A multi-source heterogeneous image data registration and fusion method

Abstract

The application discloses a multi-source heterogeneous image data registration and fusion method, and relates to the technical field of image processing. The method comprises the following steps: detecting feature points of a reference image and a to-be-registered image by using a FAST feature detector, and screening by using an aggregated feature strategy AF to obtain a feature point set; calculating a local phase response by using the odd-even symmetry characteristic of a two-dimensional Log-Gabor filter; generating a weighted phase direction feature by using a maximum and minimum bandwidth weighting function; generating a descriptor vector by constructing a regularized logarithmic polar coordinate neighborhood grid; performing feature matching according to the descriptor vector, solving a geometric transformation model, and performing clipping to obtain a registration image; obtaining a significant structure matrix by using significant structure information extracted by local Wiener filtering; performing iterative joint filtering, and obtaining a convergence result after iteration; and fusing the convergence result and the registration image to obtain a fusion image. The stable matching between different modal images is ensured.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for registration and fusion of multi-source heterogeneous image data. Background Technology

[0002] With the rapid development of multi-source sensing and imaging technologies, image data from different sensors, platforms, or imaging modes are becoming increasingly abundant. These data exhibit significant differences in spectral response, spatial resolution, imaging angle, and noise characteristics, representing typical multi-source heterogeneous image data. Multi-source heterogeneous image fusion, as a crucial component of information fusion technology, aims to comprehensively utilize the complementary information from various source images, integrating useful features from different sensors into a single fused image to obtain a more complete and accurate scene description. The results can provide more reliable visual information support for subsequent tasks such as target detection, recognition, monitoring, and analysis. In practical applications, such as remote sensing mapping, environmental monitoring, medical imaging, industrial inspection, and autonomous driving, the differences between different imaging sources mean that a single modal image often cannot fully reflect the characteristics of the target or scene. Through the registration and fusion of multi-source heterogeneous images, the advantageous features of each source image can be preserved simultaneously, thereby improving the richness of image information and the reliability of decision-making.

[0003] Image registration of multi-source heterogeneous images is a prerequisite and key step for image fusion. Its task is to calibrate image data from different sources to the same spatial coordinate system to ensure the consistency of the position of the same target in different images. However, due to significant differences in imaging mechanisms and resolutions, traditional grayscale feature-based registration methods are often affected by factors such as illumination, scale, noise, and geometric distortion in heterogeneous scenes, making it difficult to achieve ideal registration accuracy. Meanwhile, image fusion algorithms need to synthesize and optimize the features of multi-source images based on the registration results to balance detail preservation, contrast enhancement, and overall visual consistency. However, existing fusion methods often face problems such as information redundancy, uneven weight distribution, and detail loss, resulting in the quality and realism of the fused image still needing improvement. In summary, existing multi-source heterogeneous image registration and fusion technologies still suffer from insufficient robustness and unstable fusion results, making it difficult to simultaneously meet the requirements of accuracy, efficiency, and reliability in diverse application scenarios. Summary of the Invention

[0004] The purpose of this invention is to provide a method for registration and fusion of multi-source heterogeneous image data, aiming to solve or improve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, the present invention provides the following solution: A method for registration and fusion of multi-source heterogeneous image data includes: Using an improved phase coherence model PC to analyze the reference image and the image to be registered Preprocessing is required; Preprocessed reference image and the image to be registered The original set of feature points is obtained by using the FAST feature detector to detect feature points. The aggregation feature strategy (AF) is used to process the original feature point set. The feature points are then filtered to obtain a set of feature points. Using the parity-even symmetry property of a two-dimensional Log-Gabor filter, the reference image is... and the image to be registered Perform multi-scale, multi-directional convolution to calculate the local phase response; A reference image is generated by using a maximum and minimum bandwidth weighting function combined with the local phase response. and the image to be registered The weighted phase direction characteristics; Based on the weighted phase direction features and feature point set, a reference image is generated by constructing a regularized logarithmic polar coordinate neighborhood grid. and the image to be registered The descriptor vector; Reference image based on descriptor vector and the image to be registered Feature matching is performed, the geometric transformation model is solved, and image spatial alignment is completed before cropping to obtain the registered image. and ; For registered images and After normalization and noise suppression, an improved local Wiener filter is used to extract salient structural information, resulting in a salient structure matrix. ; Using the registered normalized image G as the guide map, combined with the salient structure matrix Perform iterative joint filtering, and obtain convergent results after iteration. ; Based on the convergence results and registered images and The images are then fused to obtain the fused image F.

[0006] Furthermore, the expression for the phase consistency model PC is: In the formula, , , Intermediate quantities for calculating phase consistency moments; and These are the maximum and minimum moments of phase coherence, respectively; for exist Directional mapping; In order to be in The angle between directions.

[0007] Furthermore, the aggregation feature strategy (AF) is used on the original feature point set. After filtering, a set of feature points is obtained, including: For the original set of feature points Non-maximum suppression is performed, and filtering is based on the PC intensity value of each feature point. The expression is as follows: In the formula, For the set of feature points; These are the feature points selected based on their significance scores; The intensity threshold for feature points; This represents the PC strength value. These are the feature points retained after nonmaximum suppression.

[0008] Furthermore, the parity-even symmetry property of a two-dimensional Log-Gabor filter is used to refine the reference image. and the image to be registered Perform multi-scale, multi-directional convolution to calculate the local phase response, including: The expression for a two-dimensional Log-Gabor filter in the frequency domain is: In the formula, Logarithmic polar coordinates; and These represent the scale and orientation of the two-dimensional Log-Gabor; The center frequency of the two-dimensional Log-Gabor; and They are respectively and bandwidth; The frequency domain two-dimensional Log-Gabor filter is converted into a spatial domain two-dimensional Log-Gabor filter using inverse Fourier transform, as expressed by: In the formula, It is an even-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; for; It is an odd-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; Reference image and the image to be registered The responses of even-symmetric and odd-symmetric Log-Gabor filters at multiple scales are expressed as follows: In the formula, Input image; This is the response vector of the input image on the even-symmetric filter; This is the response vector of the input image on the odd-symmetric filter; This is a convolution operation; L2 energy normalization is performed on the vector in each direction o to obtain the local phase response, expressed as: In the formula, This is a local phase response; It is the minimum non-zero value.

[0009] Furthermore, a reference image is generated by using a maximum and minimum bandwidth weighting function combined with the local phase response. and the image to be registered The weighted phase direction characteristics include: The expression for the weighted bandwidth function model of a two-dimensional Log-Gabor filter is as follows: In the formula, and These are the maximum weighting coefficient and the minimum weighting coefficient, respectively; For exponentiation operator; A fractional measure of frequency propagation; The fraction of image frequency; control The sharpness of transitions in the model; It is the minimum non-zero value; Based on the maximum weighting coefficient and minimum weighting coefficient The initial weighted phase direction feature is calculated as follows: In the formula, This represents the initial weighted phase direction feature; for Odd-symmetric local phase response in the direction; for Even-symmetric local phase response in the direction; The rotation angle; It is the minimum non-zero value; Initial weighted phase direction features By transforming the direction angle, the weighted phase direction feature is obtained, expressed as: In the formula, Initial weighted phase direction features The radian value; Weighted phase direction characteristics; It is a non-negative constant term; It is the minimum non-zero value; This is a convolution operation.

[0010] Furthermore, based on the weighted phase direction features and feature point set, a reference image is generated by constructing a regularized logarithmic polar coordinate neighborhood grid. and the image to be registered The descriptor vector includes: A fixed-radius neighborhood is established with each feature point in the feature point set as the center. The neighborhood is divided into 12 angular sectors and 4 radial zones, for a total of 48 sub-regions. For each sub-region, iterate through all pixel positions and calculate based on the corresponding weighted phase direction value. The weighted phase direction value Mapping to 8 preset directions yields an 8-dimensional direction histogram vector, which is then concatenated to generate a 384-dimensional descriptor vector.

[0011] Furthermore, the reference image is determined based on the descriptor vector. and the image to be registered Feature matching is performed, the geometric transformation model is solved, and image spatial alignment is completed before cropping to obtain the registered image. and ,include: Based on the descriptor vector, use Euclidean distance to compare the reference image. and the image to be registered Feature matching is performed, and the consistency of matching is ensured through a bidirectional matching strategy. The Fast Random Consensus (FSC) algorithm is used to remove mismatched point pairs, obtain reliable corresponding point pairs, and establish a geometric transformation model. The image to be registered is obtained using a geometric transformation model. Transform to reference image In the coordinate system, the registered image is obtained. ; Calculate reference image With the registered image The effective overlapping region, and the reference image based on the effective overlapping region. and the registered image Cropping is performed to obtain the registered image. and .

[0012] Furthermore, for the registered images and After normalization and noise suppression, an improved local Wiener filter is used to extract salient structural information, resulting in a salient structure matrix. ,include: Registered images and The intensity is normalized to The normalized image is obtained. and ; The normalized image was processed using adaptive local Wiener filtering. and Denoising is performed to obtain the filtered image. and ; The expression for adaptive local Wiener filtering is: In the formula, For pixels The filtering results; To normalize the pixels in the image The value; and Local windows The mean and variance; In pixels A local window centered on the user; It is the minimum non-zero value; The preset noise variance; Calculate the filtered image and The gradient magnitude is expressed as: In the formula, and Filtered images and In spatial coordinates Gradient magnitude at; According to the filtered image and Gradient difference constructs decision graph The expression is: Decision graph Local mean filtering is performed, and a threshold function is used to generate a salient structure matrix of binary salient structures. The expression is: In the formula, It is a local mean; A window centered at pixel p; The number of pixels within the window; For window All pixels in; For decision graph The value of pixel p; It is a significant structure matrix; This is the feature threshold.

[0013] Furthermore, using the registered normalized image G as the guide map, combined with the salient structure matrix... Perform iterative joint filtering, and obtain convergent results after iteration. ,include: In the formula, for The significant structural value at pixel p after the next iteration; for The significant structural value at pixel p after the next iteration; This is a joint filtering operation; and These are the spatial domain scale parameters and the distance domain scale parameters, respectively. This represents the current iteration number; This represents the maximum number of iterations, after which a convergence result is obtained. .

[0014] Furthermore, the expression for the fused image F is: In the formula, To merge images.

[0015] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention discloses a registration and fusion method for multi-source heterogeneous image data. The method introduces a strategy combining weighted phase consistency and aggregated feature detection during the registration stage. It utilizes a two-dimensional Log-Gabor filter to extract weighted phase direction features, replacing traditional grayscale gradient information, effectively overcoming the influence of factors such as illumination variations, noise interference, and radiometric inconsistencies. Simultaneously, it employs a regularized logarithmic polar coordinate descriptor to construct a feature representation with scale and rotation invariance, ensuring stable matching and high-precision registration between images of different modalities.

[0016] In the fusion stage, this invention introduces a salient structure extraction and iterative joint filtering mechanism, taking into account both pixel spatial distance and intensity differences. This approach gradually transfers detailed features while preserving key structural information, achieving a balance between structure preservation and detail enhancement. This method eliminates the need for complex multi-scale transformations, boasts high computational efficiency, and produces fusion results with better clarity, hierarchy, and information complementarity. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a reference image for this embodiment. With the image to be registered A schematic diagram; Figure 3 The image to be registered in this embodiment Image registration A schematic diagram of superimposed images; Figure 4 The image to be registered in this embodiment Image registration A schematic diagram of the fused image F. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The purpose of this invention is to provide a method for registration and fusion of multi-source heterogeneous image data, aiming to solve or improve at least one of the above-mentioned technical problems.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, this invention provides a method for registration and fusion of multi-source heterogeneous image data, including: Step 1: Use the improved phase consistency model PC to analyze the reference image. and the image to be registered Preprocessing includes: The expression for the phase consistency model PC is: In the formula, , , Intermediate quantities for calculating phase consistency moments; and These are the maximum and minimum moments of phase coherence, respectively; for exist Directional mapping; In order to be in The angle between directions.

[0023] The above steps enhance structurally significant features in the image (such as edges, corners, and abrupt texture changes) while suppressing interference from variations in illumination and contrast, thus providing a robust and stable response map for subsequent feature detection.

[0024] Step 2, preprocess the reference image and the image to be registered The original set of feature points is obtained by using the FAST feature detector to detect feature points. The aggregation feature strategy (AF) is used to process the original feature point set. After filtering, a set of feature points is obtained, including: For the original set of feature points Non-maximum suppression is performed, and filtering is based on the PC intensity value of each feature point. The expression is as follows: In the formula, For the set of feature points; These are the feature points selected based on their significance scores; The intensity threshold for feature points; This represents the PC strength value. These are the feature points retained after nonmaximum suppression.

[0025] Step 3: Use the parity-even symmetry property of the two-dimensional Log-Gabor filter to process the reference image. and the image to be registered Perform multi-scale, multi-directional convolution to calculate the local phase response, including: The expression for a two-dimensional Log-Gabor filter in the frequency domain is: In the formula, Logarithmic polar coordinates; and These represent the scale and orientation of the two-dimensional Log-Gabor; The center frequency of the two-dimensional Log-Gabor; and They are respectively and bandwidth; The frequency domain two-dimensional Log-Gabor filter is converted into a spatial domain two-dimensional Log-Gabor filter using inverse Fourier transform, as expressed by: In the formula, It is an even-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; for; It is an odd-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; Reference image and the image to be registered The responses of even-symmetric and odd-symmetric Log-Gabor filters at multiple scales are expressed as follows: In the formula, Input image; This is the response vector of the input image on the even-symmetric filter; This is the response vector of the input image on the odd-symmetric filter; This is a convolution operation; L2 energy normalization is performed on the vector in each direction o to obtain the local phase response, expressed as: In the formula, This is a local phase response; The minimum non-zero value is preferred. To prevent the denominator from being 0.

[0026] Step 4: Using the maximum and minimum bandwidth weighting function, combined with the local phase response, generate a reference image. and the image to be registered The weighted phase direction characteristics include: The expression for the weighted bandwidth function model of a two-dimensional Log-Gabor filter is as follows: In the formula, and These are the maximum weighting coefficient and the minimum weighting coefficient, respectively; For exponentiation operator; A fractional measure of frequency propagation; The fraction of image frequency; control The sharpness of transitions in the model; It is the minimum non-zero value; Based on the maximum weighting coefficient and minimum weighting coefficient The initial weighted phase direction feature is calculated as follows: In the formula, This represents the initial weighted phase direction feature; for Odd-symmetric local phase response in the direction; for Even-symmetric local phase response in the direction; The rotation angle; It is the minimum non-zero value; Initial weighted phase direction features By transforming the direction angle, the weighted phase direction feature is obtained, expressed as: In the formula, Initial weighted phase direction features The radian value; Weighted phase direction characteristics; It is a non-negative constant term; It is the minimum non-zero value; This is a convolution operation.

[0027] Step 5: Based on the weighted phase direction features and feature point set, a reference image is generated by constructing a regularized logarithmic polar coordinate neighborhood grid. and the image to be registered The descriptor vector includes: A fixed-radius neighborhood is established with each feature point in the feature point set as the center. The neighborhood is divided into 12 angular sectors and 4 radial zones, for a total of 48 sub-regions. For each sub-region, iterate through all pixel positions and calculate based on the corresponding weighted phase direction value. The weighted phase direction value Mapping to eight preset directions yields an 8-dimensional direction histogram vector, which is then concatenated to generate a 384-dimensional descriptor vector. The eight preset directions are... The range is divided into 8 equally spaced directional intervals.

[0028] Step 6: Pair the reference image with the descriptor vector. and the image to be registered Feature matching is performed, the geometric transformation model is solved, and image spatial alignment is completed before cropping to obtain the registered image. and ,include: Based on the descriptor vector, use Euclidean distance to compare the reference image. and the image to be registered Feature matching is performed, and the consistency of matching is ensured through a bidirectional matching strategy. The Fast Random Consensus (FSC) algorithm is used to remove mismatched point pairs, obtain reliable corresponding point pairs, and establish a geometric transformation model. The image to be registered is obtained using a geometric transformation model. Transform to reference image In the coordinate system, the registered image is obtained. ; Calculate reference image With the registered image The effective overlapping region, and the reference image based on the effective overlapping region. and the registered image Cropping is performed to obtain the registered image. and .

[0029] Step 7, register the images and After normalization and noise suppression, an improved local Wiener filter is used to extract salient structural information, resulting in a salient structure matrix. ,include: Registered images and The intensity is normalized to The normalized image is obtained. and ; The normalized image was processed using adaptive local Wiener filtering. and Denoising is performed to obtain the filtered image. and ; The expression for adaptive local Wiener filtering is: In the formula, For pixels The filtering results; To normalize the pixels in the image The value; and Local windows The mean and variance; In pixels A local window centered on the user; It is the minimum non-zero value; The preset noise variance; Calculate the filtered image and The gradient magnitude is expressed as: In the formula, and Filtered images and In spatial coordinates Gradient magnitude at; According to the filtered image and Gradient difference constructs decision graph The expression is: Decision graph Local mean filtering is performed, and a threshold function is used to generate a salient structure matrix of binary salient structures. The expression is: In the formula, It is a local mean; A window centered at pixel p; The number of pixels within the window; For window All pixels in; For decision graph The value of pixel p; It is a significant structure matrix; In this embodiment, the feature threshold is used. .

[0030] Step 8: Using the registered normalized image G as the guide map, and combining it with the salient structure matrix... Perform iterative joint filtering, and obtain convergent results after iteration. The expression is: In the formula, for The significant structural value at pixel p after the next iteration; for The significant structural value at pixel p after the next iteration; This is a joint filtering operation; and These are the spatial domain scale parameters and the distance domain scale parameters, respectively. This represents the current iteration number; This represents the maximum number of iterations, after which a convergence result is obtained. .

[0031] The above steps smoothly update the salient structure matrix within the large-scale structural neighborhood. , guide map The details are gradually transferred to the salient structural regions, and convergence results are obtained after iteration. .

[0032] Step 9, based on the convergence results and registered images and The image is fused to obtain the fused image F, which is expressed as: In the formula, To merge images.

[0033] To verify the effectiveness of this method, specific implementations were carried out, including: like Figure 2 As shown, in the initial state, the reference image With the image to be registered There are obvious geometric distortions and spatial misalignments. Features in the scene in the attached figure, such as flower beds and vehicle distribution, cannot be accurately matched. The introduced geometric errors cause the multi-source information association to fail, resulting in feature re-detection or missed detection during feature extraction, which hinders the complementarity and fusion of multimodal image information.

[0034] For the above images to be registered Existing problems, such as Figure 3 As shown, the registered image on the left As a reference, the images to be registered Geometric transformations such as rotation and translation were performed to correct the errors, resulting in a registered image. With registered images This enables precise alignment of core scene features between two images.

[0035] To more intuitively demonstrate the registration effect, the two registered images are superimposed in a mosaic checkerboard pattern. From the superimposed image, it can be clearly observed that key features such as the outline of the flower bed and the edges of the vehicle completely overlap at the stitching point. The visualized image directly shows the registration effect.

[0036] like Figure 4 As shown, the registered image With registered images The fused image F is obtained by inputting it into the image fusion module. Observe the fused image. It can be seen that the fusion result preserves both the registered images and the fused images. The infrared image clearly shows structural features such as the arrangement of vehicles and the outline of flower beds, and also includes registered images. Nighttime features in the (visible light image) enhanced information that was previously insufficient for single-modality recognition. The overall image showed no ghosting, blurring, or information loss; target recognition and information richness were significantly improved, effectively achieving complementary advantages from multiple information sources and providing a higher-quality visual foundation for subsequent tasks.

[0037] In addition to qualitative analysis, the results showed good performance on common quantitative indicators of image fusion, including mutual information index (MI): 7.2616, fusion quality (Q_abf): 0.8116, and structural similarity (SSIM): 0.9856.

[0038] This invention introduces a strategy combining weighted phase consistency and aggregated feature detection in the registration stage. It utilizes a two-dimensional Log-Gabor filter to extract weighted phase direction features, replacing traditional grayscale gradient information, effectively overcoming the influence of factors such as illumination variations, noise interference, and radiometric inconsistencies. Simultaneously, it employs a regularized logarithmic polar coordinate descriptor to construct a feature representation with scale and rotation invariance, ensuring stable matching and high-precision registration between images of different modalities.

[0039] In the fusion stage, this invention introduces a salient structure extraction and iterative joint filtering mechanism, taking into account both pixel spatial distance and intensity differences. This approach gradually transfers detailed features while preserving key structural information, achieving a balance between structure preservation and detail enhancement. This method eliminates the need for complex multi-scale transformations, boasts high computational efficiency, and produces fusion results with better clarity, hierarchy, and information complementarity.

[0040] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0041] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for registration and fusion of multi-source heterogeneous image data, characterized in that, include: Using an improved phase coherence model PC to analyze the reference image and the image to be registered Preprocessing is required; Preprocessed reference image and the image to be registered The original set of feature points is obtained by using the FAST feature detector to detect feature points. The aggregation feature strategy (AF) is used to process the original feature point set. The feature points are then filtered to obtain a set of feature points. Using the parity-even symmetry property of a two-dimensional Log-Gabor filter, the reference image is... and the image to be registered Perform multi-scale, multi-directional convolution to calculate the local phase response; A reference image is generated by using a maximum and minimum bandwidth weighting function combined with the local phase response. and the image to be registered The weighted phase direction characteristics; Based on the weighted phase direction features and feature point set, a reference image is generated by constructing a regularized logarithmic polar coordinate neighborhood grid. and the image to be registered The descriptor vector; Reference image based on descriptor vector and the image to be registered Feature matching is performed, the geometric transformation model is solved, and image spatial alignment is completed before cropping to obtain the registered image. and ; For registered images and After normalization and noise suppression, an improved local Wiener filter is used to extract salient structural information, resulting in a salient structure matrix. ; Using the registered normalized image G as the guide map, combined with the salient structure matrix Perform iterative joint filtering, and obtain convergent results after iteration. ; Based on the convergence results and registered images and The images are then fused to obtain the fused image F.

2. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The expression for the phase consistency model PC is: In the formula, , , Intermediate quantities for calculating phase consistency moments; and These are the maximum and minimum moments of phase coherence, respectively; for exist Directional mapping; In order to be in The angle between directions.

3. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The aggregation feature strategy AF is used to process the original feature point set. After filtering, a set of feature points is obtained, including: For the original set of feature points Non-maximum suppression is performed, and filtering is based on the PC intensity value of each feature point. The expression is as follows: In the formula, For the set of feature points; These are the feature points selected based on their significance scores; The intensity threshold for feature points; This represents the PC strength value. These are the feature points retained after nonmaximum suppression.

4. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The parity-even symmetry property of the two-dimensional Log-Gabor filter is used for the reference image. and the image to be registered Perform multi-scale, multi-directional convolution to calculate the local phase response, including: The expression for a two-dimensional Log-Gabor filter in the frequency domain is: In the formula, Logarithmic polar coordinates; and These represent the scale and orientation of the two-dimensional Log-Gabor; The center frequency of the two-dimensional Log-Gabor; and They are respectively and bandwidth; The frequency domain two-dimensional Log-Gabor filter is converted into a spatial domain two-dimensional Log-Gabor filter using inverse Fourier transform, as expressed by: In the formula, It is an even-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; for; It is an odd-symmetric filter for a two-dimensional Log-Gabor filter in the spatial domain; Reference image and the image to be registered The responses of even-symmetric and odd-symmetric Log-Gabor filters at multiple scales are expressed as follows: In the formula, Input image; This is the response vector of the input image on the even-symmetric filter; This is the response vector of the input image on the odd-symmetric filter; This is a convolution operation; L2 energy normalization is performed on the vector in each direction o to obtain the local phase response, expressed as: In the formula, This is a local phase response; It is the minimum non-zero value.

5. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The reference image is generated by using a maximum and minimum bandwidth weighting function combined with the local phase response. and the image to be registered The weighted phase direction characteristics include: The expression for the weighted bandwidth function model of a two-dimensional Log-Gabor filter is as follows: In the formula, and These are the maximum weighting coefficient and the minimum weighting coefficient, respectively; For exponentiation operator; A fractional measure of frequency propagation; The fraction of image frequency; control The sharpness of transitions in the model; It is the minimum non-zero value; Based on the maximum weighting coefficient and minimum weighting coefficient The initial weighted phase direction feature is calculated as follows: In the formula, This represents the initial weighted phase direction feature; for Odd-symmetric local phase response in the direction; for Even-symmetric local phase response in the direction; The rotation angle; It is the minimum non-zero value; Initial weighted phase direction features By transforming the direction angle, the weighted phase direction feature is obtained, expressed as: In the formula, Initial weighted phase direction features The radian value; Weighted phase direction characteristics; It is a non-negative constant term; It is the minimum non-zero value; This is a convolution operation.

6. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The reference image is generated by constructing a regularized logarithmic polar coordinate neighborhood grid based on the weighted phase direction features and feature point set. and the image to be registered The descriptor vector includes: A fixed-radius neighborhood is established with each feature point in the feature point set as the center. The neighborhood is divided into 12 angular sectors and 4 radial zones, for a total of 48 sub-regions. For each sub-region, iterate through all pixel positions and calculate based on the corresponding weighted phase direction value. The weighted phase direction value Mapping to 8 preset directions yields an 8-dimensional direction histogram vector, which is then concatenated to generate a 384-dimensional descriptor vector.

7. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The reference image is based on the descriptor vector. and the image to be registered Feature matching is performed, the geometric transformation model is solved, and image spatial alignment is completed before cropping to obtain the registered image. and ,include: Based on the descriptor vector, use Euclidean distance to compare the reference image. and the image to be registered Feature matching is performed, and the consistency of matching is ensured through a bidirectional matching strategy. The Fast Random Consensus (FSC) algorithm is used to remove mismatched point pairs, obtain reliable corresponding point pairs, and establish a geometric transformation model. The image to be registered is obtained using a geometric transformation model. Transform to reference image In the coordinate system, the registered image is obtained. ; Calculate reference image With the registered image The effective overlapping region, and the reference image based on the effective overlapping region. and the registered image Cropping is performed to obtain the registered image. and .

8. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The registered image and After normalization and noise suppression, an improved local Wiener filter is used to extract salient structural information, resulting in a salient structure matrix. ,include: Registered images and The intensity is normalized to The normalized image is obtained. and ; The normalized image was processed using adaptive local Wiener filtering. and Denoising is performed to obtain the filtered image. and ; The expression for adaptive local Wiener filtering is: In the formula, For pixels The filtering results; To normalize the pixels in the image The value; and Local windows The mean and variance; In pixels A local window centered on the user; It is the minimum non-zero value; The preset noise variance; Calculate the filtered image and The gradient magnitude is expressed as: In the formula, and Filtered images and In spatial coordinates Gradient magnitude at; According to the filtered image and Gradient difference constructs decision graph The expression is: Decision graph Local mean filtering is performed, and a threshold function is used to generate a salient structure matrix of binary salient structures. The expression is: In the formula, It is a local mean; A window centered at pixel p; The number of pixels within the window; For window All pixels in; For decision graph The value of pixel p; It is a significant structure matrix; This is the feature threshold.

9. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The normalized image G after registration is used as the guide map, combined with the salient structure matrix. Perform iterative joint filtering, and obtain convergent results after iteration. ,include: In the formula, for The significant structural value at pixel p after the next iteration; for The significant structural value at pixel p after the next iteration; This is a joint filtering operation; and These are the spatial domain scale parameters and the distance domain scale parameters, respectively. This represents the current iteration number; This represents the maximum number of iterations, after which a convergence result is obtained. .

10. The registration and fusion method for multi-source heterogeneous image data according to claim 1, characterized in that, The expression for the fused image F is: In the formula, To merge images.

Images