A remote sensing image non-uniformity correction method based on moment matching method

Abstract

The application discloses a remote sensing image non-uniformity correction method based on a moment matching method, comprising: dividing a to-be-corrected remote sensing image from which obvious column strips are removed into blocks; for each image block: detecting strip positions in the image block; for each strip position: calculating a column standard deviation mean value of the image block, and setting minimum, initial and maximum detection windows according to the column standard deviation mean value; calculating maximum and minimum column mean variance thresholds according to the minimum, initial and maximum detection windows; calculating a column mean variance with the strip position as a center and a window width as the initial detection window, and updating the initial detection window according to the column mean variance and the maximum and minimum column mean variance thresholds; calculating a Gaussian weight, a column mean value and a column standard deviation of each column in the updated initial detection window, calculating a reference column mean value and a reference column standard deviation according to the Gaussian weight, the column mean value and the column standard deviation, and calculating a correction gain and a calibration offset according to the reference column mean value and the reference column standard deviation to correct the strip position. The application improves the correction effect.

Description

Technical Field

[0001] This invention belongs to the field of image denoising technology, specifically relating to a method for correcting non-uniformity in remote sensing images based on moment matching. Background Technology

[0002] Remote sensing cameras are widely used in military early warning, agricultural monitoring, and marine monitoring, and their imaging quality significantly impacts the depth and breadth of their applications. In remote sensing imaging systems, CCD (Charge-coupled Device) / CMOS (Complementary Metal Oxide Semiconductor) sensors experience inconsistent responses to the imaging target due to factors such as differences in detector pixel size, the influence of dark current in the circuitry, inhomogeneities in the optical coating on the sensor surface, and the effects of external environmental and temperature changes on the sensor's photoelectric system. This inconsistency results in non-uniform noise. Optical pushbroom remote sensing cameras use a focal plane composed of multiple sensor arrays stitched together, performing pushbroom imaging along a specific direction. Due to inconsistencies in the photoelectric responses between different detector units, non-uniform noise in the imaging results is mainly distributed in a stripe shape along the pushbroom direction in the image. This stripe noise degrades image quality and severely damages image information.

[0003] Currently, non-uniformity correction algorithms for strip noise fall into two main categories: those based on radiometric calibration and those based on image data. Radiometric calibration-based methods are simple in principle and easy to implement, but their applicability decreases as the calibration parameters change with the detector's operating environment, requiring equipment pause and recalibration, which is inconvenient. Image data-based methods can be further divided into non-uniformity correction based on frequency domain filtering, non-uniformity correction based on total variation algorithms, and non-uniformity correction based on statistical matching, among others. Non-uniform correction based on frequency domain filtering and non-uniform correction based on total variation algorithms suffer from problems such as severe loss of non-noise information, large computational load, complex process, and inability to meet real-time processing requirements. Therefore, non-uniform correction based on statistical matching is more widely used. For example, Gadallah et al. proposed the moment matching algorithm to remove strip noise. This algorithm has a good denoising effect in relatively flat areas, but it performs poorly in images with rich details and many types of objects. The denoised image has obvious banding effects. The basic idea of ​​other correction methods based on standard moment matching is to use the average of all column means and column standard deviations of the entire image as reference values ​​and substitute them into the moment matching formula for non-uniform correction. After correction, the column means of the image are too smooth, resulting in a significant loss of image detail information. Therefore, ZWYang et al. proposed a motion window moment matching method, which combines the idea of ​​local filtering and uses the column mean and standard deviation of the local window image to replace the column mean and standard deviation of the entire image, thus suppressing the striping effect of the moment matching algorithm. WWChang et al. proposed an algorithm based on wavelet transform and moment matching. This algorithm first calculates the average value of each column of pixels, then uses wavelet transform to denoise the average pixel value of each column. The denoised value is the true average value of each column. Finally, the original mean and the denoised mean are moment matched. J.Wang et al. obtained the corrected output image based on the idea of ​​moment matching and the relationship between the mean and vertical gradient of each column image and the reference value. Then, they combined the idea of ​​histogram matching to further correct the stripe non-uniformity of the output image in the previous step.

[0004] However, the method proposed by ZWYang et al. requires setting different moving window widths for remote sensing images of varying complexity to achieve the best correction effect, resulting in poor versatility. The method proposed by WWChang et al. indiscriminately applies moment matching correction to all stripe noise, ignoring certain stripe noise with large information deviations, leading to noise residue and artifacts. The method proposed by J.Wang et al. uses a total variational model to calculate the standard deviation of the Gaussian function during histogram matching, resulting in high overall algorithm complexity and computational burden, which is not conducive to real-time hardware implementation. Therefore, there is an urgent need for an efficient, universal method suitable for real-time processing of remote sensing image non-uniformity correction. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for correcting non-uniformity in remote sensing images based on moment matching. The technical problem to be solved by this invention is achieved through the following technical solution:

[0006] This invention provides a method for correcting non-uniformity in remote sensing images based on moment matching, the method comprising:

[0007] Detect obvious column bands in the remote sensing image to be corrected, and remove all detected obvious column bands using the interpolation method of the gradient minimum direction;

[0008] The remote sensing image to be corrected, after removing obvious column bands, is divided into several image blocks;

[0009] For each image patch, the process includes: calculating the normalized column mean gradient of each column in the image patch; detecting the strip positions in the image patch based on the normalized column mean gradients of adjacent columns; and for each detected strip position, the process includes:

[0010] Calculate the mean of the column standard deviations of the image patch, and set the minimum detection window, the initial detection window, and the maximum detection window based on the mean of the column standard deviations; wherein the minimum detection window is smaller than the initial detection window, and the initial detection window is smaller than the maximum detection window;

[0011] Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window;

[0012] Calculate the column mean variance centered on the strip position and with the window width as the initial detection window, and update the initial detection window based on the column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold;

[0013] Calculate the Gaussian weight, column mean, and column standard deviation for each column within the updated initial detection window; and calculate the reference column mean and reference column standard deviation based on the Gaussian weight, column mean, and column standard deviation.

[0014] The correction gain and calibration offset are calculated based on the mean and standard deviation of the reference column, and the strip position is corrected based on the correction gain and calibration offset.

[0015] In one embodiment of the present invention, detecting prominent column bands in the remote sensing image to be corrected and removing all detected prominent column bands includes:

[0016] Using the Pearson correlation coefficient of adjacent columns in the remote sensing image to be corrected, all obvious column bands in the remote sensing image to be corrected are detected;

[0017] For each prominent column band, the following steps are taken: the prominent column band is corrected using an interpolation method based on the gradient minimum direction.

[0018] In one embodiment of the present invention, calculating the normalized column mean gradient of each column in the image patch includes:

[0019] The formula for calculating the mean of the column standard deviations of each column in the image patch is as follows:

[0020]

[0021] Where, μ j DN represents the column mean of the j-th column in the image patch, M represents the number of rows in the image patch, and DN represents the column mean of the j-th column in the image patch. (i,j) This represents the grayscale value at pixel position (i,j);

[0022] The normalized column mean gradient of each column in the image patch is calculated using the following formula:

[0023]

[0024] Where, Δμ j G_μ represents the normalized column mean gradient of the j-th column in the image patch. j =|μ j -μ j-1 |,μ j-1 This represents the column mean of the (j-1)th column in the image patch. N represents the number of columns in the image patch.

[0025] In one embodiment of the present invention, the mean of the column standard deviations of the image patch is calculated, and a minimum detection window, an initial detection window, and a maximum detection window are set based on the mean of the column standard deviations, including:

[0026] The mean of the column standard deviations of this image patch is calculated using the following formula:

[0027]

[0028] in, σ represents the mean of the column standard deviations of an image patch. j This represents the column standard deviation of the j-th column in the image patch.

[0029] The widths of the minimum detection window, initial detection window, and maximum detection window, set based on the mean of the standard deviations of the columns, are expressed by the following formula:

[0030]

[0031] Among them, W min W represents the width of the minimum detection window, and W0 represents the width of the initial detection window.max This indicates the width of the maximum detection window.

[0032] In one embodiment of the present invention, the formula for calculating the column mean and variance with the strip position as the center and the window width as the initial detection window is expressed as:

[0033]

[0034] Where V represents the mean and variance of the columns in the initial detection window, centered at the strip position and with the window width as the initial width, and n represents the total number of columns in the initial detection window. This represents the column mean of column j1 in the initial detection window. This represents the average of the values ​​of all columns in the initial detection window.

[0035] In one embodiment of the present invention, calculating the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window includes:

[0036] Using the minimum detection window, the initial detection window, and the maximum detection window as moving windows respectively, the corresponding column mean and variance sets are calculated by traversing the entire image block.

[0037] Select the maximum value from the set of column mean variances corresponding to the minimum detection window, select the maximum and minimum values ​​from the set of column mean variances corresponding to the initial detection window, and select the minimum value from the set of column mean variances corresponding to the maximum detection window;

[0038] Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values.

[0039] In one embodiment of the present invention, calculating the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values ​​includes:

[0040] The formula for calculating the threshold of maximum column mean variance is as follows:

[0041]

[0042] Among them, V max This represents the threshold for the maximum column mean variance. This represents the maximum value selected from the set of column mean and variance corresponding to the minimum detection window. This indicates the maximum value selected from the set of column mean and variance corresponding to the initial detection window;

[0043] The formula for calculating the minimum column mean variance threshold is as follows:

[0044]

[0045] Among them, V min This represents the minimum column mean and variance threshold. This represents the minimum value selected from the set of column mean and variance corresponding to the maximum detection window. This represents the minimum value selected from the set of column mean variances corresponding to the initial detection window.

[0046] In one embodiment of the present invention, updating the initial detection window based on the column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold includes:

[0047] The column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold are compared. If the column mean variance is greater than the maximum column mean variance threshold, the initial detection window is reduced. If the column mean variance is less than the minimum column mean variance threshold, the initial detection window is expanded. If the column mean variance is between the maximum column mean variance threshold and the minimum column mean variance threshold, the initial detection window remains unchanged.

[0048] In one embodiment of the present invention, calculating the reference column mean and reference column standard deviation based on the Gaussian weights, the column mean, and the column standard deviation includes:

[0049] The formula for calculating the mean of the reference column is as follows:

[0050]

[0051] Where, μ τ W represents the mean of the reference column. j' This indicates the width of the updated initial detection window, a and b represent the starting and ending column numbers of the updated initial detection window, respectively, and j2 represents the j2-th column in the updated initial detection window. This represents the column mean of column j2 in the updated initial detection window. This represents the Gaussian weight of column j2 in the updated initial detection window. j' indicates the column where the stripe is located;

[0052] The standard deviation of the reference column is calculated using the following formula:

[0053]

[0054] Where, σ τ Indicates the standard deviation of the reference column. This represents the column standard deviation of column j2 in the updated initial detection window.

[0055] In one embodiment of the present invention, calculating the correction gain and calibration offset based on the mean of the reference column and the standard deviation of the reference column includes:

[0056] The correction gain is calculated using the following formula:

[0057]

[0058] Among them, a j' σ represents the correction gain at column j' where the stripe is located. τ σ represents the standard deviation of the reference column. j' This represents the column standard deviation at column j' where the band is located;

[0059]

[0060] Among them, b j' This represents the calibration offset at column j' where the stripe is located, μ τ μ represents the mean of the reference column. j' This represents the column mean at column j' where the band is located.

[0061] The beneficial effects of this invention are:

[0062] The remote sensing image non-uniformity correction method proposed in this invention includes: detecting obvious column bands in the remote sensing image to be corrected, and removing all detected obvious column bands using an interpolation method with the minimum gradient direction; dividing the remote sensing image to be corrected after removing obvious column bands into several image blocks; for each image block, the method includes: calculating the normalized column mean gradient of each column in the image block; detecting the position of the bands in the image block based on the normalized column mean gradients of adjacent columns; for each detected band position, the method includes: calculating the mean of the column standard deviation of the image block, and setting a minimum detection window, an initial detection window, and a maximum detection window based on the mean of the column standard deviation; wherein, the minimum detection window is smaller than the initial detection window. The initial detection window is smaller than the maximum detection window. The maximum and minimum column mean variance thresholds are calculated based on the minimum, initial, and maximum detection windows. The column mean variance is calculated with the strip position as the center and the window width as the initial detection window. The initial detection window is updated based on the column mean variance, the maximum and minimum column mean variance thresholds. The Gaussian weight, column mean, and column standard deviation of each column within the updated initial detection window are calculated. The reference column mean and reference column standard deviation are calculated based on these values. The correction gain and calibration offset are calculated based on the reference column mean and reference column standard deviation, and the strip position is corrected based on the correction gain and calibration offset. Therefore, this invention first detects and corrects obvious column stripes in the remote sensing image to be corrected, then divides the corrected remote sensing image into blocks, and then detects and corrects strip noise in each image block separately. Linear interpolation correction is performed on obvious stripes based on the direction of minimum gradient, avoiding the problem of the moment matching algorithm not being ideal for correcting obvious stripes. The remote sensing image to be corrected is first processed into blocks, and then the strip position in each image block is corrected. This preserves the information of non-strip column regions of the image, improves the quality of the corrected image, increases the hardware processing speed and reduces resource consumption. Moreover, the gray-level changes within the image block are relatively small compared to the entire image to be corrected, the correction reference value is more accurate, and the correction effect is improved.

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating a remote sensing image non-uniformity correction method based on moment matching provided in an embodiment of the present invention.

[0065] Figure 2 This is a schematic diagram of the gradient direction at a certain location of a strip provided in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram of the image after segmentation provided in an embodiment of the present invention;

[0067] Figure 4 (a)~ Figure 4 (c) is a schematic diagram of the remote sensing image to be corrected and the process of correcting the remote sensing image to be corrected, provided in an embodiment of the present invention. Detailed Implementation

[0068] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0069] Remote sensing images often exhibit stripe noise, which severely impacts image quality and causes information bias. For a better solution to this problem, please refer to [link / reference needed]. Figure 1 This invention provides a method for correcting non-uniformity in remote sensing images based on moment matching, specifically including the following steps:

[0070] S10. Detect obvious column bands in the remote sensing image to be corrected, and remove all detected obvious column bands using the interpolation method of the gradient minimum direction.

[0071] Because some columns in the remote sensing image to be corrected have grayscale values ​​that differ significantly from adjacent normal columns, they can hardly reflect the information of the detected pixels. It is inappropriate to calculate the bias and gain of these detected pixels using the moment matching method. Therefore, this embodiment of the invention first detects and corrects these obvious column bands. Specifically:

[0072] Embodiments of the present invention detect obvious column bands in a remote sensing image to be corrected and remove all detected obvious column bands, including:

[0073] The Pearson correlation coefficient between adjacent columns in the remote sensing image to be corrected is used to detect all prominent column bands in the image. The formula for detecting prominent column bands using the Pearson correlation coefficient is expressed as:

[0074]

[0075] Where X and Y represent the pixel values ​​of two adjacent columns in the remote sensing image to be corrected, Cov(X,Y) represents the covariance of X and Y, Cov(X,Y)=E(XY)-E(X)·E(Y), E(X), E(Y), and E(XY) represent the means of X, Y, and XY, respectively, and σ X σ Y Let X and Y represent the standard deviations, respectively. The Pearson correlation coefficient r(j) between each column and its left adjacent column is calculated, and the average of all Pearson correlation coefficients is taken. If satisfied and Then the j-th column is considered to be a clear strip column.

[0076] Furthermore, for each prominent column band, the prominent column band is corrected using an interpolation method based on the gradient minimum direction.

[0077] like Figure 2 As shown, column b represents the prominent stripe, and columns a and c are the nearest neighbor normal columns. A1, A2, A3, B, C1, C2, and C3 are pixel values. When performing interpolation correction on pixel B in the prominent stripe, the values ​​are first calculated... Figure 2 The gradient values ​​in the three directions are expressed by the formula:

[0078]

[0079] Where G1 represents the gradient value in direction 1, G2 represents the gradient value in direction 2, and G3 represents the gradient value in direction 3. The direction with the smallest gradient value is selected as the correction direction for pixel B using formula (2), and different weights are assigned to linear interpolation considering the distance from neighboring normal pixels to the pixel to be corrected. If direction 2 is the direction with the smallest gradient value, then the correction value formula for pixel B is expressed as:

[0080] B=ω'·A2+ω"·C2 (3)

[0081] Where ω' and ω" represent interpolation weights, the magnitude of which is related to the distance from the normal column pixels to the strip column pixels; the larger the distance, the smaller the weight. The calculation formula is expressed as:

[0082]

[0083] Where L1 represents the distance between the obvious column band and the nearest normal column on the left, and L2 represents the distance between the obvious column band and the nearest normal column on the right.

[0084] Unlike the traditional indiscriminate correction of each column, the embodiments of the present invention pre-detect and correct obvious column bands, that is, the gray values ​​of the banded column show obvious bright or dark bands compared with the surrounding neighboring columns, thus avoiding the problem of band residue and "artifacts" still existing in the image after moment matching correction.

[0085] S20. The remote sensing image to be corrected, after removing obvious column bands, is divided into blocks to obtain several image blocks.

[0086] Because remote sensing images are large in area and have complex ground features, they contain both flat areas with small grayscale variations and detailed areas with large grayscale variations. Furthermore, the image changes along the column direction as the detected pixels are scanned. Therefore, the correction bias and correction gain values ​​should differ within the same column. Dividing the image into multiple sub-blocks and using local image information to obtain the correction bias and correction gain values ​​instead of global image information can increase correction accuracy. On the other hand, block processing reduces hardware resource consumption and latency. The number of blocks is determined based on the target scene and image resolution. Figure 3 As shown, each image block after splitting has a size of M rows and N columns. If the target scene is rich, the grayscale values ​​vary greatly, and the image resolution is high, then the number of blocks will be large. At the same time, it is necessary to ensure that the resolution of the sub-images after segmentation is not too small, to avoid insufficient information in the sub-images, which would lead to a decrease in the accuracy of the correction bias and correction gain values.

[0087] Furthermore, for each image patch, including:

[0088] S30. Calculate the normalized column mean gradient of each column in the image patch.

[0089] For images containing stripe noise, the mean grayscale value of the stripe columns differs significantly from that of their adjacent columns. Therefore, the column mean gradient can be used to reflect the difference in mean grayscale values ​​between adjacent columns. However, for images with rich detail, the column mean gradient is generally large, while for smoother images, it is generally small. Therefore, a normalized column mean gradient can be used to detect stripes, improving detection accuracy. Specifically:

[0090] This invention embodiment calculates the normalized column mean gradient of each column in the image patch, including:

[0091] First, calculate the column mean of each column in the image patch, expressed by the formula:

[0092]

[0093] Where, μ j DN represents the column mean of the j-th column in the image patch, M represents the number of rows in the image patch, and DN represents the column mean of the j-th column in the image patch. (i,j) This represents the grayscale value at pixel position (i,j);

[0094] Next, the normalized column mean gradient of each column in the image patch is calculated, as expressed by the formula:

[0095]

[0096] Where, Δμ j G_μ represents the normalized column mean gradient of the j-th column in the image patch. j =|μ j -μj-1 |,μ j-1 This represents the mean of the standard deviations of the (j-1)th column in the image patch. N represents the number of columns in the image patch.

[0097] S40. Detect the position of the strip in the image block based on the gradient of the normalized column mean of adjacent columns.

[0098] For strip position detection in an image patch, determine if Δμ is satisfied. j >Th and Δμ j+1 If Th is defined, then the j-th column is determined to be a stripe column, and the stripe position is recorded. There may be one or more stripe positions within an image patch. Here, Th represents the gradient threshold.

[0099] Furthermore, for each detected band location, including:

[0100] Since fixed window detection has poor noise reduction performance, this invention proposes an adaptive window detection scheme, which specifically includes the following steps S50 to S70:

[0101] S50. Calculate the mean of the column standard deviations of the image patch, and set the minimum detection window, the initial detection window, and the maximum detection window based on the mean of the column standard deviations; wherein, the minimum detection window is smaller than the initial detection window, and the initial detection window is smaller than the maximum detection window.

[0102] In this embodiment of the invention, the range of the detection window width is determined based on the mean of the column standard deviations of an image patch. Therefore, the mean of the column standard deviations of the image patch is first calculated, and the minimum detection window, initial detection window, and maximum detection window are set based on the mean of the column standard deviations, including:

[0103] The mean of the column standard deviations of this image patch is calculated using the following formula:

[0104]

[0105] in, σ represents the mean of the column standard deviations of an image patch. j This represents the column standard deviation of the j-th column in the image patch. When determining the width of the adaptive window based on the mean of the column standard deviations for different remote sensing images, A larger value indicates more image information, and the smaller the detection window should be; conversely, a smaller value indicates more information. The smaller the value, the less image information there is, and the larger the detection window should be. Therefore, in this embodiment of the invention, the minimum, initial, and maximum values ​​of the detection window width are set according to an inverse proportional function. Here, the ratio is set to 1:2:4, and the corresponding ratio factors are set to 50, 100, and 200. The widths of the minimum, initial, and maximum detection windows, set according to the mean of the column standard deviations, are expressed by the following formula:

[0106]

[0107] Among them, W min W represents the width of the minimum detection window, and W0 represents the width of the initial detection window. max This indicates the width of the maximum detection window.

[0108] S60. Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window.

[0109] This invention designs the width of an adaptive window based on the column mean variance within the minimum detection window, the initial detection window, and the maximum detection window. Specifically, this invention calculates the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window, including:

[0110] S601. Using the smallest detection window, the initial detection window, and the largest detection window as moving windows respectively, traverse the entire image block and calculate the corresponding column mean and variance set.

[0111] In S601, the method for calculating the column mean variance during the traversal of the entire image block is as follows: Formula (11). The minimum detection window, the initial detection window, and the maximum detection window are moved to traverse the entire image block respectively to calculate the corresponding column mean variance set.

[0112] S602. Select the maximum value from the set of column mean and variance corresponding to the smallest detection window, denoted as . Select the maximum and minimum values ​​from the set of column means and variances corresponding to the initial detection window, denoted as and respectively. Select the minimum value from the set of column mean and variance corresponding to the largest detection window, denoted as .

[0113] S603. Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values.

[0114] This embodiment of the invention calculates the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values, including:

[0115] The formula for calculating the threshold of maximum column mean variance is as follows:

[0116]

[0117] Among them, V max This represents the threshold for the maximum column mean variance. This indicates the maximum value selected from the set of column mean and variance corresponding to the smallest detection window. This indicates the maximum value selected from the set of column mean and variance corresponding to the initial detection window;

[0118] The formula for calculating the minimum column mean variance threshold is as follows:

[0119]

[0120] Among them, V min This represents the minimum column mean and variance threshold. This represents the minimum value selected from the set of column mean and variance corresponding to the maximum detection window. This represents the minimum value selected from the set of column mean variances corresponding to the initial detection window.

[0121] S70. Calculate the column mean variance with the strip position as the center and the window width as the initial detection window, and update the initial detection window based on the column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold.

[0122] The formula for calculating the column mean and variance of the detection window centered on the strip position and with the window width as the initial window width in this embodiment of the invention is expressed as follows:

[0123]

[0124] Where V represents the mean and variance of the columns in the initial detection window, centered at the strip position and with the window width as the initial width, and n represents the total number of columns in the initial detection window. This represents the column mean of column j1 in the initial detection window. This represents the average of the values ​​of all columns in the initial detection window.

[0125] This embodiment of the invention updates the initial detection window based on the column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold, including:

[0126] Compare column mean and variance V, and the threshold V for maximum column mean and variance. max and minimum column mean variance threshold V min If the column mean variance V is greater than the maximum column mean variance threshold V max Then, the initial detection window is narrowed. If the column mean variance V is less than the minimum column mean variance threshold V... min Then expand the initial detection window. If the column mean variance V is within the maximum column mean variance threshold V...max and minimum column mean variance threshold V min During this period, the initial detection window remains unchanged. Finally, the detection window W, centered on the stripe location, is determined. j' This refers to the updated initial detection window.

[0127] S80. Calculate the Gaussian weight, column mean, and column standard deviation of each column in the updated initial detection window, and calculate the reference column mean and reference column standard deviation based on the Gaussian weight, column mean, and column standard deviation.

[0128] In this embodiment of the invention, after determining the width of the adaptive window (i.e., the updated initial detection window), considering the spatial correlation between the neighboring columns and the column j' where the stripe is located, weights are assigned to each column based on the distances of other columns to the column j' where the stripe is located. Gaussian weights are used for these weights, and the formula is as follows:

[0129]

[0130] in, This represents the Gaussian weight of column j2 in the updated initial detection window, where j2 represents column j2 in the updated initial detection window, j' represents the column where the band is located, and W... j' This indicates the width of the updated initial detection window.

[0131] In this embodiment of the invention, the reference column mean and reference column standard deviation are calculated using a weighted average in the moment matching algorithm. Specifically, this embodiment calculates the reference column mean and reference column standard deviation based on Gaussian weights, column mean, and column standard deviation, including:

[0132] The formula for calculating the mean of the reference column is as follows:

[0133]

[0134] Where, μ τ This represents the mean of the reference column, where a and b represent the starting and ending column numbers of the updated initial detection window, respectively. This represents the column mean of column j2 in the updated initial detection window.

[0135] The standard deviation of the reference column is calculated using the following formula:

[0136]

[0137] Where, σ τ Indicates the standard deviation of the reference column. This represents the column standard deviation of column j2 in the updated initial detection window.

[0138] S90. Calculate the correction gain and calibration offset based on the reference column mean and reference column standard deviation, and correct the strip position based on the correction gain and calibration offset.

[0139] This invention embodiment calculates the correction gain and calibration offset based on the reference column mean and reference column standard deviation, including:

[0140] The correction gain is calculated using the following formula:

[0141]

[0142] Among them, a j' σ represents the correction gain at column j' where the stripe is located. τ σ represents the standard deviation of the reference column. j' This represents the column standard deviation at column j' where the band is located;

[0143]

[0144] Among them, b j' This represents the calibration offset at column j' where the stripe is located, μ τ μ represents the mean of the reference column. j' This represents the column mean at column j' where the strip position is located. For instructions on how to correct the strip position using correction gain and calibration offset, please refer to existing techniques.

[0145] In this embodiment of the invention, each image block after the remote sensing image to be corrected is divided into blocks, and the above-described S30 to S90 processes are performed on each strip position detected in the image block, so as to achieve non-uniform correction of the entire remote sensing image to be corrected.

[0146] This invention divides the remote sensing image to be corrected after obvious stripe correction into multiple image blocks, detects and corrects the stripe positions in each image block, avoiding the loss of information in non-striped areas, improving the correction effect while reducing hardware resource consumption. During stripe position correction, based on local image spatial correlation, the width of the detection window is adaptively adjusted with the stripe position as the center. Gaussian weight values ​​are designed using the distance from non-striped columns (normal columns) to striped columns (the columns containing the stripe positions) within the adaptive detection window. The mean and standard deviation of the reference columns for moment matching are obtained through these designed Gaussian weight values. The stripe positions are then corrected using the mean and standard deviation of the reference columns, increasing the correction accuracy and effect.

[0147] It should be noted that the method proposed in this invention can also be applied to other image information processing, and is not limited to remote sensing images.

[0148] Please see Figure 4 (a)~ Figure 4(c), Figure 4 (a) is a remote sensing image to be corrected provided in an embodiment of the present invention. The resolution of the remote sensing image to be corrected is 2000*2000. Figure 4 (b) is the result of processing the remote sensing image to be corrected into blocks, such as... Figure 4 The image shown in (a) is divided into 16 4x4 image blocks, meaning each image block has a resolution of 500x500. Figure 4 (c) shows the corrected result. It can be seen that the method proposed in this embodiment of the invention effectively eliminates the original column stripes and no "artifact" phenomenon appears after correction.

[0149] In summary, the remote sensing image non-uniformity correction method based on moment matching proposed in this invention includes: detecting obvious column bands in the remote sensing image to be corrected, and removing all detected obvious column bands using an interpolation method with the minimum gradient direction; dividing the remote sensing image to be corrected after removing obvious column bands into blocks to obtain several image blocks; for each image block, the method includes: calculating the normalized column mean gradient of each column in the image block; detecting the position of the bands in the image block based on the normalized column mean gradients of adjacent columns; for each detected band position, the method includes: calculating the mean of the column standard deviation of the image block, and setting a minimum detection window, an initial detection window, and a maximum detection window based on the mean of the column standard deviation; wherein, the minimum detection window is smaller than... An initial detection window is established, which is smaller than the maximum detection window. The maximum and minimum column mean variance thresholds are calculated based on the minimum, initial, and maximum detection windows. The column mean variance is calculated with the strip position as the center and the window width as the initial detection window. The initial detection window is updated based on the column mean variance, the maximum and minimum column mean variance thresholds. The Gaussian weight, column mean, and column standard deviation of each column within the updated initial detection window are calculated. The reference column mean and reference column standard deviation are calculated based on these values. The correction gain and calibration offset are calculated based on the reference column mean and reference column standard deviation, and the strip position is corrected based on the correction gain and calibration offset. Therefore, this embodiment of the invention first detects and corrects obvious column stripes in the remote sensing image to be corrected, then divides the corrected remote sensing image into blocks, and then detects and corrects strip noise in each image block separately. For obvious stripes, linear interpolation correction is performed based on the direction of minimum gradient, avoiding the problem of unsatisfactory correction of obvious stripes by the moment matching algorithm. The remote sensing image to be corrected is first divided into blocks, and then the strip position in each image block is corrected. This preserves the information of non-strip column regions of the image, improves the quality of the corrected image, increases hardware processing speed and reduces resource consumption. Moreover, the gray-level changes within the image block are relatively small compared to the entire image to be corrected, and the correction reference value is more accurate, thus improving the correction effect. The method proposed in this invention has strong versatility and is applicable to the non-uniformity correction of remote sensing images in any scene, as well as the non-uniformity correction of other non-remote sensing images.

[0150] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0151] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the specification and accompanying drawings, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0152] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for correcting non-uniformity in remote sensing images based on moment matching, characterized in that, The method includes: Detect obvious column bands in the remote sensing image to be corrected, and remove all detected obvious column bands using the interpolation method of the gradient minimum direction; The remote sensing image to be corrected, after removing obvious column bands, is divided into several image blocks; For each image patch, the process includes: calculating the normalized column mean gradient of each column in the image patch; detecting the strip positions in the image patch based on the normalized column mean gradients of adjacent columns; and for each detected strip position, the process includes: Calculate the mean of the column standard deviations of the image patch, and set the minimum detection window, the initial detection window, and the maximum detection window based on the mean of the column standard deviations; wherein the minimum detection window is smaller than the initial detection window, and the initial detection window is smaller than the maximum detection window; Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window; Calculate the column mean variance centered on the strip position and with the window width as the initial detection window, and update the initial detection window based on the column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold; Calculate the Gaussian weight, column mean, and column standard deviation for each column within the updated initial detection window; and calculate the reference column mean and reference column standard deviation based on the Gaussian weight, column mean, and column standard deviation. The correction gain and calibration offset are calculated based on the mean and standard deviation of the reference column, and the strip position is corrected based on the correction gain and calibration offset.

2. The remote sensing image non-uniformity correction method based on moment matching according to claim 1, characterized in that, Detect prominent stripes in the remote sensing image to be corrected, and remove all detected prominent stripes, including: Using the Pearson correlation coefficient of adjacent columns in the remote sensing image to be corrected, all obvious column bands in the remote sensing image to be corrected are detected; For each prominent column band, the following steps are taken: the prominent column band is corrected using an interpolation method based on the gradient minimum direction.

3. The remote sensing image non-uniformity correction method based on moment matching according to claim 1, characterized in that, Calculating the normalized column mean gradient for each column in the image patch includes: The formula for calculating the mean of the column standard deviations of each column in the image patch is as follows: Where, μ j DN represents the column mean of the j-th column in the image patch, M represents the number of rows in the image patch, and DN represents the column mean of the j-th column in the image patch. (i,j) This represents the grayscale value at pixel position (i,j). The normalized column mean gradient of each column in the image patch is calculated using the following formula: Where, Δμ j G_μ represents the normalized column mean gradient of the j-th column in the image patch. j =|μ j -μ j-1 |,μ j-1 This represents the column mean of the (j-1)th column in the image patch. N represents the number of columns in the image patch.

4. The remote sensing image non-uniformity correction method based on moment matching according to claim 3, characterized in that, Calculate the mean of the column standard deviations of the image patch, and set the minimum detection window, initial detection window, and maximum detection window based on the mean of the column standard deviations, including: The mean of the column standard deviations of this image patch is calculated using the following formula: in, σ represents the mean of the column standard deviations of an image patch. j This represents the column standard deviation of the j-th column in the image patch. The widths of the minimum detection window, initial detection window, and maximum detection window, set based on the mean of the standard deviations of the columns, are expressed by the following formula: Among them, W min W represents the width of the minimum detection window, and W0 represents the width of the initial detection window. max This indicates the width of the maximum detection window.

5. The remote sensing image non-uniformity correction method based on moment matching according to claim 1, characterized in that, The formula for calculating the column mean and variance with the strip position as the center and the window width as the initial detection window is expressed as: Where V represents the mean and variance of the columns in the initial detection window, centered at the strip position and with the window width as the initial width, and n represents the total number of columns in the initial detection window. This represents the column mean of column j1 in the initial detection window. This represents the average of the values ​​of all columns in the initial detection window.

6. The method for correcting non-uniformity of remote sensing images based on moment matching according to claim 1, characterized in that, Calculating the maximum column mean variance threshold and the minimum column mean variance threshold based on the minimum detection window, the initial detection window, and the maximum detection window includes: Using the minimum detection window, the initial detection window, and the maximum detection window as moving windows respectively, the corresponding column mean and variance sets are calculated by traversing the entire image block. Select the maximum value from the set of column mean variances corresponding to the minimum detection window, select the maximum and minimum values ​​from the set of column mean variances corresponding to the initial detection window, and select the minimum value from the set of column mean variances corresponding to the maximum detection window; Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values.

7. The remote sensing image non-uniformity correction method based on moment matching according to claim 6, characterized in that, Calculate the maximum column mean variance threshold and the minimum column mean variance threshold based on all selected maximum and minimum values, including: The formula for calculating the threshold of maximum column mean variance is as follows: Among them, V max This represents the threshold for the maximum column mean and variance. This represents the maximum value selected from the set of column mean and variance corresponding to the minimum detection window. This indicates the maximum value selected from the set of column mean and variance corresponding to the initial detection window; The formula for calculating the minimum column mean variance threshold is as follows: Among them, V min This represents the minimum column mean and variance threshold. This represents the minimum value selected from the set of column mean and variance corresponding to the maximum detection window. This represents the minimum value selected from the set of column mean variances corresponding to the initial detection window.

8. The method for correcting non-uniformity of remote sensing images based on moment matching according to claim 1, characterized in that, The initial detection window is updated based on the column mean and variance, the maximum column mean and variance threshold, and the minimum column mean and variance threshold, including: The column mean variance, the maximum column mean variance threshold, and the minimum column mean variance threshold are compared. If the column mean variance is greater than the maximum column mean variance threshold, the initial detection window is reduced. If the column mean variance is less than the minimum column mean variance threshold, the initial detection window is expanded. If the column mean variance is between the maximum column mean variance threshold and the minimum column mean variance threshold, the initial detection window remains unchanged.

9. The method for correcting non-uniformity of remote sensing images based on moment matching according to claim 1, characterized in that, Calculating the reference column mean and reference column standard deviation based on the Gaussian weights, the column mean, and the column standard deviation includes: The formula for calculating the mean of the reference column is as follows: Where, μ τ W represents the mean of the reference column. j' This indicates the width of the updated initial detection window, a and b represent the starting and ending column numbers of the updated initial detection window, respectively, and j2 represents the j2-th column in the updated initial detection window. This represents the column mean of column j2 in the updated initial detection window. This represents the Gaussian weight of column j2 in the updated initial detection window. j' indicates the column where the stripe is located; The standard deviation of the reference column is calculated using the following formula: Where, σ τ Indicates the standard deviation of the reference column. This represents the column standard deviation of column j2 in the updated initial detection window.

10. The method for correcting non-uniformity of remote sensing images based on moment matching according to claim 9, characterized in that, The correction gain and calibration offset are calculated based on the mean and standard deviation of the reference column, including: The correction gain is calculated using the following formula: Among them, a j' σ represents the correction gain at column j' where the stripe is located. τ σ represents the standard deviation of the reference column. j' This represents the column standard deviation at column j' where the band is located; Among them, b j' This represents the calibration offset at column j' where the stripe is located, μ τ μ represents the mean of the reference column. j' This represents the column mean at column j' where the band is located.

Images