A method and system for panchromatic sharpening against registration error

Abstract

A panchromatic sharpening method and system against registration error, the method comprising: 1, obtaining the panchromatic image and multispectral image of the same ground object area, performing up-sampling processing on the multispectral image to generate a multispectral image, and then performing superpixel segmentation on the multispectral image to construct an edge mask; 2, calculating the weighted mean of the spectral information of the multispectral image; 3, color coding and decoupling the multispectral image MSU using the YUV color space to obtain the spectral channel UV and the brightness channel Y; 4, establishing a linear transmission model of the brightness channel, and obtaining the brightness component by solving the linear transmission model; 5, taking the brightness component as the reference, combining the edge mask, and filling the non-boundary area according to the chroma initialization execution, establishing a spectral diffusion model based on local radiation similarity and solving, and finally obtaining a fused image. The present application realizes the collaborative optimization of radiation consistency and detail fidelity, and reduces the spectral distortion of the fused image.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a panchromatic sharpening method and system resistant to registration errors. BACKGROUND

[0002] In remote sensing image processing, panchromatic images (PAN) and multispectral images (MS) each have their unique advantages. Panchromatic images provide high spatial resolution but only contain structural information; while multispectral images have multiple bands and can provide rich spectral information. However, individual panchromatic images or multispectral images often cannot meet the needs in remote sensing applications, so image fusion technology emerges as the times require, aiming to combine the advantages of both and improve the spatial and spectral resolution of the image.

[0003] Panchromatic sharpening technology has emerged a large number of methods in the past few decades, which can be roughly divided into four categories: Component Substitution (CS) based methods, Multiresolution Analysis (MRA) based methods, Variational Optimization-based (VO) methods, and Deep Learning (DL) based methods. Among them, CS and MRA based methods are widely used due to their simple and efficient processing characteristics. However, these methods are sensitive to geometric registration errors between the images to be fused. For example, MRA based methods strictly depend on the accurate alignment of high-frequency information, and when registration errors exist, ringing artifacts are easily induced, which is a significant challenge in engineering practice. In recent years, VO and DL based methods have gradually emerged, and with their strong modeling ability and adaptability, they have solved many limitations of traditional methods and found a better balance between spatial and spectral quality preservation. Although VO and DL based methods perform well, they often require a large amount of computing resources, and training and application on small sample datasets may face overfitting or unstable performance problems. Therefore, in actual remote sensing image processing, how to balance the computational efficiency and fusion effect is still a problem that needs to be continuously optimized.

[0004] In practical applications, panchromatic images and multispectral images used for image fusion often have registration errors. Many existing panchromatic sharpening methods are usually developed and evaluated on image data that have undergone fine geometric correction, and when these methods are directly applied to actual images, image distortion often occurs. SUMMARY

[0005] The present application provides a panchromatic sharpening method and system resistant to registration errors to solve the technical problems mentioned in the background art.

[0006] To achieve the above object, the technical scheme of the present application is as follows:

[0007] The present application provides an anti-registration error panchromatic sharpening method, comprising the following steps:

[0008] S1, obtaining panchromatic image PAN and multispectral image MSU of the same size, then performing superpixel segmentation on the multispectral image MSU, constructing an edge mask using the superpixel boundary, and excluding the interference area caused by the registration error between the panchromatic image PAN and the multispectral image MSU in combination with the edge mask;

[0009] S2, calculating the weighted mean of R / G / B band spectral information in the neighborhood of the centroid of each superpixel;

[0010] S3, performing color encoding and decoupling on the multispectral image MSU using YUV color space to obtain spectral channel UV and luminance channel Y, and isolating the interference of the registration error on the spectral component using the spectral channel UV;

[0011] S4, fitting the linear relationship between the luminance channel Y of the multispectral image MSU and the luminance channel I of the panchromatic image PAN in the local neighborhood of each pixel, establishing a linear transmission model of the luminance channel Y according to the linear relationship, and obtaining the luminance component of the fused image by solving the linear transmission model;

[0012] S5, performing color encoding on the weighted mean of R / G / B band spectral information using YUV color space to generate the initial chrominance components U and V of the fused image; filling the non-boundary area according to the execution of chrominance initialization in combination with the edge mask and the initial chrominance components U and V with the luminance component as the reference, establishing a spectral diffusion model based on local radiometric similarity and solving, and finally obtaining the fused image.

[0013] The present application also provides an anti-registration error panchromatic sharpening system, comprising a computer device programmed or configured to perform the panchromatic sharpening method.

[0014] The present application has the following beneficial effects:

[0015] The present application provides an anti-registration error panchromatic sharpening method, which can effectively suppress the spectral coupling distortion caused by registration error, maintain high spatial resolution characteristics, and significantly improve the spectral fidelity of the fused image.

[0016] Compared with the classical panchromatic sharpening traditional method, the present application can keep good visual effect after introducing registration error into the image, has certain robustness to the registration error, and has strong practicability; the present application designs a linear transmission model to perform radiation correction. When the image is unevenly distributed and local difference of ground objects is too large, the present application has significant advantages in reducing spectral distortion, keeping spatial structure integrity and improving visual quality. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flow chart of the panchromatic sharpening method in the present application;

[0018] Figure 2 is a radiation correction effect diagram of the embodiment of the present application;

[0019] Figure 3 is a spectral information diffusion effect diagram of the embodiment of the present application;

[0020] Figure 4 is a comparison diagram of the fused image of the full resolution fusion using IKONOS satellite data set in the embodiment of the present application; in Figure 4 , figure a is a panchromatic image PAN to be fused; figure b is a multi-spectral image MSU after up-sampling and to be fused; figure c is a fusion effect diagram realized by the present application without introducing translation registration error; figures d, e, f, g, h, i respectively represent fusion effect diagrams of IHS, PCA, GS, ATWT, MTF-GLP, TV method under introducing translation registration error; figure j represents a fusion effect diagram realized by the present application;

[0021] Figure 5 is a comparison diagram of the fused image of the reduced resolution fusion using IKONOS satellite data set in the embodiment of the present application; in Figure 5 , figure a is a panchromatic image PAN to be fused; figure b is a multi-spectral image MSU after up-sampling and to be fused; figure c is an original multi-spectral image MS; figures d, e, f, g, h, i respectively represent fusion effect diagrams of IHS, PCA, GS, ATWT, MTF-GLP, TV method under introducing translation registration error; figure j represents a fusion effect diagram realized by the present application. DETAILED DESCRIPTION

[0022] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The preferred embodiments of the present application are shown in the drawings. However, the present application can be realized in many other different forms, and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0023] Reference Figure 1The embodiment of the application provides a panchromatic sharpening method against registration error, comprising the following steps:

[0024] S1, acquiring a panchromatic image (PAN) and a multi-spectral image (MS) of the same ground object region, performing 4 times spatial resolution up-sampling processing on the multi-spectral image MS to generate a multi-spectral image (MSU) with the same size as the panchromatic image PAN, then performing superpixel segmentation on the multi-spectral image MSU, constructing an edge mask using a superpixel boundary, and combining the edge mask to exclude an interference area caused by registration error between the panchromatic image PAN and the multi-spectral image MSU;

[0025] S2, calculating a weighted mean of R / G / B band spectral information in a centroid neighborhood of each superpixel;

[0026] S3, performing color coding and decoupling on the multi-spectral image MSU using a YUV color space to obtain a spectral channel UV and a luminance channel Y, and using the spectral channel UV to isolate the interference of the spectral component caused by the registration error; the spectral component is a chroma of the multi-spectral image MSU;

[0027] S4, fitting a linear relationship between the luminance channel Y of the multi-spectral image MSU and the luminance channel I of the panchromatic image PAN in a local neighborhood of each pixel, establishing a linear transmission model of the luminance channel Y according to the linear relationship, and obtaining a luminance component of a fusion image by solving the linear transmission model ;

[0028] S5, performing color coding on the weighted mean of R (red) / G (green) / B (blue) band spectral information using a YUV color space to generate initial chroma components U and V of the fusion image; the initial chroma component U is the difference between the blue channel and the luminance value, and the initial chroma component V is the difference between the red channel and the luminance value, which functions to describe the image color and saturation; taking the luminance component as a reference, combining the edge mask and the initial chroma components U and V to fill a non-boundary area according to chroma initialization, establishing a spectral diffusion model based on local radiometric similarity and solving, and finally obtaining the fusion image.

[0029] The method provided by the application can effectively suppress the spectral coupling distortion caused by the registration error, while maintaining the high spatial resolution characteristics, and significantly improves the spectral fidelity of the fusion image.

[0030] Compared with the traditional method of classical panchromatic sharpening, the application can maintain good visual effect after introducing registration error into the image, has certain robustness to registration error, and has strong practicability; the application designs a linear transmission model for radiation correction.

[0031] In some embodiments, the resolution of the panchromatic image PAN in S1 is higher than the resolution of the multispectral image MS; specifically, the panchromatic image PAN is selected as a high-resolution panchromatic image; and the multispectral image MS is selected as a low-resolution multispectral image. Preferably, the spatial resolution of the high-resolution panchromatic image PAN is 1 m (with a frame of 1024*1024 pixels); and the spatial resolution of the low-resolution multispectral image is 4 m (with a frame of 256*256 pixels).

[0032] In some embodiments, S1 specifically comprises the following steps:

[0033] S11, acquiring a panchromatic image PAN and a multispectral image MS of the same ground object region;

[0034] S12, performing 4 times spatial resolution up-sampling processing on the multispectral image MS to generate a multispectral image MSU with the same size as the panchromatic image PAN;

[0035] S13, setting a compactness parameter in the superpixel segmentation process, preferably, the compactness parameter is set to 10, and the number of superpixels is dynamically confirmed according to the size of the multispectral image MSU; then, the multispectral image MSU is segmented into superpixels according to the compactness parameter and the number of superpixels and using a SLIC (Simple Linear Iterative Clustering) superpixel segmentation algorithm;

[0036] S14, performing morphological dilation processing on the superpixel boundary region to generate an edge mask;

[0037] S15, combining the edge mask to exclude the interference area caused by the registration error between the panchromatic image PAN and the multispectral image MSU, constructing a spectral diffusion model based on local radiation similarity, and then using the spectral diffusion model to constrain the spectral features in the multispectral image MSU to be transmitted in the homogeneous region, so as to effectively suppress the spectral-spatial coupling distortion caused by the registration error.

[0038] In some embodiments, the calculation formula of the weighted mean of R / G / B band spectral information in S2 is:

[0039] ;

[0040] wherein, represents the weighted mean of R / G / B band spectral information; represents all pixels in the i-th superpixel centroid neighborhood; represents the pixel value ; represents the element k; represents the pixel value occupies the weight, and the weight is calculated as follows:

[0041] ;

[0042] wherein, σ is an adaptive Gaussian kernel width, is the Euclidean space distance from the pixel to the centroid; represents a natural exponential function.

[0043] In some embodiments, with reference to Figure 2 , the S4 specifically comprises the following steps:

[0044] S41, extracting the luminance channel I (corresponding to the PAN luminance channel in Figure 2 ) of the panchromatic image PAN and the luminance channel Y (corresponding to the MS luminance channel in Figure 2 ) of the multispectral image MSU;

[0045] S42, fitting a linear relationship between the luminance channel Y of the multispectral image MSU and the luminance channel I of the panchromatic image PAN within the local neighborhood of each superpixel, and establishing a linear transmission model of the luminance channel Y according to the linear relationship;

[0046] S43, calculating the covariance between the panchromatic image PAN and the multispectral image MSU through a sliding window, i.e. the variance and mean value within each pixel and its neighborhood between the panchromatic image PAN and the multispectral image MSU, and solving the linear transmission model by using an optimization method to obtain the optimal values of the radiation scaling factor and the offset b;

[0047] S44, solving the luminance component of the fused image according to the optimal values of the radiation scaling factor .

[0048] In some embodiments, the linear transmission model in the S42 is specifically as follows:

[0049] ;

[0050] wherein, represents a loss function; represents a radiation scaling factor; represents the pixel value of the panchromatic image PAN and the offset of the corresponding multispectral image MSU pixel value; represents the weight coefficient for controlling the size of the radiometric scaling factor ; represents the local neighborhood of the pixel point.

[0051] In some embodiments, the calculation formula of the optimal value of the radiometric scaling factor and the offset b in S43 is respectively:

[0052] ;

[0053] ;

[0054] wherein, is the covariance of the luminance channel I and the luminance channel Y; represents the variance; , respectively represents the mean of the luminance channel I and the luminance channel Y;

[0055] The calculation formula of the luminance component in S44 is:

[0056] .

[0057] In some embodiments, S5 specifically comprises the following steps:

[0058] S51, color encoding the weighted mean of the R / G / B band spectral information in the YUV color space to generate the initial chroma components U and V of the fusion image;

[0059] S52, performing chroma initialization filling on the non-boundary region based on the luminance component of the fusion image in S4, in combination with the edge mask in S1 and the initial chroma components U and V of the fusion image;

[0060] S53, after the chroma initialization filling, establishing a spectral diffusion model based on local radiometric similarity for the heterogeneous boundary region defined by the edge mask;

[0061] S54, realizing the spectral consistency propagation (diffusion effect as shown in ) of the chroma component Figure 3 in the heterogeneous boundary region by iteratively solving the established spectral diffusion model based on local radiometric similarity, to obtain the optimized YUV image;

[0062] S55, inversely transforming the optimized YUV image to the RGB space to obtain the final fusion image.

[0063] In some embodiments, the spectral diffusion model in S53 is specified as follows:

[0064] ;

[0065] wherein, represents a loss function; represents the U value of the r pixel, i.e. the chroma value; represents the U value of the s pixel; represents the neighborhood of the r pixel; represents the pixel s; represents the affinity weight, and the affinity weight is obtained based on the squared difference of the brightness between two pixels, and is expressed as:

[0066] ;

[0067] wherein, represents the brightness value of the r pixel; a represents a positive correlation; and e represents a natural constant; represents the brightness value of the s pixel; is the variance of the brightness value in the local window centered on the pixel r.

[0068] In some embodiments, after S5, the method further comprises the following steps:

[0069] Step 6: Select some classic panchromatic sharpening methods to perform fusion on the same data set, and introduce registration errors to verify the robustness of the method to spatial registration errors.

[0070] Specifically, the selected classic panchromatic sharpening methods include Principal Component Analysis (PCA), Intensity-Hue-Saturation (IHS) transformation method, Gram-Schmidt (GS) method, à trous wavelet transform (ATWT) method, Modulation Transfer Functions-Generalized Laplacian Pyramid (MTF-GLP) method, and Total Variation (TV) based method. Before fusion, the original image (i.e. the multispectral image MS) is cropped to simulate the possible registration errors that may exist in practice, so as to verify the robustness to registration errors.

[0071] To verify the robustness of each method to registration error, a test set was constructed by simulating translational registration error (PAN panchromatic image was cropped at 1:1022, MS multispectral image was cropped at 3:256 and then interpolated to recover) before fusion. Both reduced resolution (as per Wald protocol) and full resolution fusion modes were evaluated. Translational registration error refers to the deviation between the registration result and the true position due to the inaccuracy of translational transformation in the image registration process.

[0072] In the reduced resolution fusion experiment, the full-color image PAN and the multispectral image MS were first down-sampled to construct a low-resolution image pair. Then, each panchromatic sharpening method was applied to the down-sampled image for fusion to generate a reconstructed low-resolution fusion image. At this time, the original multispectral image MS was used as the reference image to facilitate quantitative analysis. In the full resolution fusion experiment, the original full-color image PAN and the up-sampled multispectral image MS (i.e., multispectral image MSU) were directly fused to generate a fusion image with high spatial resolution and multispectral information, which was used to comprehensively evaluate the fusion effect.

[0073] To illustrate the actual effect of the present application, the relevant experimental results and analysis are now provided:

[0074] For ease of description, the reduced resolution fusion is referred to as "simulation experiment", and the full resolution fusion is referred to as "real experiment". Taking IKONOS remote sensing images as an example, the results of the real experiment are shown in Figure 4 , where Fig. a is the full-color image PAN to be fused, Fig. b is the up-sampled multispectral image MSU to be fused, Fig. c is the fusion effect achieved by the present application without introducing translational registration error. Figs. d, e, f, g, h, i respectively represent the fusion effect of IHS, PCA, GS, ATWT, MTF-GLP, and TV methods with translational registration error introduced, and Fig. j represents the fusion effect achieved by the present application.

[0075] The results of the simulation experiment are shown in Figure 5 . In the simulation experiment, the spatial structure is greatly destroyed, and the images to be fused are all down-sampled versions. Fig. a is the full-color image PAN to be fused, Fig. b is the up-sampled multispectral image MSU to be fused, and Fig. c is the original multispectral image MS (as the reference image). Figs. d, e, f, g, h, i respectively represent the fusion effect of IHS, PCA, GS, ATWT, MTF-GLP, and TV methods with translational registration error introduced, and Fig. j represents the fusion effect achieved by the present application. Two local areas are selected from each image and enlarged to more intuitively observe the fusion effect.

[0076] In the real experiment, the component substitution method performs well in the spatial information preservation, but there is obvious spectral distortion phenomenon. For example, the grassland region appears white after processing, which is quite different from the dark green in the original multispectral image MSU. The TV-based fusion method also has similar spectral distortion problems. In contrast, the method proposed in this paper and the multi-resolution analysis method well preserve the spectral characteristics of the original image during the fusion process, and there is almost no obvious color difference compared with the multispectral image MSU. When the registration error is introduced, the spectral distortion of various fusion methods appears to varying degrees. Among them, IHS, PCA, GS and TV methods all have color distortion in the two local enlarged areas. The fusion images generated by ATWT and MTF-GLP methods have serious spatial distortion, and there are significant "artifacts" at the boundary, which seriously affect the overall visual effect. In contrast, the fusion image generated by the present application performs well in spatial details and color restoration, and can effectively resist the interference of registration error, and has strong robustness.

[0077] In the simulation experiment, due to the desampling processing of the source image, a large amount of spatial information is lost, so the main evaluation index is the spectral fidelity of the image. Consistent with the results of the real experiment, after introducing the registration error, by observing the local enlarged area, it can be found that all methods have a certain degree of loss of spectral information in the edge area. The multi-resolution analysis method has serious loss of edge details in the fusion process, and the spatial distortion is obvious. In comparison, the present application still maintains a good balance in the preservation of spatial details and spectral restoration, and the color information is relatively complete, and the fusion effect is relatively superior.

[0078] In summary, compared with the existing traditional algorithms, the present application has obvious advantages in spatial information preservation and spectral fidelity. Whether in full resolution evaluation or in the simulation experiment of desampling, the present application can effectively balance the integrity of spatial details and spectral information, especially in the case of introducing registration error, the present application has strong anti-interference and stability. Compared with the spectral distortion of IHS, PCA, GS and TV methods, and the serious spatial distortion and "artifact" phenomenon of ATWT and MTF-GLP methods, the fusion image generated by the present application has more natural visual effect, higher color restoration degree and clearer spatial details.

[0079] Therefore, the present application has high practical value and popularization potential in the field of remote sensing image fusion, and can provide higher quality image data for related applications.

[0080] Another aspect of the present application also provides an anti-registration error panchromatic sharpening system, which comprises a computer device programmed or configured to perform the panchromatic sharpening method.

[0081] The above merely describes specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Furthermore, the technical solutions of each embodiment of the present application can be combined with each other, but it must be based on the realization of the ordinary skilled in the art, when the combination of the technical solutions appears contradictory or unachievable, it should be considered that the combination of the technical solutions does not exist, and is not within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method of panchromatic sharpening against registration errors, characterized in that, The method comprises the following steps: S1, obtaining a panchromatic image PAN and a multispectral image MSU of the same size, and then performing superpixel segmentation on the multispectral image MSU, and constructing an edge mask using superpixel boundaries; S2, calculating a weighted average of R / G / B band spectral information in a neighborhood of a centroid of each superpixel; S3, performing color encoding and decoupling on the multispectral image MSU in a YUV color space to obtain a spectral channel UV and a luminance channel Y; S4, fitting a linear relationship between the luminance channel Y of the multispectral image MSU and a luminance channel I of the panchromatic image PAN in a local neighborhood of each pixel, establishing a linear transmission model of the luminance channel Y according to the linear relationship, and obtaining a luminance component of the fused image by solving the linear transmission model; S5, performing color encoding on the weighted average of R / G / B band spectral information in a YUV color space to generate initial chrominance components U and V of the fused image; performing chrominance initialization filling on a non-boundary region according to the luminance component as a reference, combining the edge mask and the initial chrominance components U and V, establishing a spectral diffusion model based on local radiometric similarity, and solving the spectral diffusion model to finally obtain the fused image; The S4 specifically comprises the following steps: S41, extracting a luminance channel I of the panchromatic image PAN and a luminance channel Y of the multispectral image MSU; S42, fitting a linear relationship between the luminance channel Y of the multispectral image MSU and the luminance channel I of the panchromatic image PAN in a local neighborhood of each pixel, and establishing a linear transmission model of the luminance channel Y according to the linear relationship; S43. Calculate the covariance between the panchromatic image PAN and the multispectral image MSU by a sliding window, i.e. the variance and the mean value of each pixel and its neighborhood between the panchromatic image PAN and the multispectral image MSU, and solve the linear transfer model by an optimization method to get the optimal value of the radiometric scaling factor a and the offset b. and the offset b. S44, determining a radiation scaling factor and the optimal value of the offset b to solve the luminance component of the fused image ; The linear transmission model in the S42 is specifically as follows: ； wherein, represents a loss function; represents an offset of the pixel value of the panchromatic image PAN and the pixel value of the corresponding multispectral image MSU; represents a weight coefficient for controlling the size of the radiometric scaling factor ; represents a local neighborhood of a pixel.

2. The method of anti-registration error panchromatic sharpening of claim 1, wherein, The resolution of the panchromatic image PAN in the S1 is higher than the resolution of the multispectral image MS.

3. The method of anti-registration error panchromatic sharpening of claim 1, wherein, The S1 specifically comprises the following steps: S11, obtaining a panchromatic image PAN and a multispectral image MS of the same ground object region; S12, performing up-sampling processing on the multispectral image MS to generate a multispectral image MSU of the same size as the panchromatic image PAN; S13, setting a compactness parameter in a superpixel segmentation process, dynamically confirming a superpixel number according to the size of the multispectral image MSU, and then performing superpixel segmentation on the multispectral image MSU according to the compactness parameter and the superpixel number; S14, performing morphological dilation processing on a superpixel boundary region to generate an edge mask; S15, excluding an interference region caused by registration errors between the panchromatic image PAN and the multispectral image MSU in combination with the edge mask.

4. The method of anti-registration error panchromatic sharpening of claim 1, wherein, The calculation formula of the weighted average of R / G / B band spectral information in the S2 is as follows: ； wherein, represents the R / G / B band spectral information weighted mean value; represents all pixels in the i-th superpixel centroid neighborhood; represents the pixel value ; represents the element k; represents the weight of the pixel value occupies, and the weight is calculated as follows: ； where σ is the adaptive Gaussian kernel width, is the pixel Euclidean distance to the centroid; denotes the natural exponential function.

5. The method of anti-registration error panchromatic sharpening of claim 1, wherein, The radiation scaling factor in S43 The calculation formula of the optimal value of the offset b is respectively: ； ； wherein, covariance of the luminance channel I and the luminance channel Y; denotes the variance; , denotes the mean of the luminance channel I and the luminance channel Y, respectively; S44 in the luminance component The calculation formula is: 。 6. The method of anti-registration error panchromatic sharpening of claim 5, wherein, The S5 specifically comprises the following steps: S51, performing color encoding on the weighted average of R / G / B band spectral information in a YUV color space to generate initial chrominance components U and V of the fused image; S52, with the luminance component of the fused image in S4 For reference, perform chroma initialization padding for the non-boundary regions in conjunction with the edge mask in S1 and the initial chroma components U and V of the fused image. S53, after the chrominance initialization filling, establishing a spectral diffusion model based on local radiometric similarity for a heterogeneous boundary region defined by the edge mask; S54, realizing the chroma component by iteratively solving the established spectral diffusion model based on local radiation similarity In the spectral consistency propagation of the heterogeneous boundary region, an optimized YUV image is obtained. S55, inversely transforming the optimized YUV image to an RGB space to obtain a final fused image.

7. The method of anti-registration error panchromatic sharpening of claim 6, wherein, The spectral diffusion model in the S53 is specifically as follows: ； wherein, represents a loss function; represents the U value of the r pixel, the U value being a chroma value; represents the U value of the s pixel; represents the neighborhood of the r pixel; represents the pixel s; represents an affinity weight, and the affinity weight is obtained based on the square difference of the brightness between two pixels, and is expressed as: ； wherein, represents the luminance value of the r-pixel; a represents a positive correlation; e represents the natural constant; represents the luminance value of the s-pixel; is the variance of the luminance values within the local window centered on the pixel r.

8. A panchromatic sharpening system that is resistant to registration errors, comprising a computer device, characterized in that, The computer device is programmed or configured to perform the panchromatic sharpening method in any one of claims 1 to 7.

Images