Image alignment method and device, storage medium and electronic device
By chunking the image to be measured and building a deep learning network model with different weight parameters, the homography matrix between images is directly estimated, which solves the problem of insufficient image alignment accuracy and adaptive learning ability in the prior art, and achieves the following High alignment accuracy and robustness.
Patent Information
- Application Number
- CN202310603741.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2024-12-03
AI Technical Summary
In the prior art, the image alignment method has problems with poor accuracy and adaptive learning ability, especially in image alignment in different shooting scenes. The traditional method relies on feature extraction and homography matrix estimation to be more complex and prone to deviations.
By chunking the image to be tested, a first deep learning network model with different weight parameters is constructed, a second deep learning network model is constructed using the response feature map, and the homography matrix between the images is directly estimated, thereby realizing image alignment. The feature extraction and error point removal steps in traditional methods are avoided.
Improve image alignment accuracy and adaptive learning ability, reduce the impact of wrong blocked areas on alignment, and achieve higher alignment accuracy and robustness.
Smart Images

Figure CN119068219A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the field of communications, and in particular, to an image alignment method and device, a storage medium, and an electronic device. Background Art
[0002] In the existing deep learning networks based on re-identification, the field of image alignment is almost not involved. Most of the re-identification technologies are mainly used for pedestrian or vehicle recognition. In images or videos of other non-overlapping shooting scenes, various changing features can be learned and the learned features can be mapped to a new space, making pedestrians or vehicles with the same features closer and pedestrians or vehicles with different features farther away. The same pedestrian or vehicle is recognized, but it lacks adaptive functions and cannot accurately align images. In addition, traditional homography matrix estimation methods rely on feature extraction and error point elimination, and the calculation is relatively complex. If there is a deviation, it will affect the accuracy of alignment. The homography matrix estimation methods based on deep learning networks rarely involve the field of image alignment.
[0003] Therefore, existing image alignment methods suffer from poor accuracy and adaptive learning capabilities. Summary of the invention
[0004] Embodiments of the present invention provide an image alignment method and device, a storage medium, and an electronic device to at least solve the problems of poor image alignment accuracy and poor adaptive learning ability in the related art.
[0005] According to one embodiment of the present invention, there is provided an image alignment method, comprising: performing block processing on an image to be tested to obtain a plurality of training blocks, and constructing a first deep learning network model with different weight parameters for each of the training blocks; inputting a target image into the first deep learning network model to obtain a response feature map corresponding to the training block; constructing a second deep learning network model according to the response feature map to obtain a homography matrix representing a mapping relationship between the image to be tested and the target image; and performing image alignment on the image to be tested and the target image based on the homography matrix.
[0006] According to another embodiment of the present invention, there is provided an image alignment device, comprising: a block training module, used to perform block processing on the image to be tested, obtain multiple training blocks, and construct a first deep learning network model with different weight parameters for each training block; a target response module, used to input the target image into the first deep learning network model, and obtain a response feature map corresponding to the training block; a target mapping module, used to construct a second deep learning network model according to the response feature map, and obtain a homography matrix representing the mapping relationship between the image to be tested and the target image; and an image alignment module, used to perform image alignment on the image to be tested and the target image based on the homography matrix.
[0007] According to yet another embodiment of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.
[0008] According to yet another embodiment of the present invention, there is provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
[0009] Through the present invention, by processing the image to be tested in blocks, a plurality of training blocks are obtained, and a first deep learning network model with different weight parameters is constructed for each training block; the target image is input into the first deep learning network model to obtain a response feature map corresponding to the training block; a second deep learning network model is constructed according to the response feature map to obtain a homography matrix representing the mapping relationship between the image to be tested and the target image; and the image to be tested and the target image are aligned based on the homography matrix. The problem of poor image alignment accuracy and adaptive learning ability in the related art is solved, and the effect of improving the image alignment accuracy and adaptive learning ability is achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1 It is a hardware structure block diagram of a computer terminal of an image alignment method according to an embodiment of the present invention;
[0011] Figure 2 is a flow chart of an image alignment method according to an embodiment of the present invention;
[0012] Figure 3 is a structural block diagram of an image alignment device according to an embodiment of the present invention;
[0013] Figure 4 is a structural block diagram of a block training module according to an embodiment of the present invention;
[0014] Figure 5is a flow chart of an image alignment method according to a scenario embodiment of the present invention;
[0015] Figure 6 It is a schematic diagram of the UNet network structure according to a scenario embodiment of the present invention;
[0016] Figure 7 It is a schematic diagram of the deep learning network training alignment principle of the image alignment method according to the scenario embodiment of the present invention. DETAILED DESCRIPTION
[0017] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings and in combination with the embodiments.
[0018] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0019] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a computer terminal as an example, Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal of an image alignment method according to an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the above-mentioned computer terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.
[0020] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the image alignment method in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0021] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0022] In this embodiment, an image alignment method running on the above-mentioned computer terminal is provided. Figure 2 is a flowchart of an image alignment method according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:
[0023] Step S202, dividing the image to be tested into blocks to obtain multiple training blocks, and constructing a first deep learning network model with different weight parameters for each training block;
[0024] In an exemplary embodiment, a first deep learning network model with different weight parameters is constructed for each training block, including: performing data augmentation operations on N training blocks respectively to obtain N feature data sets corresponding to the training blocks, where N is a positive integer; performing deep learning network training according to the N feature data sets respectively to obtain N first deep learning network models with different weight parameters.
[0025] In the actual implementation process, the above-mentioned data augmentation operation may include flipping, cropping, padding, etc. After the above-mentioned data augmentation operation, N feature data sets corresponding to the training blocks are obtained, that is, each training block corresponds to a feature data set. According to the feature data set of each training block, the first deep learning network model corresponding to the data set is constructed, and each training block corresponds to a first deep learning network model. The type of the first deep learning network model is the same, but the weight parameters of the first deep learning network models corresponding to each training block are different from each other. The weight parameters of each first deep learning network model are related to the feature data set of the training block corresponding to it. Through the above-mentioned technical means, it is possible to achieve adaptive adjustment of the weight parameters to learn the features of the image to be tested to effectively and accurately improve the accuracy of image alignment, more targeted judgment of the accuracy of alignment, and reduce the influence of erroneous block areas on image alignment.
[0026] Step S204, inputting the target image into the first deep learning network model to obtain a response feature map corresponding to the training block;
[0027] In an exemplary embodiment, a target image is input into a first deep learning network model to obtain a response feature map corresponding to a training block, including: inputting the target image into corresponding first deep learning network models of N first deep learning network models, respectively, to obtain N response feature maps corresponding to the training blocks, respectively.
[0028] In the actual implementation process, the target images are respectively input into the first deep learning network model to obtain response feature maps corresponding to the training blocks, and the specific number of response feature maps is equal to the number of training blocks.
[0029] Step S206, constructing a second deep learning network model according to the response feature graph to obtain a homography matrix representing a mapping relationship between the image to be tested and the target image;
[0030] In the actual implementation process, the multiple response feature maps obtained in the above steps are input into the deep learning network for training and learning to obtain the second deep learning network model. The type of the second deep learning network model can be the same as or different from the type of the first deep learning network model. The homography matrix that characterizes the mapping relationship between the image to be tested and the target image can be estimated through the second deep learning network model, avoiding the shortcomings of the traditional method that relies on feature extraction, error point removal and other steps that are complex to calculate and prone to deviations. The deep learning network can avoid these steps and directly estimate the homography matrix through the network, which can adaptively achieve a good alignment effect with high accuracy.
[0031] Step S208: aligning the image to be tested and the target image based on the homography matrix.
[0032] In an exemplary embodiment, the type of the first deep learning network model or the second deep learning network model includes at least one of the following: a convolutional neural network; a residual network; a UNet network.
[0033] Through the above steps, the image to be tested is processed into blocks to obtain multiple training blocks, and a first deep learning network model with different weight parameters is constructed for each training block; the target image is input into the first deep learning network model to obtain a response feature map corresponding to the training block; a second deep learning network model is constructed according to the response feature map to obtain a homography matrix representing the mapping relationship between the image to be tested and the target image; and the image to be tested and the target image are aligned based on the homography matrix. The problem of poor image alignment accuracy and adaptive learning ability in the related art is solved, and the effect of improving the accuracy and adaptive learning ability of image alignment is achieved.
[0034] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.
[0035] In this embodiment, an image alignment device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.
[0036] Figure 3 is a structural block diagram of an image alignment device according to an embodiment of the present invention. Figure 3As shown, the image alignment device 30 includes: a block training module 310, which is used to perform block processing on the image to be tested, obtain multiple training blocks, and construct a first deep learning network model with different weight parameters for each training block; a target response module 320, which is used to input the target image into the first deep learning network model, and obtain a response feature map corresponding to the training block; a target mapping module 330, which is used to construct a second deep learning network model according to the response feature map, and obtain a homography matrix that characterizes the mapping relationship between the image to be tested and the target image; an image alignment module 340, which is used to perform image alignment on the image to be tested and the target image based on the homography matrix.
[0037] In an exemplary embodiment, the type of the first deep learning network model or the second deep learning network model includes at least one of the following: a convolutional neural network; a residual network; a UNet network.
[0038] In an exemplary embodiment, the target response module 320 is further used to input the target image into the corresponding first deep learning network models among the N first deep learning network models, and obtain N response feature maps corresponding to the training blocks respectively.
[0039] Figure 4 is a structural block diagram of a block training module according to an embodiment of the present invention. Figure 4 As shown, the block training module 310 includes: a data augmentation unit 410, used to perform data augmentation operations on N training blocks respectively, to obtain N feature data sets corresponding to the training blocks, where N is a positive integer; a network training unit 420, used to perform deep learning network training according to the N feature data sets respectively, to obtain N first deep learning network models with different weight parameters.
[0040] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.
[0041] An embodiment of the present invention further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.
[0042] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.
[0043] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.
[0044] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.
[0045] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.
[0046] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
[0047] In order to enable those skilled in the art to better understand the technical solution of the present invention, it is described below in conjunction with specific scenario embodiments.
[0048] Scenario Example 1
[0049] Among the existing image alignment methods, one is to use the optical flow method for image alignment, obtain the optical flow information map of the image to be tested and the target image, and divide them into superpixels, and calculate the homography matrix of each area, calculate the mapping map of each superpixel area, and finally fuse them to obtain the aligned image. However, the two basic assumptions of the optical flow method are constant brightness and small movement, which are not easy to meet in reality. The second is a method based on deep learning, which selects sub-images or key point sets at the same position of the image to be tested and the target image, and then uses a deep learning network to extract the feature points of the two images and match them, and calculates the homography matrix to complete the alignment operation. Although it can effectively improve the accuracy of image alignment, since the images may be obtained from different devices, different camera resolutions, and different shooting angles, the obtained feature information is always unclear and cannot achieve accurate alignment. The re-identification technology can accurately obtain feature information in scenes such as cross-device and different shooting angles, but the re-identification technology lacks adaptive functions. Therefore, combining the re-identification technology with a deep learning network with adaptive weights can solve these problems well. It can adaptively identify features that change under different brightness and different scenes, and accurately obtain image feature information to achieve accurate image alignment.
[0050] Figure 5 is a flow chart of an image alignment method according to a scenario embodiment of the present invention, such as Figure 5 As shown, the following steps are included:
[0051] Step 1: Obtain the image A to be tested and divide it into N blocks.
[0052] Wherein, N is a positive integer, and the value of N is not specifically limited, wherein the size of the block includes but is not limited to the size of a pixel block, and the size of half the input image;
[0053] Step 2: Perform data augmentation operations (flipping, cropping, filling, etc.) on the first area of the image to be tested to obtain a set of data, and use the data set to train a deep learning network 1 to obtain a deep learning network 1 with new weights for the first area.
[0054] Among them, the types of deep learning network 1 include but are not limited to convolutional neural network, residual network, UNet network structure, etc. Figure 6 is a schematic diagram of the UNet network structure according to a scenario embodiment of the present invention, such as Figure 6 As shown in the figure, the image to be tested is input into the UNet network structure, and the specific network structure characteristics can be obtained through the connection sampling of multiple convolution blocks.
[0055] Step 3, obtaining a target image B, inputting the deep learning network 1 with new weights of the first area obtained in step 2, and obtaining a response feature map of the first area of the image to be tested on B;
[0056] Step 4: Repeat steps 2 and 3 for the remaining blocks of the image to be tested until all the block areas of the image to be tested are trained, and a set of response feature maps is obtained, the number of which is N. The deep learning network structure of each prediction remains unchanged, and the weight parameters change according to the characteristics of each block area of the input image. There is no specific limit on the number of parameters;
[0057] Step 5: Using the set of response feature maps obtained in step 4 as a feature data set, a deep learning network 2 is trained using the feature data set to predict a homography matrix H representing the mapping relationship between the image to be tested and the target image, and an alignment operation is performed according to the homography matrix to obtain an aligned image. The deep learning network 2 for predicting the homography matrix H includes but is not limited to a convolutional neural network, a residual network, and Figure 6 The UNet network structure shown, etc.
[0058] The main innovation of the present invention is to propose a new image alignment method based on re-identification, divide the image to be tested into blocks and use each area separately for data augmentation as a data set, predict a deep learning network based on re-identification, let the network with new weights better extract the response feature map of the target image, make the same features closer, and different features farther, and at the same time use all the response feature maps as data sets, estimate the homography matrix through a deep learning network, and then perform alignment operations. The specific structure and network training parameters of the deep learning network for re-identification and deep learning network for estimating the homography matrix involved in the method are not specifically limited, including but not limited to convolutional neural network, residual network, UNet network structure, etc.
[0059] Figure 7 is a schematic diagram of deep learning network training alignment according to an image alignment method of a scenario embodiment of the present invention, such as Figure 7 As shown, the image to be tested is first divided into blocks, and then data augmentation is performed on each area as a data set. The obtained data set is used to train the deep learning network 1 to obtain the weight corresponding to each area, and then the deep learning network 1 with different weights is obtained. The target image B is respectively input into the deep learning network 1 with different weights obtained above to obtain a set of response feature maps corresponding to each area of the image to be tested A on the target image B. Afterwards, the obtained set of response feature maps is used to train the deep learning network to obtain the deep learning network 2. The homography matrix H can be estimated through the deep learning network 2. The image to be tested A and the Mu Bai image B are aligned based on the homography matrix H to obtain the aligned image.
[0060] In summary, the present invention provides an image alignment method, which applies the re-identification technology used to identify pedestrians or vehicles to image alignment. The re-identification technology can adapt to scenes with low camera resolution and biased shooting angles, learn various changing features of pedestrians or vehicles under different cameras, and map the learned features to a new space. Therefore, the re-identification idea is transplanted to the field of image alignment, which solves the problem of image feature extraction in different time periods across devices, and improves the accuracy and robustness of feature extraction. Use a network with adaptively adjusted weight parameters to extract features. For the image to be tested, data augmentation is performed on the same deep learning network to predict the same deep learning network separately, and deep learning networks with different new weights are obtained respectively. The weight parameters change according to the characteristics of each area of the input image, and then the target image is input into these networks with new weights to obtain the response feature map of the first area of the image to be tested on the target image. The adaptive adjustment of the weight parameters can learn the characteristics of the image to be tested to effectively and accurately improve the accuracy of image alignment, more targeted judgment of the accuracy of alignment, and reduce the influence of the wrong block area on image alignment. Based on a set of response feature maps obtained as a data set, a deep learning network is used to estimate the homography matrix, avoiding the shortcomings of traditional methods that rely on feature extraction, error point removal and other steps that are complex to calculate and prone to deviations. The deep learning network can avoid these steps and directly estimate the homography matrix through the network, which can adaptively achieve a good alignment effect with high accuracy.
[0061] In the above implementation process, re-identification technology is used to accurately obtain feature information in scenarios such as cross-device and different shooting angles. It can include data augmentation operations (flipping, cropping, padding, etc.) and other conventional technical means of re-identification technology. Image features, that is, the above-mentioned feature image set, can be obtained according to actual conditions.
[0062] Compared with the existing image alignment methods, the present invention mainly proposes a method for realizing image alignment by using a deep learning network based on re-identification technology, wherein the re-identification technology can adapt to scenes such as low camera resolution, biased shooting angle, inconsistent brightness, and large motion range, and adopts an adaptive network to extract the features of the target image, wherein the adaptive network can adjust the weight parameters according to the input feature features, thereby improving the robustness and accuracy of feature extraction, more targetedly judging the accuracy of alignment, and reducing the influence of erroneous pixel areas on image alignment. At the same time, a deep learning network is used to estimate the homography matrix, avoiding the shortcomings of traditional methods that rely on feature extraction, error point elimination, etc., which are complex to calculate and prone to deviations. The deep learning network can avoid these steps and directly estimate the homography matrix through the network, which can adaptively achieve a good alignment effect, and the alignment effect has high accuracy.
[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. An image alignment method, characterized in that: include: Perform block processing on the image to be tested to obtain multiple training blocks, and construct a first deep learning network model with different weight parameters for each training block; Inputting the target image into the first deep learning network model to obtain a response feature map corresponding to the training block; Constructing a second deep learning network model according to the response feature graph to obtain a homography matrix representing a mapping relationship between the image to be tested and the target image; The image to be tested and the target image are aligned based on the homography matrix.
2. The method according to claim 1, characterized in that in, The type of the first deep learning network model or the second deep learning network model includes at least one of the following: Convolutional Neural Networks; Residual Networks; UNet network.
3. The method according to claim 1, characterized in that The constructing of a first deep learning network model with different weight parameters for each training block comprises: Performing data augmentation operations on the N training blocks respectively to obtain N feature data sets corresponding to the training blocks, where N is a positive integer; Deep learning network training is performed respectively according to the N feature data sets to obtain N first deep learning network models with different weight parameters.
4. The method according to claim 3, characterized in that The step of inputting the target image into the first deep learning network model to obtain a response feature map corresponding to the training block includes: The target images are respectively input into the corresponding first deep learning network models among the N first deep learning network models to obtain N response feature maps respectively corresponding to the training blocks.
5. An image alignment device, characterized in that: include: A block training module is used to perform block processing on the image to be tested to obtain multiple training blocks, and to construct a first deep learning network model with different weight parameters for each training block; A target response module, used for inputting a target image into the first deep learning network model to obtain a response feature map corresponding to the training block; A target mapping module, used to construct a second deep learning network model according to the response feature map, and obtain a homography matrix representing a mapping relationship between the image to be tested and the target image; An image alignment module is used to align the image to be tested and the target image based on the homography matrix.
6. The device according to claim 5, characterized in that in, The type of the first deep learning network model or the second deep learning network model includes at least one of the following: Convolutional Neural Networks; Residual Networks; UNet network.
7. The device according to claim 5, characterized in that The block training module includes: A data augmentation unit, used to perform data augmentation operations on the N training blocks respectively to obtain N feature data sets corresponding to the training blocks, where N is a positive integer; A network training unit is used to perform deep learning network training according to the N feature data sets respectively to obtain N first deep learning network models with different weight parameters.
8. The device according to claim 5, characterized in that The target response module is also used to: The target images are respectively input into the corresponding first deep learning network models among the N first deep learning network models to obtain N response feature maps respectively corresponding to the training blocks.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the method described in any one of claims 1 to 4 when executed by a processor.
10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method described in any one of claims 1 to 4 is implemented.
Citation Information
Patent Citations
Image processing method and device, training method of re-identification network and electronic equipment
CN112668374A
Image alignment method and device, computer readable storage medium and electronic equipment
CN113962846A
Vehicle face image alignment method based on feature matching
CN115588110A
Image registration method
US20180150962A1
