A method for correcting fringe nonuniformity based on spatiotemporal adaptive one-dimensional filtering

Abstract

This invention discloses a method for correcting stripe nonuniformity based on spatiotemporal adaptive one-dimensional filtering. It employs one-dimensional discrete wavelet transform to extract initial stripe nonuniformity, combining temporal mean filtering and spatial horizontal one-dimensional weighted guided filtering to construct a spatiotemporal adaptive one-dimensional filter. The filter parameters are adaptively adjusted in the spatial domain based on Median Absolute Deviation (MAD), and in the intensity domain based on two-dimensional discrete wavelet transform and one-dimensional edge-aware weighting. Based on an iterative correction strategy, inaccurate estimates are removed and error accumulation is suppressed by increasing the constraint threshold, enabling the filter to converge. This invention balances scene generalization ability, correction accuracy, and computational efficiency, possessing the ability to distinguish stripe nonuniformity and scene edges, achieving adaptive correction of stripe nonuniformity in infrared images regardless of scene changes and content.

Description

Technical Field

[0001] This invention pertains to infrared imaging non-uniformity correction technology, and particularly relates to a stripe non-uniformity correction method based on spatiotemporal adaptive one-dimensional filtering. Background Technology

[0002] Infrared imaging systems can be categorized into scanning and staring types based on their imaging method. Scanning systems use a strip-shaped one-dimensional infrared focal plane, acquiring infrared images row by row or column in a time-division manner, and then stitching these rows or columns together to form an image. They are commonly used for long-range satellite remote sensing imaging to acquire high-resolution images. Due to the discontinuity of the scanning method, the image exhibits stripe non-uniformity. Staring systems use a planar two-dimensional infrared focal plane, commonly used for short-range imaging. They output images through a shared amplifier and analog-to-digital converter for each row or column. Differences in electronic components can also lead to stripe non-uniformity in the image. Stripe non-uniformity is a type of noise superimposed on the real scene; because of its slow change, it is also known as fixed-pattern noise and severely degrades image quality.

[0003] To restore image quality, correction methods are divided into two categories: calibration-based correction and scene-based correction. The commonly used calibration correction method in engineering is single-point occlusion correction, which corrects the effects of drift non-uniformity through periodic occlusion with uniform occluders. However, the blind-viewing effect generated during occlusion makes it unsuitable for real-time applications. Scene-based correction methods mainly include statistical and registration-based methods. Statistical methods rely on sufficient scene motion, and slow-moving scenes are easily treated as non-uniformity errors, leading to negative ghosting. Registration-based correction methods rely on scene-specific global translational motion. Deep learning methods have been a hot topic in recent years, but limited by the availability of training data, the individual differences in non-uniformity of infrared imaging systems, and high computational costs, they are still far from practical engineering applications. There are also model optimization correction methods that utilize prior information about images and non-uniformity, solving them by constructing a cost function. These methods typically have high computational costs and forgo utilizing video stream information, limiting correction accuracy. In summary, it is difficult for the above methods to balance scene generalization ability, correction accuracy, and computational efficiency. Especially when the stripes are non-uniform and coincide with the scene edge, under-correction or over-correction of the stripes and blurry edges are likely to occur. Summary of the Invention

[0004] The purpose of this invention is to provide a stripe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering, which has the ability to distinguish stripe nonuniformity and scene edges, and can adaptively correct stripe nonuniformity of infrared images regardless of scene changes and content. Even when stripe nonuniformity coincides with scene edges, it can avoid stripe undercorrection or overcorrection and edge blurring.

[0005] The technical solution for implementing this invention is as follows: a method for correcting fringe non-uniformity based on spatiotemporal adaptive one-dimensional filtering, comprising the following steps:

[0006] Step 1: Acquire images using an infrared imaging system with an image resolution of m×n;

[0007] Step 2: For the first frame image z(1), calculate the spatial mean by column to obtain the column spatial mean vector of z(1) and normalize it. Use one-dimensional single-scale discrete wavelet transform to extract high-frequency detail components from the normalized vector, copy m rows and expand it into an initial fringe non-uniformity correction image of m×n.

[0008] Step 3: Continuously acquire images using an infrared imaging system, and statistically analyze the temporal mean image of the j-th frame. An iterative correction strategy is used to obtain an iteratively updated corrected image for stripe non-uniformity. and iteratively updated time-domain mean-corrected image Initialize stripe non-uniformity correction image Image update with residual stripe non-uniformity

[0009] Step 4: Adjust the temporal mean of the iteratively updated image. The intensity of overall stripe nonuniformity in the image is estimated using the Median Absolute Deviation (MAD) method, and the radius r of the filter is adaptively adjusted from the spatial domain.

[0010] Step 5: Adjust the temporal mean of the iteratively updated image. An improved horizontal one-dimensional edge-sensing weighting matrix Γ is used to enhance image processing. Mid-band non-uniformity and scene edge resolution. Image extraction using two-dimensional single-scale discrete wavelet transform. The vertical detail component matrix V is obtained, and the median of the absolute values ​​of each column of matrix V is taken as the estimated value of the non-uniformity intensity of each column of stripes. Combined with Γ, the regularization factor matrix λ′ of the filter is adaptively adjusted from the intensity domain.

[0011] Step 6: Correct the temporal mean of the iteratively updated image. Combining the horizontally weighted guided filter in the spatial domain, and substituting the adaptive filter radius r from step 4 and the regularization factor matrix λ′ from step 5, a spatiotemporal adaptive one-dimensional filter is constructed, which is then applied to the input image. The filtered image is obtained by subtracting the output image q of the spatiotemporal adaptive one-dimensional filter. Will This serves as an initial image of the residual stripe non-uniformity.

[0012] Step 7: Process the initially extracted residual stripe non-uniformity image. By increasing the constraint threshold TH, the inaccurate stripe non-uniformity estimates at the scene edges are set to 0, resulting in a constrained residual stripe non-uniformity image, denoted as... right The mean of the non-zero spatial domain is calculated column-wise to obtain a 1×n vector. After copying m rows, it is expanded into an m×n residual stripe non-uniformity update image.

[0013] Step 8: Calculate the corrected image of the j-th frame.

[0014] Compared with the prior art, the significant advantages of this invention are:

[0015] (1) The MAD method is used for the first time to estimate the overall stripe non-uniformity intensity of the current corrected image. It is not limited by scene content and stripe intensity. It can estimate the overall stripe non-uniformity intensity of the image and adjust the radius r of the spatial-temporal adaptive one-dimensional filter from the spatial domain adaptive adjustment.

[0016] (2) For the first time, the median of the absolute values ​​of each column in the vertical detail component image V was extracted using two-dimensional single-scale discrete wavelet transform as an estimate of the non-uniformity intensity of each column of stripes. Combined with the horizontal one-dimensional edge-sensing weighting matrix Γ, the global, local and each column were associated, and the regularization factor matrix λ′ of the spatiotemporal adaptive one-dimensional filter was adaptively adjusted from the intensity domain.

[0017] (3) For the first time, the constraint threshold TH is added to set the inaccurate stripe non-uniformity estimate at the scene edge to 0, suppress the error accumulation caused by iterative correction, enable the spatiotemporal adaptive one-dimensional filter to have convergence ability, and enhance the robustness and correction accuracy of the method. Attached Figure Description

[0018] Figure 1 This is a flowchart of the fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to the present invention.

[0019] Figure 2 This is a model diagram of the fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to the present invention.

[0020] Figure 3 The figure shows the subjective comparison results of five stripe non-uniformity correction methods with the method proposed in this invention. The two images in the same column are grouped together, and from left to right, they are the non-uniformity degradation image group, SSS image group, DDBP image group, TVBP image group, WDGF image group, WGFMF image group, and Proposed image group.

[0021] Figure 4The figures show the quantitative comparison of five stripe non-uniformity correction methods with the method proposed in this invention. (a) is the PSNR figure; (b) is the SSIM figure. Detailed Implementation

[0022] 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 a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] The technical solutions of the various embodiments of the present invention can be combined with each other, but only if they can be implemented by those skilled in the art. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0024] The following section will further introduce the specific implementation method, as well as the technical difficulties and inventive points of this invention, using examples from this design.

[0025] Combination Figure 1 and Figure 2 A method for correcting fringe nonuniformity based on spatiotemporal adaptive one-dimensional filtering includes the following steps:

[0026] Step 1: Acquire images using an infrared imaging system with an image resolution of m×n.

[0027] Step 2: For the first frame image z(1), calculate the spatial mean by column to obtain the column spatial mean vector of z(1) and normalize it. Use one-dimensional single-scale discrete wavelet transform to extract high-frequency detail components from the normalized vector, copy m rows and expand it into an initial fringe non-uniformity correction image of m×n. Specifically as follows:

[0028] For the first frame image z(1), calculate the spatial mean by column to obtain the column spatial mean vector of the first frame image z(1). This represents a vector, where the subscript y represents the column number. This represents the operation of spatial mean.

[0029] The spatial mean of the y-th column of the first frame image z(1) Represented as:

[0030]

[0031] (x,y) represents the pixel position.

[0032] right The normalized vector is denoted as

[0033] The normalized column spatial mean vector of the y-th column Represented as:

[0034]

[0035] right One-dimensional single-scale discrete wavelet transform is used to extract high-frequency detail components, and the images are copied m rows and expanded to form an initial m×n fringe non-uniformity corrected image. Specifically as follows:

[0036] For vectors High-frequency detail component vectors are extracted using one-dimensional single-scale discrete wavelet transform.

[0037]

[0038] ^ represents an estimated value, DWT 1D sym8 represents the one-dimensional single-scale discrete wavelet transform operation, and sym8 represents the Symlets8 wavelet basis. This represents the high-frequency detail component vector, primarily including fringe non-uniformity. To preserve the high-frequency detail component vector... The length is consistent with the image column width n, DWT 1D Do not downsampling.

[0039] High-frequency detail component vectors The initial stripe non-uniformity corrected image is expanded to m×n after copying m rows.

[0040]

[0041] The symbol `repmat` represents the vector copy operation. Combined with `[m 1]`, it represents copying m rows and 1 column, thus creating a 1×n high-frequency detail component vector. Expanded to an initial fringe non-uniformity correction image of m×n

[0042] Step 3: Continuously acquire images using an infrared imaging system, and statistically analyze the temporal mean image of the j-th frame. The fringe non-uniformity correction image is updated iteratively using an iterative correction strategy. and time-domain mean-corrected image Initialize stripe non-uniformity correction image Image update with residual stripe non-uniformity Specifically as follows:

[0043] Images are continuously acquired using an infrared imaging system, and the temporal mean of the j-th frame is statistically analyzed.

[0044]

[0045] This indicates the operation of the time-domain mean.

[0046] An iterative correction strategy is used to obtain an iteratively updated corrected image for stripe non-uniformity. and iteratively updated time-domain mean-corrected image

[0047]

[0048] Represents the stripe non-uniformity corrected image of frame j. The image representing the residual fringe nonuniformity update in frame (j-1) is used to initialize the fringe nonuniformity correction image. and residual stripe non-uniformity update image

[0049] Step 4: Adjust the temporal mean of the iteratively updated image. Estimation using the Median Absolute Deviation (MAD) method Overall stripe non-uniformity intensity σ s The radius r of the filter is adaptively adjusted in the spatial domain, as follows:

[0050] S4-1, Calculate the iteratively updated time-domain mean-corrected image. Average values ​​of the spatial domains in each column:

[0051]

[0052] in, Represents the time-domain mean-corrected image updated iteratively. The mean of the spatial domain in the y-th column is denoted as the vector formed by it.

[0053] S4-2, Calculate the image The forward difference of the means of two adjacent columns forms the forward difference vector.

[0054]

[0055] S4-3, Calculate the MAD value to exclude... Outliers in:

[0056]

[0057] med represents the median operation.

[0058] S4-4, Remove The vector after the outlier is denoted as

[0059]

[0060] S4-5, Estimating the iteratively updated time-domain mean-corrected image Overall stripe non-uniformity intensity σ s :

[0061]

[0062] std represents the standard deviation calculation.

[0063] S4-6, Based on the confidence interval of the normal distribution, adaptively adjust the radius r of the filter in the spatial domain:

[0064]

[0065] in, It represents rounding up.

[0066] Step 5: Adjust the temporal mean of the iteratively updated image. An improved horizontal one-dimensional edge-sensing weighting matrix Γ is used to enhance image processing. Mid-band non-uniformity and scene edge resolution. Image extraction using two-dimensional single-scale discrete wavelet transform. The vertical detail component matrix V is used, and the median of the absolute values ​​of each column of matrix V is taken as the estimated value of the non-uniformity intensity of each column of stripes. Combined with Γ, the regularization factor matrix λ′ of the filter is adaptively adjusted from the intensity domain, as follows:

[0067]

[0068] Where L is The dynamic range, where ε represents the sensitivity coefficient and K represents the proportional coefficient.

[0069] This indicates the time-domain mean-corrected image during iterative updates. The variance of a 1×3 local window centered at pixel p′. This represents the variance of a 1×3 local window centered at pixel p, where pixel p represents... All pixels. Γ(p′) represents the horizontal one-dimensional edge-aware weighting at p′. To correlate the local and global, Γ(p′) uses Variance normalization for all pixels within their respective 1×3 local windows.

[0070] This invention is the first to propose adding a proportional coefficient K to the sensitivity coefficient ε formula to adjust the resolution sensitivity for stripe non-uniformity and scene edges, typically setting K to 0.1; changing the two-dimensional edge perception weighting to one-dimensional edge perception weighting, making it focus on the resolution capability for vertical stripe non-uniformity and scene edges, further improving the resolution sensitivity.

[0071] Image extraction using two-dimensional single-scale discrete wavelet transform Vertical detail component matrix V:

[0072]

[0073] DWT 2D This represents a two-dimensional single-scale discrete wavelet transform, where haar represents the Haar wavelet basis, and V represents the detail component matrix in the vertical direction. To maintain the relationship between matrix V and the image... Consistent resolution, DWT 2D Do not downsampling.

[0074] Take the column vectors of matrix V The median absolute value is used as an estimate of the non-uniformity intensity of each stripe column:

[0075]

[0076] Let σ represent the y-th column vector of matrix V. y λ represents the standard deviation of the estimated non-uniformity intensity of the y-th column fringe. y The regularization factor of the y-th column is represented by the vector as follows: vector After copying m rows, expand to an m×n regularization factor matrix λ:

[0077]

[0078] Combining Γ, λ is improved into a regularization factor matrix λ′ of the intensity domain adaptively adjusting the dimensional filter, thus relating the global, local, and column-specific aspects:

[0079]

[0080] symbol It represents the Hadamaji.

[0081] Step 6: Correct the temporal mean of the iteratively updated image. Combining the horizontally weighted guided filter in the spatial domain, and substituting the adaptive filtering radius r from step 4 and the adaptive regularization factor matrix λ′ from step 5, a spatiotemporal adaptive one-dimensional filter is constructed, which is then applied to the input image. The filtered image is obtained by subtracting the output image q of the spatiotemporal adaptive one-dimensional filter. Will The preliminary extracted image of residual stripe non-uniformity is shown below:

[0082] The output image q of the spatiotemporal adaptive one-dimensional filter is as follows:

[0083]

[0084] Where w represents a horizontal window of 1×(2r+1), denoted by , represents the variance within the w-window centered at pixel p′, a(p′) is the linear multiplication coefficient of the filtering transformation of pixel p′, and b(p′) is the linear addition coefficient of the filtering transformation of pixel p′. It is the spatial mean of a(p′) within the w window centered at pixel p′. It is the spatial mean of b(p′) within the w window centered at pixel p′. This indicates that within the w window, centered at pixel p′... The spatial mean of the input image. q(p′) is the input image. The filter output at pixel p′ is used to iterate through all pixels of the window image using w to obtain the output image q.

[0085] By inputting an image The filtered image is obtained by subtracting the output image q of the spatiotemporal adaptive one-dimensional filter.

[0086]

[0087] Will This serves as an initial image of the residual stripe non-uniformity.

[0088] Step 7: Process the initially extracted residual stripe non-uniformity image. By increasing the constraint threshold TH, the inaccurate stripe non-uniformity estimates at the scene edges are set to 0, resulting in a constrained residual stripe non-uniformity image, denoted as... right The mean of the non-zero spatial domain is calculated column-wise to obtain a 1×n vector. After copying m rows, it is expanded into an m×n residual stripe non-uniformity update image. Specifically as follows:

[0089] Image of residual stripe non-uniformity extracted in the preliminary stage Due to the edge-preserving effect of the one-dimensional weighted guided filter, the fringe non-uniformity estimates at the scene edges are close to 0, resulting in inaccurate estimates. By increasing the constraint threshold TH, the inaccurate fringe non-uniformity estimates at the scene edges are set to 0, yielding the constrained residual fringe non-uniformity image, denoted as... It can suppress the accumulation of errors generated by iteration, enabling the spatiotemporal adaptive one-dimensional filter to have convergence capability.

[0090]

[0091] For 8-bit or 16-bit images, TH is typically set to 0.1.

[0092] right Calculating the mean of the non-zero spatial region column by column yields a 1×n vector. This represents the mean of the non-zero spatial domain.

[0093] Represented as:

[0094]

[0095] Representing an image The mean of the non-zero spatial domain in the y-th column, where ||||0 represents the L0 norm. Representing an image The number of non-zero elements in the y-th column.

[0096] The image is updated after copying m rows and expanding to an m×n residual stripe non-uniformity.

[0097]

[0098] Here, the symbol `repmat` represents the vector copying operation, and `[m 1]` represents copying m rows and 1 column, transforming a 1×n vector... Expanded to m×n

[0099] Step 8: Calculate the corrected image of the j-th frame.

[0100]

[0101] Example 1

[0102] In this embodiment, the verification was performed using Matlab R2020b on a computer with a 2.10 GHz processor and 16 GB of memory. To better evaluate the effectiveness of the proposed method for stripe nonuniformity correction based on spatiotemporal adaptive one-dimensional filtering, this section selects the following methods: Spectral Shaping Statistics (SSS) (Liu T, Sui X, Wang Y, et al. Strong non-uniformity correction algorithm based on spectral shaping statistics and LMS[J]. Optics Express, 2023, 31(19):30693-30709.), Dual Domain Filtering Backpropagation Neural Network (DDBP) (Liu Y, Qiu B, Tian Y, et al. Scene-based dual domain non-uniformity correction algorithm for stripe and optics-caused fixed pattern noise removal[J]. Optics Express, 2024, 32(10):16591-16610.), and Total Variation Backpropagation Neural Network (TVBP) (Vera E, Meza P, Torres S. Total variation approach for adaptive nonuniformity correction in focal-plane arrays[J]. Optics Express, 2023, 31(19):30693-30709.), Dual Domain Filtering Backpropagation Neural Network (DDBP) (Liu Y, Qiu B, Tian Y, et al. Scene-based dual domain non-uniformity correction algorithm for stripe and optics-caused fixed pattern noise removal[J]. Optics Express, 2024, 32(10):16591-16610.), and Total Variation Backpropagation Neural Network (TVBP) (Vera E, Meza P, Torres S. Total variation approach for adaptive nonuniformity correction in focal-plane arrays[J]. Optics Express, 2023, 31(19):30693-30709.). Letters, 2011, 36(2): 172-174.), Multi-scale wavelet decomposition and guided filtering (WDGF) (Cao Y, He Z, Yang J, et al. A multi-scale non-uniformity correction method based on wavelet decomposition and guided filtering for uncooled long waveinfrared camera[J]. Signal Processing: Image Communication, 2018, 60: 13-21.), Weighted guided filtering and multi-domain fusion (WGFMF) (Hong Y, Rao P, Zhou Y, et al. A weighted guided filtering-based multi-domain fusion destriping method[J].(IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024.) Five methods for fringe nonuniformity correction were compared. On 500 frames (16-bit data format, image resolution 512×640) of the FLIR Advanced Driver Assistance Systems Thermal Infrared Public Dataset, Gaussian distributed fringe nonuniformity with a mean of 0 and a standard deviation of 10 was added, with the same intensity in each column, simulating nonuniformity degradation. A comparative experiment was designed. The correction results for frames 100 and 500 are shown below. Figure 3 As shown in the figure. The SSIM curve is quantitatively evaluated using peak signal-to-noise ratio (PSNR) and structural similarity index (SSI). Figure 4 As shown. (Summary) Figure 3 and Figure 4 As a result, the fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering proposed in this invention can converge within 100 frames, effectively correcting fringe nonuniformity without generating ghosting. The PSNR of the last frame increased from 57.08dB to 78.04dB, and the SSIM increased from 0.537 to 0.997, both higher than the other five fringe nonuniformity correction methods. Moreover, the single-frame processing speed can reach within 80ms, which effectively demonstrates the excellent performance of the fringe nonuniformity correction method proposed in this invention.

[0103] In summary, this invention balances scene generalization ability, correction accuracy, and computational efficiency, and possesses the ability to distinguish between stripe non-uniformity and scene edges. It achieves adaptive correction of stripe non-uniformity in infrared images regardless of scene changes or content. Even when stripe non-uniformity coincides with scene edges, it can avoid under-correction or over-correction of stripes and blurred edges.

Claims

1. A method for correcting fringe nonuniformity based on spatiotemporal adaptive one-dimensional filtering, characterized in that, The steps are as follows: Step 1: Acquire images using an infrared imaging system; the image resolution is [missing information]. ; Step 2: For the first frame image Calculate the spatial mean by column to obtain The column spatial mean vector is normalized, and then a one-dimensional single-scale discrete wavelet transform is used to extract high-frequency detail components from the normalized vector. (This process is repeated in the original text.) Line expansion to Initial stripe non-uniformity corrected image ; Step 3: Continuously acquire images using an infrared imaging system and statistically analyze the results. Temporal mean image of the frame An iterative correction strategy is used to obtain an iteratively updated corrected image for stripe non-uniformity. and iteratively updated time-domain mean-corrected image Initialize the stripe non-uniformity correction image. Image update with residual stripe non-uniformity ; Step 4: Adjust the temporal mean of the iteratively updated image. The MAD method is used to estimate the overall stripe non-uniformity intensity of the image, and the radius of the filter is adaptively adjusted in the spatial domain. ; Step 5: Adjust the temporal mean of the iteratively updated image. An improved horizontal one-dimensional edge-sensing weighted matrix is ​​adopted. Improve image Mid-band non-uniformity and scene edge resolution; Image extraction using two-dimensional single-scale discrete wavelet transform Vertical detail component matrix Take the matrix The median of the absolute values ​​of each column is used as an estimate of the non-uniformity intensity of the stripes in each column, combined with... The regularization factor matrix of the filter is adaptively adjusted from the intensity domain. ; Step 5, as follows: (13)， in, yes The dynamic range, Represents the sensitivity coefficient. Indicates the proportionality coefficient; This indicates the time-domain mean-corrected image during iterative updates. of At, in pixels The variance of a 1×3 local window centered at the specified value; Represented in pixels The variance of a 1×3 local window centered on the x-axis, in pixels. represent All pixels; Indicates in One-dimensional edge perception weighting in the horizontal direction; in order to link the local and global perspectives. use Variance normalization for all pixels within their respective 1×3 local windows; Step 6: Correct the temporal mean of the iteratively updated image. Combining the horizontal one-dimensional weighted guided filter in the spatial domain, and substituting the adaptive filtering radius from step 4... and the regularization factor matrix in step 5 This forms a spatiotemporal adaptive one-dimensional filter, which is applied to the input image. Output image of spatiotemporal adaptive one-dimensional filter Subtraction yields the filtered image ,Will As a preliminary extracted image of residual stripe non-uniformity; Step 7: Process the initially extracted residual stripe non-uniformity image. By increasing the constraint threshold By setting the inaccurate fringe non-uniformity estimates at the scene edges to 0, the constrained residual fringe non-uniformity image is obtained, denoted as... ;right The mean value of the non-zero spatial domain is calculated by column. The vector, copy Line expansion to Residual stripe non-uniformity update image ; Step 8: Calculate the first... Corrected image of the frame .

2. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 1, characterized in that, In step 2, the first frame image... Calculate the spatial mean by column to obtain The column spatial mean vector is obtained and normalized as follows: For the first frame image The spatial mean is calculated column by column to obtain the first frame image. column space mean vector , Represents a vector, with subscripts. Represents the column number. Represents the operation of spatial mean; Frame 1 image No. Spatial mean of the column Represented as: (1)， Represents pixel position; right The normalized vector is denoted as ; No. Column-normalized column spatial mean vector Represented as: (2)。 3. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 2, characterized in that, In step 2, for High-frequency detail components are extracted using one-dimensional single-scale discrete wavelet transform, and the results are reproduced. Line expansion to Initial stripe non-uniformity corrected image The details are as follows: For vectors High-frequency detail component vectors are extracted using one-dimensional single-scale discrete wavelet transform. : (3)， Represents the estimated value. This represents a one-dimensional, single-scale discrete wavelet transform operation. Represents the Symlets 8 wavelet base; Represents the high-frequency detail component vector; to preserve the high-frequency detail component vector Length and image column width Consistent Do not downsampling; High-frequency detail component vectors copy Line expansion to Initial stripe non-uniformity corrected image : (4)， symbol This represents the operation of copying vectors, in conjunction with... Represents copying Row 1, Column 1 High-frequency detail component vector Expand to Initial stripe non-uniformity corrected image .

4. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 3, characterized in that, In step 3, images are continuously acquired using an infrared imaging system, and the results are statistically analyzed. Temporal mean image of the frame An iterative correction strategy is used to obtain an iteratively updated corrected image for stripe non-uniformity. and iteratively updated time-domain mean-corrected image Initialize the stripe non-uniformity correction image. Image update with residual stripe non-uniformity The details are as follows: Continuously acquire images using an infrared imaging system and statistically analyze the results. Temporal mean image of the frame : (5)， This represents the time-domain mean operation; An iterative correction strategy is used to obtain an iteratively updated corrected image for stripe non-uniformity. and iteratively updated time-domain mean-corrected image : (6)， Representing the The frame's stripe non-uniformity corrected image, Representing the The residual fringe non-uniformity of the frame is updated in the image, and the fringe non-uniformity correction image is initialized. and the image updated by residual stripe non-uniformity. .

5. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 4, characterized in that, In step 4, the time-domain mean-corrected image is iteratively updated. The MAD method is used to estimate the overall stripe non-uniformity intensity of the image, and the radius of the filter is adaptively adjusted in the spatial domain. The details are as follows: S4-1, Calculate the iteratively updated time-domain mean-corrected image. Average values ​​of the spatial domains in each column: (7)， in, Represents the time-domain mean-corrected image updated iteratively. No. The spatial mean of the columns, denoted as the vector formed by them, is denoted as... ; S4-2, Calculate the image The forward difference of the means of two adjacent columns forms the forward difference vector. : (8)， S4-3, Calculate the MAD value to exclude... Outliers in: (9)， Represents median operations; S4-4, Remove The vector after the outlier is denoted as : (10)， S4-5, Estimating the iteratively updated time-domain mean-corrected image Overall stripe non-uniformity intensity : (11)， Represents standard deviation calculation; S4-6, Based on the confidence interval of the normal distribution, adaptively adjust the radius of the filter in the spatial domain. : (12)， in, It represents rounding up.

6. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 5, characterized in that, In step 5, the image is extracted using two-dimensional single-scale discrete wavelet transform. Vertical detail component matrix Take the matrix The median of the absolute values ​​of each column is used as an estimate of the non-uniformity intensity of the stripes in each column, combined with... The regularization factor matrix of the filter is adaptively adjusted from the intensity domain. The details are as follows: Image extraction using two-dimensional single-scale discrete wavelet transform Vertical detail component matrix : (14)， Represents a two-dimensional single-scale discrete wavelet transform. Representing Halboki, The matrix represents the detail components in the vertical direction; to preserve the matrix With images The resolution is consistent. Do not downsampling; Take matrix Each column vector The median absolute value is used as an estimate of the non-uniformity intensity of each stripe column. : (15)， Representation matrix The Column vector; Indicates the first The regularization factor of a column, represented as a vector, is , will vector copy Line expansion to Regularity factor matrix : (16)， Combination ,Will Improved to adaptively adjust the regularization factor matrix of the dimensional filter from the intensity domain. This links the global, local, and individual columns: (17)， symbol It represents the Hadamaji.

7. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 6, characterized in that, Step 6: Correct the temporal mean of the iteratively updated image. Combining the horizontal one-dimensional weighted guided filter in the spatial domain, and substituting the adaptive filtering radius from step 4... and the adaptive regularization factor matrix in step 5 This forms a spatiotemporal adaptive one-dimensional filter, which is applied to the input image. Output image of a spatiotemporal adaptive one-dimensional filter Subtraction yields the filtered image ,Will The preliminary extracted image of residual stripe non-uniformity is shown below: Output image of a spatiotemporal adaptive one-dimensional filter as follows: (18)， in, express Horizontal window, Represented by pixels Centered on Variance within the window It is a pixel The linear multiplication coefficients of the filter transform It is a pixel Linear addition coefficients for the filter transform; Therefore Pixel-centered Inside the window The spatial mean, It is based on pixels Centered on Inside the window The spatial mean; Represented in pixels Centered on Inside the window The spatial mean; It is the input image exist The filter output at the pixel is passed through Iterate through all pixels of the window image to obtain the output image. ; By inputting an image Output image of a spatiotemporal adaptive one-dimensional filter Subtraction yields the filtered image : (19)， Will This serves as an initial image of the residual stripe non-uniformity.

8. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 7, characterized in that, Step 7 is detailed below: Image of residual stripe non-uniformity extracted in the preliminary stage By increasing the constraint threshold By setting the inaccurate fringe non-uniformity estimates at the scene edges to 0, the constrained residual fringe non-uniformity image is obtained, denoted as... : (20)， right The mean value of the non-zero spatial domain is calculated by column. vector ; This represents the mean of the non-zero spatial domain; Represented as: (21) Representing an image No. The non-zero spatial mean of the column, Describing the L0 norm, Representing an image No. The number of non-zero elements in the column; copy Line expansion to Residual stripe non-uniformity update image : (22)， Here, the symbol repmat represents the copy vector operation, in conjunction with... Represents copying Row 1, Column 1 vector Expand to of .

9. The fringe nonuniformity correction method based on spatiotemporal adaptive one-dimensional filtering according to claim 8, characterized in that, In step 8, calculate the first... Corrected image of the frame : (23)。

Images