair.

[0111] The second preset rule is: if the distance between any two first points meets the preset threshold, the first point with the smallest matching value among any two first points will be eliminated.

[0112] In some examples, the second preset rule can also be understood as: counting whether there are other first points whose distance range is within a set threshold for each first point; if so, comparing the matching values ​​of all first points within the range, retaining the first point with the largest matching value, and deleting the other first points and their corresponding matching points.

[0113] S303: Divide the first image into blocks according to the image size of the first image to obtain multiple image regions.

[0114] The first image can be divided into 8*8 image regions.

[0115] S304: Determine the target matching point pairs that match each image region based on the matching values ​​corresponding to the target matching point pairs.

[0116] Among them, the matching value corresponding to the target matching point pair is the matching value between the first point and the second point shown in the target matching point pair.

[0117] Optionally, the process of determining the target matching point pair that matches each image region based on the matching value corresponding to the target matching point pair can be found in the steps S401-S405 of the following embodiments.

[0118] S305: Determine the calculation point pairs based on the target matching point pairs matched for each image region.

[0119] Optionally, the process of determining the calculation point pair based on the target matching point pair matched by each image region can be found in the steps shown in S501-S502 of the following embodiments.

[0120] The processes shown in S301-S305 above utilize the first and second preset rules to determine target matching point pairs. Based on the matching values ​​of the target matching point pairs, and combined with multiple image regions, the target matching point pairs matched by each image region are determined, thereby improving the accuracy of the final calculated point pairs and effectively improving the matching accuracy of feature points.

[0121] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0122] S401: Determine the matching values ​​and reference points corresponding to multiple target matching point pairs.

[0123] The reference point is the feature point belonging to the first image.

[0124] S402: For each image region, map the coordinates of the reference point whose matching value meets the first threshold to the image region.

[0125] S403: Determine the total number of reference points mapped to the image region.

[0126] S404: If the total number of reference points meets the specified number, determine the target matching point pair that matches the image region based on the target matching point pair to which the reference points are mapped to the image region.

[0127] S405: If the total number of reference points does not meet the specified number, the coordinates of the reference points whose matching values ​​meet the second threshold are mapped to the image region so that the total number of mapped reference points meets the specified number, and the target matching point pairs to which the reference points currently mapped to the image region belong are determined as the target matching point pairs that match the image region.

[0128] The second threshold is less than the first threshold.

[0129] In some examples, to make the distribution of matching point pairs more uniform and the results more reliable, the process shown in S401-S405 can be understood as follows: First, obtain and retain all matching point pairs with a matching value greater than or equal to 0.8, and map them to the corresponding image regions according to the coordinates of the reference points; count the number of reference points in each image region; then lower the matching value threshold to 0.65 to obtain all matching point pairs with a matching value greater than or equal to 0.65; if the number of reference points in a certain image region is greater than or equal to 5, then the image region is no longer mapped; otherwise, the matching point pairs are retained and mapped according to the coordinates of the reference points until the number of reference points in the image region is greater than or equal to 5.

[0130] The processes shown in S401-S405 above utilize each image region, combined with the matching value corresponding to the target matching point pair and the reference point, to determine the target matching point pair matched by each image region, thereby improving the matching accuracy of feature points.

[0131] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0132] S501: Using the random sampling consensus algorithm, the target matching point pairs matched in each image region are filtered to obtain multiple second anchor point pairs.

[0133] Among them, the random sampling consensus algorithm can be the ransac algorithm. The calculation process of the ransac algorithm is a conventional technique and will not be described in detail here.

[0134] S502: Determine the calculation point pair based on the first anchor point pair and multiple second anchor point pairs.

[0135] Optionally, the process of determining the calculation point pair based on the first anchor point pair and multiple second anchor point pairs can be found in the steps shown in S601-S606 of the following embodiments.

[0136] It should be noted that the second anchor point pair includes the anchor points of the third image. Therefore, the first transformation matrix M1 is also needed to perform an inverse transformation on the anchor points of the third image in the second anchor point pair. Finally, the second anchor point pair after the inverse transformation is used to determine the calculation point pair.

[0137] The process shown in S501-S502 above can improve the accuracy of the calculated point pairs by using the first anchor point pair and multiple second anchor point pairs.

[0138] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0139] S601: Based on the first anchor point pair and multiple second anchor point pairs, determine the first anchor point set corresponding to the first image and the second image respectively.

[0140] The first set of anchor points includes multiple anchor points.

[0141] It is understandable that multiple anchor points corresponding to the first image and the second image can be determined by parsing the first anchor point pair and multiple second anchor point pairs.

[0142] S602: For any one of the first image and the second image, according to the third preset rule, eliminate some feature points of the image to obtain the effective feature points of the image.

[0143] The third preset rule is: for any feature point in any image, if the distance between any feature point and any anchor point in the first anchor point set meets the preset calibration distance, then any feature point will be eliminated.

[0144] In some examples, the third preset rule can be understood as: in any image, feature points that appear within a threshold range of a certain anchor point are removed and do not participate in subsequent matching.

[0145] S603: From the first set of anchor points, select anchor points according to the filtering rules and determine them as valid anchor points corresponding to valid feature points.

[0146] The selection rules are not limited. They can be as follows: after determining the distance between each anchor point in the first anchor point set and the valid feature point, anchor points that satisfy the rules of minimum, second minimum, and maximum distance are selected from the first anchor point set and identified as valid anchor points corresponding to valid feature points. Furthermore, multiple rules can be used simultaneously, meaning one feature point can correspond to multiple anchor points, thus providing richer spatial information.

[0147] S604: Based on the effective feature points of any image and the corresponding effective anchor points, obtain the second composite feature points of any image.

[0148] Specifically, the effective feature points of any image are combined with the corresponding effective anchor points to obtain the second composite feature points of any image.

[0149] See in some examples Figure 4 As shown, if the effective feature point of any image is A5 and the nearest anchor point to it is A4, then the second composite feature point obtained by combining them can be (A4, A5); if the effective feature point of any image is A7 and the nearest anchor point to it is A6, then the second composite feature point obtained by combining them can be (A6, A7).

[0150] S605: Based on the second composite feature point of any image, determine the spatial composite descriptor of the effective feature points of any image.

[0151] The implementation process and execution principle of determining the spatial composite descriptor of the corresponding effective feature point based on the second composite feature point can be found in steps C1-C5, and will not be repeated here.

[0152] S606: Match the effective feature points of the first image and the second image based on the spatial composite descriptor of the effective feature points to obtain effective matching point pairs.

[0153] Furthermore, the image can be divided into blocks to obtain effective matching point pairs that match each image region. The implementation process and execution principle can be found in the steps S301-S305 and S401-S405 of the above embodiments.

[0154] S607: Determine the calculation point pairs based on valid matching point pairs, as well as the first anchor point pair and multiple second anchor point pairs.

[0155] Optionally, the process of determining the calculation point pair based on the valid matching point pairs, the first anchor point pair, and multiple second anchor point pairs can be referred to the steps shown in S701-S708 of the following embodiments.

[0156] It should be noted that when determining the calculation point pairs based on the valid matching point pairs, since the valid matching point pairs include the valid feature points of the third image, it is also necessary to use the first transformation matrix M1 to perform an inverse transformation on the valid feature points of the third image, and finally determine the calculation point pairs based on the valid matching point pairs after the inverse transformation.

[0157] The process described in S601-S607 above determines the second composite feature points of the first image and the second image based on the first anchor point pair and multiple second anchor point pairs. Using the second composite feature points, the spatial composite descriptors of the effective feature points of the first image and the second image are determined. Based on the spatial composite descriptors of the effective feature points, the effective feature points of the first image and the second image are matched to obtain calculated point pairs, thereby improving the matching accuracy of feature points.

[0158] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0159] S701: Based on valid matching point pairs, obtain multiple third anchor point pairs.

[0160] Among them, the random sampling consensus algorithm can be used to filter valid matching point pairs to obtain multiple third anchor point pairs.

[0161] S702: Based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, determine multiple anchor points corresponding to each candidate image.

[0162] Each candidate image includes a first image and a fourth image. The fourth image is obtained by transforming the second image, further aligning the first and second images. For anchor points belonging to the third image in the second and third anchor point alignments, an inverse transformation is performed using the first transformation matrix M1 to obtain the corresponding anchor points in the second image.

[0163] Optionally, the process of determining multiple anchor points corresponding to the fourth image based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs can be found in the steps shown in S801-S804 of the following embodiments.

[0164] S703: Based on multiple anchor points corresponding to each candidate image, determine the candidate feature points of each candidate image.

[0165] Specifically, for each candidate image, the process of determining candidate feature points based on multiple anchor points corresponding to the candidate image includes: obtaining a mask image corresponding to the candidate image; specifically, filling the area defined by the anchor points; then, not extracting feature points from the filled area in the mask image, and extracting feature points from the unfilled area to obtain candidate feature points.

[0166] In some examples, the candidate anchor points include points A and B. The defined region corresponding to point A is a first circular region centered on the coordinates of point A and with a radius equal to a preset calibration value. The defined region corresponding to point B is a second circular region centered on the coordinates of point B and with a radius equal to a preset calibration value. Therefore, feature points are extracted from the regions outside the first and second circular regions and determined as candidate feature points of the candidate image.

[0167] S704: From multiple anchor points corresponding to the candidate image, select anchor points according to the filtering rules and determine them as candidate anchor points corresponding to the candidate feature points.

[0168] The filtering rules used are not limited. They can be determined by identifying the distances between multiple anchor points corresponding to candidate images and candidate feature points, and then filtering out anchor points that meet the rules of minimum, second minimum, maximum, etc., and determining them as candidate anchor points corresponding to candidate feature points. Multiple rules can be used simultaneously.

[0169] S705: Based on the candidate feature points and corresponding candidate anchor points of each candidate image, obtain the third composite feature points of each candidate image.

[0170] In this process, the candidate feature points of each candidate image are combined with the corresponding candidate anchor points to obtain the third composite feature points of each candidate image.

[0171] In some examples, the effective feature point of the candidate image is A8, and the corresponding candidate anchor point is A7. Then the third composite feature point obtained by combining them can be (A7, A8).

[0172] S706: Based on the third composite feature points of each candidate image, determine the spatial composite descriptor of the candidate feature points of each candidate image.

[0173] The implementation process and execution principle of determining the spatial composite descriptor of the corresponding candidate feature point based on the third composite feature point can be found in steps C1-C5, and will not be repeated here.

[0174] S707: Based on the spatial composite descriptor of the candidate feature points of each candidate image, the candidate feature points of each candidate image are matched to obtain candidate matching point pairs.

[0175] Furthermore, the image can be divided into blocks to obtain candidate matching point pairs that match each image region. The implementation process and execution principle can be found in the steps shown in S301-S305 and S401-S405 of the above embodiments, and will not be repeated here.

[0176] S708: Based on candidate matching point pairs, as well as first anchor point pairs, multiple second anchor point pairs, and multiple third anchor point pairs, determine the calculation point pairs.

[0177] Optionally, the process of determining the calculation point pair based on the candidate matching point pairs matched for each image region, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, can be found in the steps shown in S901-S906 of the following embodiments.

[0178] It is important to note that when determining the calculation point pairs based on candidate matching point pairs, since the candidate matching point pairs include candidate feature points from the fourth image, it is necessary to use the second transformation matrix M2 to perform an inverse transformation on the candidate feature points of the fourth image. Finally, the calculation point pairs are determined based on the candidate matching point pairs after the inverse transformation. Similarly, since the second and third anchor point pairs include anchor points from the third image, it is necessary to use the first transformation matrix M1 to perform an inverse transformation on the anchor points of the third image in the second and third anchor point pairs. Finally, the calculation point pairs are determined based on the anchor point pairs after the inverse transformation.

[0179] The process described in S701-S708 above determines the third composite feature point of each candidate image based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs. Using the third composite feature point, the spatial composite descriptor of the candidate feature point of each candidate image is determined. Based on the spatial composite descriptor of the candidate feature point, the candidate feature point of each candidate image is matched to obtain the calculated point pair, thereby improving the matching accuracy of the feature point.

[0180] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0181] S801: Generate a first set of point pairs based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs.

[0182] S802: Calculate the second transformation matrix based on the first set of point pairs.

[0183] The calculation of the second transformation matrix M2 based on the first set of point pairs is common knowledge and will not be elaborated here.

[0184] S803: Use the second transformation matrix to transform the second image to obtain the fourth image.

[0185] S804: Using the second transformation matrix, transform the multiple anchor points corresponding to the second image shown in the first point pair set to obtain the multiple anchor points corresponding to the fourth image.

[0186] The processes shown in S801-S804 above can determine multiple anchor points corresponding to each candidate image based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs.

[0187] Optionally, another image registration method based on spatial composite descriptors provided in the embodiments of this application includes the following steps.

[0188] S901: Based on the candidate matching point pairs, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, generate a second point pair set.

[0189] Among them, the candidate matching point pairs, the second anchor point pairs, and the third anchor point pairs need to be inversely transformed.

[0190] S902: The third transformation matrix is ​​calculated based on the second set of point pairs.

[0191] The calculation of the third transformation matrix M3 based on the second set of point pairs is common knowledge and will not be elaborated here.

[0192] S903: Base candidate matching point pairs, which identify key feature points in the fourth image and the first image that have a matching relationship.

[0193] The key feature points of the fourth image include the candidate feature points of the fourth image shown in the candidate matching point pair matched by each image region, and the key feature points of the first image include the candidate feature points of the first image shown in the candidate matching point pair matched by each image region.

[0194] S904: Using the third transformation matrix, transform the key feature points of the fourth image to obtain the corresponding usable feature points.

[0195] S905: If the distance between the available feature points of the fourth image and the key feature points in the first image that have a matching relationship meets the preset spacing, the available feature points of the fourth image and the key feature points in the first image that have a matching relationship are determined as available matching point pairs.

[0196] Specifically, the distance between the available feature points in the fourth image and the key feature points in the first image that have a matching relationship with the available feature points can be determined based on the coordinates of the available feature points in the fourth image and the coordinates of the key feature points in the first image that have a matching relationship with the available feature points. Furthermore, the image can be divided into blocks to obtain available matching point pairs that match each image region. The implementation process and execution principle can be found in steps S301-S305 and S401-S405 of the above embodiments, and will not be repeated here.

[0197] S906: Determine the calculation point pairs based on the available matching point pairs, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs.

[0198] The available matching point pairs, along with the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, can be determined as calculation point pairs. In a possible implementation, the feature points of the second image shown in the final obtained calculation point pairs can be found in [reference needed]. Figure 5 As shown.

[0199] It should be noted that when determining the calculation point pairs based on the available matching point pairs, since the available matching point pairs include the available feature points of the fourth image, it is also necessary to use the third transformation matrix M3 to perform an inverse transformation on the available feature points of the fourth image, and finally determine the calculation point pairs based on the available matching point pairs after the inverse transformation.

[0200] The process described in S901-S906 above, based on candidate matching point pairs and combined with the third transformation matrix, obtains available matching point pairs, and incorporates the available matching point pairs into the calculated point pairs, thereby effectively improving the accuracy of the calculated point pairs.

[0201] like Figure 2 The diagram shown is a schematic of the architecture of an image registration device based on spatial composite descriptors provided in an embodiment of this application, including the following units.

[0202] The matching unit 100 is used to obtain the matching result between the first image and the second image; the first image and the second image are both fundus images of the same retina; the matching result includes multiple matching point pairs; the matching point pair includes two feature points that have a matching relationship; one of the two feature points belongs to the first image and the other belongs to the second image.

[0203] Anchor point pair filtering unit 200 is used to calculate the similarity index of the image using the similarity transformation matrix corresponding to each matching point pair, and to filter out the matching point pair corresponding to the maximum similarity index, and use the matching point pair as the first anchor point pair.

[0204] Anchor point determination unit 300 is used to determine the anchor points corresponding to each target image based on the first anchor point pair; each target image includes a first image and a third image; the third image is obtained by transforming the second image.

[0205] The composite feature point determination unit 400 is used to obtain the first composite feature point of each target image based on the feature points of each target image and the corresponding anchor points.

[0206] The spatial composite descriptor determination unit 500 is used to determine the spatial composite descriptor of the feature points of each target image based on the first composite feature points of each target image.

[0207] The feature point matching unit 600 is used to match feature points of each target image according to the spatial composite descriptor of the feature points to obtain calculated point pairs.

[0208] The matrix determination unit 700 is used to determine the corresponding transformation matrix based on the calculated point pairs; the transformation matrix is ​​used to transform the second image to obtain the registered second image.

[0209] Optionally, the feature point matching unit 600 is specifically used for: obtaining multiple first points and multiple second points; the first points and second points belong to feature points of different target images respectively; for each first point, determining a corresponding defined region based on the coordinates of the first point; the defined region includes a region within a preset range centered on the coordinates of the first point in the target image to which the second point belongs; determining at least one valid second point from the multiple second points according to the defined region; the coordinates of the valid second point are located within the defined region; determining a matching value between the first point and the valid second point based on their respective spatial composite descriptors; determining the valid second point with the largest matching value as the matching point of the first point; and determining a calculation point pair based on each first point and its corresponding matching point.

[0210] Optionally, the feature point matching unit 600 is specifically used for: correcting the matching points of each first point according to a first preset rule, so that the matching points of each first point are different from each other; the first preset rule is: for multiple first points of the same matching point, the matching point of the first point with the largest matching value is determined to remain unchanged, and other valid second points with the largest matching value in the limited area corresponding to other first points (excluding the same matching point) are determined as the matching points of other first points; according to a second preset rule, some first points are eliminated, and the remaining first points and their corresponding matching points are determined as target matching point pairs; the second preset rule is: if the distance between any two first points meets a preset threshold, the first point with the smallest matching value among any two first points is eliminated; the first image is divided into blocks according to the image size of the first image to obtain multiple image regions; the target matching point pairs matched with each image region are determined according to the matching value corresponding to the target matching point pairs; and calculation point pairs are determined based on the target matching point pairs matched with each image region.

[0211] Optionally, the feature point matching unit 600 is specifically used to: determine the matching values ​​and reference points corresponding to multiple target matching point pairs; the reference points are feature points belonging to the first image; for each image region, map the coordinates of the reference points whose matching values ​​meet a first threshold to the image region; determine the total number of reference points mapped to the image region; if the total number of reference points meets a specified number, determine the target matching point pairs that match the image region based on the target matching point pairs to which the reference points mapped to the image region belong; if the total number of reference points does not meet a specified number, map the coordinates of the reference points whose matching values ​​meet a second threshold to the image region, so that the total number of mapped reference points meets a specified number, and determine the target matching point pairs that match the image region based on the target matching point pairs to which the reference points currently mapped to the image region belong; the second threshold is less than the first threshold.

[0212] Optionally, the feature point matching unit 600 is specifically used to: use a random sampling consensus algorithm to filter the target matching point pairs matched in each image region to obtain multiple second anchor point pairs; and determine the calculation point pairs based on the first anchor point pairs and the multiple second anchor point pairs.

[0213] Optionally, the feature point matching unit 600 is specifically used for: determining a first anchor point set corresponding to the first image and the second image respectively based on a first anchor point pair and multiple second anchor point pairs; the first anchor point set includes multiple anchor points; for any image in the first image and the second image, eliminating some feature points of any image according to a third preset rule to obtain effective feature points of any image; the third preset rule is: for any feature point in any image, if the distance between any feature point and any anchor point in the first anchor point set meets a preset calibration distance, the feature point is eliminated; from the first anchor point set, anchor points are selected according to a filtering rule and determined as effective anchor points corresponding to effective feature points; based on the effective feature points of any image and the corresponding effective anchor points, obtaining a second composite feature point of any image; based on the second composite feature points of any image, determining a spatial composite descriptor of the effective feature points of any image; matching the effective feature points of the first image and the second image according to the spatial composite descriptor of the effective feature points to obtain effective matching point pairs; and determining calculation point pairs based on the effective matching point pairs, the first anchor point pair, and multiple second anchor point pairs.

[0214] Optionally, the feature point matching unit 600 is specifically used for: obtaining multiple third anchor point pairs based on valid matching point pairs; determining multiple anchor points corresponding to each candidate image based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs; each candidate image includes a first image and a fourth image; the fourth image is obtained by transforming the second image; determining candidate feature points of each candidate image based on the multiple anchor points corresponding to each candidate image; selecting anchor points from the multiple anchor points corresponding to the candidate image according to a filtering rule, and determining them as candidate anchor points corresponding to the candidate feature points; obtaining third composite feature points of each candidate image based on the candidate feature points and the corresponding candidate anchor points; determining spatial composite descriptors of the candidate feature points of each candidate image based on the third composite feature points of each candidate image; matching the candidate feature points of each candidate image based on the spatial composite descriptors of the candidate feature points of each candidate image to obtain candidate matching point pairs; and determining calculation point pairs based on the candidate matching point pairs, the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs.

[0215] Optionally, the feature point matching unit 600 is specifically used to: generate a first point pair set based on a first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs; calculate a second transformation matrix based on the first point pair set; transform the second image using the second transformation matrix to obtain a fourth image; and transform multiple anchor points corresponding to the second image shown in the first point pair set using the second transformation matrix to obtain multiple anchor points corresponding to the fourth image.

[0216] Optionally, the feature point matching unit 600 is specifically used for: generating a second point pair set based on candidate matching point pairs, a first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs; calculating a third transformation matrix based on the second point pair set; determining key feature points in the fourth image and the first image that have a matching relationship based on the candidate matching point pairs; transforming the key feature points in the fourth image using the third transformation matrix to obtain corresponding usable feature points; if the distance between the usable feature points in the fourth image and the key feature points in the first image that have a matching relationship conforms to a preset spacing, determining the usable feature points in the fourth image and the key feature points in the first image that have a matching relationship as usable matching point pairs; and determining calculation point pairs based on the usable matching point pairs, the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs.

[0217] Each of the units described above determines a spatial composite descriptor of the feature points based on the anchor points and feature points of each image. Based on the spatial composite descriptor of the feature points, the feature points of each image are matched, which improves the feature point matching accuracy of the first image and the second image, thereby improving the accuracy and efficiency of image registration.

[0218] This application also provides a computer-readable storage medium including a stored program, wherein the program executes the image registration method based on spatial composite descriptors provided in this application.

[0219] This application also provides an electronic device, including a processor, a memory, and a bus. The processor and the memory are connected via the bus. The memory is used to store a program, and the processor is used to run the program. During program execution, the image registration method based on spatial composite descriptors provided in this application is executed.

[0220] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0221] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for image registration based on spatial compound descriptors, characterized in that, include: Obtain the matching result between the first image and the second image; The first image and the second image are both fundus images of the same retina; the matching result includes multiple matching point pairs; each matching point pair includes two feature points that have a matching relationship; one of the two feature points belongs to the first image and the other belongs to the second image; Using the similarity transformation matrix corresponding to each matching point pair, the similarity index of the image is calculated, and the matching point pair corresponding to the maximum similarity index is selected and used as the first anchor point pair. Based on the first anchor point pair, anchor points corresponding to each target image are determined; each target image includes the first image and the third image; the third image is obtained by transforming the second image. Based on the feature points of each target image and the corresponding anchor points, the first composite feature points of each target image are obtained. Based on the first composite feature points of each of the target images, a spatial composite descriptor of the feature points of each of the target images is determined; Based on the spatial composite descriptor of the feature points, the feature points of each target image are matched to obtain computed point pairs; Based on the calculated point pairs, a corresponding first transformation matrix is ​​determined; the first transformation matrix is ​​used to transform the second image to obtain a registered second image.

2. The method of claim 1, wherein, Based on the spatial composite descriptor of the feature points, feature points of each target image are matched to obtain computed point pairs, including: Multiple first points and multiple second points are obtained; the first points and the second points are respectively attributed to feature points of different target images; For each of the first points, a corresponding defined region is determined based on the coordinates of the first point; the defined region includes a region within a preset range centered on the coordinates of the first point in the target image to which the second point belongs; Based on the defined region, at least one valid second point is determined from a plurality of second points; the coordinates of the valid second point are located within the defined region. Based on the spatial composite descriptors of the first point and the valid second point, determine the matching value between the first point and the valid second point; The second valid point with the largest matching value is determined as the matching point of the first point. Based on each of the first points and its corresponding matching points, a calculation point pair is determined.

3. The method of claim 2, wherein, Based on each of the first points and its corresponding matching points, a pair of calculation points is determined, including: According to the first preset rule, the matching point of each first point is corrected so that the matching points of each first point are different from each other; the first preset rule is: for multiple first points of the same matching point, the matching point of the first point with the largest matching value is determined to remain unchanged, and other valid second points with the largest matching value in the limited area corresponding to other first points (excluding the same matching point) are determined as the matching points of the other first points; According to the second preset rule, some of the first points are eliminated, and the remaining first points and their corresponding matching points are determined as target matching point pairs; the second preset rule is: if the distance between any two first points meets a preset threshold, the first point with the smallest matching value among the two first points is eliminated; Based on the image size of the first image, the first image is divided into blocks to obtain multiple image regions; Based on the matching value corresponding to the target matching point pair, determine the target matching point pair that matches each of the image regions; Based on the target matching point pairs matched for each of the image regions, the calculation point pairs are determined.

4. The method of claim 3, wherein, Based on the matching values ​​corresponding to the target matching point pairs, determine the target matching point pairs that match each of the image regions, including: Determine the matching values ​​and reference points corresponding to multiple target matching point pairs; the reference points are feature points belonging to the first image. For each image region, the coordinates of the reference point whose matching value meets the first threshold are mapped to the image region; Determine the total number of reference points mapped to the image region; If the total number of reference points meets the specified number, the target matching point pairs to which the reference points mapped to the image region belong are determined as the target matching point pairs that match the image region; If the total number of reference points does not meet the specified number, the coordinates of reference points whose matching values ​​meet the second threshold are mapped to the image region, so that the total number of mapped reference points meets the specified number, and the target matching point pairs to which the reference points currently mapped to the image region belong are determined as target matching point pairs that match the image region; the second threshold is less than the first threshold.

5. The method of claim 3, wherein, Based on the target matching point pairs matched for each of the image regions, the calculation point pairs are determined, including: Using a random sampling consensus algorithm, the target matching point pairs matched in each image region are filtered to obtain multiple second anchor point pairs; Based on the first anchor point pair and multiple second anchor point pairs, a calculation point pair is determined.

6. The method of claim 5, wherein, Based on the first anchor point pair and multiple second anchor point pairs, calculation point pairs are determined, including: Based on the first anchor point pair and multiple second anchor point pairs, a first anchor point set corresponding to each of the first image and the second image is determined; the first anchor point set includes multiple anchor points. For any one of the first image and the second image, according to a third preset rule, some feature points of the image are eliminated to obtain valid feature points of the image; the third preset rule is: for any feature point in the image, if the distance between the feature point and any anchor point in the first anchor point set meets the preset calibration distance, the feature point is eliminated; From the first set of anchor points, anchor points are selected according to the filtering rules and determined as valid anchor points corresponding to the valid feature points; Based on the effective feature points and corresponding effective anchor points of any image, a second composite feature point of any image is obtained. Based on the second composite feature points of any image, determine the spatial composite descriptor of the effective feature points of any image; Based on the spatial composite descriptor of the effective feature points, the effective feature points of the first image and the second image are matched to obtain effective matching point pairs; Based on the effective matching point pairs, the first anchor point pair, and multiple second anchor point pairs, calculation point pairs are determined.

7. The method of claim 6, wherein, Based on the effective matching point pairs, the first anchor point pair, and multiple second anchor point pairs, the calculation point pairs are determined, including: Based on the effective matching point pairs, multiple third anchor point pairs are obtained; Based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, multiple anchor points corresponding to each candidate image are determined; each candidate image includes the first image and the fourth image; the fourth image is obtained by transforming the second image; Based on multiple anchor points corresponding to each candidate image, candidate feature points of each candidate image are determined. From the multiple anchor points corresponding to the candidate image, anchor points are selected according to the filtering rules and determined as candidate anchor points corresponding to the candidate feature points; Based on the candidate feature points and corresponding candidate anchor points of each candidate image, the third composite feature points of each candidate image are obtained. Based on the third composite feature points of each candidate image, a spatial composite descriptor of the candidate feature points of each candidate image is determined. Based on the spatial composite descriptor of the candidate feature points of each candidate image, the candidate feature points of each candidate image are matched to obtain candidate matching point pairs. Based on the candidate matching point pairs, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, calculation point pairs are determined.

8. The method of claim 7, wherein, The process of determining multiple anchor points corresponding to the fourth image based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs includes: Based on the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, a first point pair set is generated; Based on the first set of point pairs, the second transformation matrix is ​​calculated; The second image is transformed using the second transformation matrix to obtain the fourth image; Using the second transformation matrix, multiple anchor points corresponding to the second image shown in the first point pair set are transformed to obtain multiple anchor points corresponding to the fourth image.

9. The method of claim 7, wherein, Based on the candidate matching point pairs, and the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, a calculation point pair is determined, including: Based on the candidate matching point pairs, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, a second point pair set is generated; Based on the second set of point pairs, the third transformation matrix is ​​calculated; Based on the candidate matching point pairs, key feature points in the fourth image and the first image that have a matching relationship are determined; The key feature points of the fourth image are transformed using the third transformation matrix to obtain the corresponding usable feature points; If the distance between the available feature points of the fourth image and the key feature points in the first image that have a matching relationship meets the preset spacing, the available feature points of the fourth image and the key feature points in the first image that have a matching relationship are determined to be a pair of available matching points; Based on the available matching point pairs, as well as the first anchor point pair, multiple second anchor point pairs, and multiple third anchor point pairs, calculation point pairs are determined.

10. An image registration apparatus based on spatially compounded descriptors, characterized by include: A matching unit is used to obtain the matching result between the first image and the second image; The first image and the second image are both fundus images of the same retina; the matching result includes multiple matching point pairs; each matching point pair includes two feature points that have a matching relationship; one of the two feature points belongs to the first image and the other belongs to the second image; Anchor point pair filtering unit is used to calculate the similarity index of the image using the similarity transformation matrix corresponding to each matching point pair, and to filter out the matching point pair corresponding to the maximum similarity index, and use the matching point pair as the first anchor point pair. An anchor point determination unit is used to determine anchor points corresponding to each target image based on the first anchor point pair; each target image includes the first image and a third image; the third image is obtained by transforming the second image; A composite feature point determination unit is used to obtain the first composite feature point of each target image based on the feature points of each target image and the corresponding anchor points; A spatial composite descriptor determining unit is used to determine the spatial composite descriptor of the feature points of each of the target images based on the first composite feature points of each of the target images; The feature point matching unit is used to match the feature points of each target image according to the spatial composite descriptor of the feature points to obtain computed point pairs; A matrix determination unit is used to determine the corresponding first transformation matrix based on the calculated point pair; The first transformation matrix is ​​used to transform the second image to obtain the registered second image.

Images