An Infrared Image Denoising Method Based on Scene Correction

Abstract

The present invention discloses an infrared image denoising method based on scene correction, and the steps are as follows: S1: Compensate the current frame image and the historical frame image using the gain image G and the offset image O data; S2: Calculate the true value of the compensated current frame image at each pixel position; S3: Calculate the learning factor at each pixel position; S4: Update the gain image G and the offset image O data; S5: Update the threshold; S6: Output the compensated image of the current frame as the denoising result, collect the next frame of image data, and restart the steps. After the conventional two-point correction, the present invention enables a scene-based correction method to indirectly correct the gain data and offset data of the two-point correction, achieving the denoising effect.

Description

Technical Field

[0001] The present invention relates to the field of infrared image denoising, and particularly to an infrared image denoising method based on scene correction. Background Art

[0002] Infrared detectors have non-uniformity characteristics and often use a two-point calibration method for non-uniformity correction. However, infrared detectors have a temperature drift effect. After a period of time after the blocking NUC correction, the patch noise in the image becomes obvious. This short-term change in detector performance caused by temperature rise makes the offset data obtained at the moment of blocking the shim no longer applicable. Moreover, the long-term temperature change, such as the gain data obtained by the detector in winter is not applicable in summer, and the image quality drops significantly.

[0003] Existing infrared image denoising algorithms are based on the idea of edge-preserving denoising and perform denoising in the spatial domain, time domain, frequency domain, transform domain, etc. Classical algorithms include bilateral filtering, guided filtering, and NLM filtering in the spatial domain, algorithms combining the time domain and spatial domain such as 3dnr, wavelet domain denoising, and algorithms combining the spatial domain and frequency domain ideas such as BM3D. However, when the noise is particularly severe, increasing the denoising strength will result in the loss of image details.

[0004] Regarding the problem of significant degradation of image quality caused by temperature drift, the essence lies in the inapplicability of gain data and offset data. The scene-based correction method continuously updates the offset data and gain data through continuous scene data. Different from this, the method based on scene correction is used in the NUC link and is an algorithm to replace the conventional two-point correction.

[0005] The methods based on scene correction include time-domain high-pass algorithms, constant statistics algorithms, neural network algorithms, and algebraic averaging methods. Among them, the time-domain high-pass method does not involve the update of gain data. The constant statistics method involves the update of gain data and offset data but heavily depends on scene motion. The neural network method involves the update of gain data and offset data but is limited by the acquisition of true values. The algebraic averaging method depends on the transfer and accumulation of matrix elements and is prone to incorrect calculations and transfer errors at the edges. These methods have different computational complexities, each with its own advantages and disadvantages, but all require scene motion. Summary of the Invention

[0006] To solve the technical problem of significant degradation of image quality, the present invention proposes an infrared image denoising method based on scene correction. After the conventional two-point correction, that is, after the non-uniformity of the infrared image data is corrected to a certain extent, the scene-based correction method is then enabled to indirectly correct the gain data and offset data of the two-point correction, achieving the effect of denoising.

[0007] An infrared image denoising method based on scene correction includes the following steps:

[0008] S1: Compensate the current frame image and the historical frame image using the gain image G and the offset image O data;

[0009] S2: Calculate the true value of the compensated current frame image at each pixel position;

[0010] S3: Calculate the learning factor at each pixel position;

[0011] S4: Update the gain image G and the offset image O data;

[0012] S5: Update the threshold;

[0013] S6: Output the compensated image of the current frame as the denoising result, collect the next frame of image data, and restart the steps.

[0014] Furthermore, an infrared image denoising method based on scene correction, where S1 includes the following sub-steps:

[0015] S11: Compensate the current frame image and the historical frame image at the coordinate (i, j) using the gain image G and the offset image O data. The compensation formula is:

[0016] ;

[0017] Wherein, represents the image value at the coordinate (i, j) of the current frame image, represents the image value at the coordinate (i, j) of the historical frame image, represents the current frame image data with height H and width W, represents the historical frame image data with height H and width W;

[0018] Wherein, represents the offset image value at the coordinate (i, j), represents the gain image value at the coordinate (i, j); The gain image G is initialized, and each element of the gain image G is set to 1; The offset image O is initialized, and each element is set to 0; The width and height of the gain image G and the offset image O are the same as those of the current frame image;

[0019] Wherein, represents the pixel value of the compensated current frame image at the coordinate (i, j), represents the pixel value of the compensated historical frame image at the coordinate (i, j).

[0020] Furthermore, an infrared image denoising method based on scene correction, where S2 includes the following sub-steps:

[0021] S21: Calculate the image value at the coordinate (i, j) of the compensated current frame image True value statistical information on the historical frame image compensation map ;

[0022] S22: Calculate the image value at the coordinate (i, j) of the compensated current frame image True value statistical information on the current frame image compensation map ;

[0023] S23: Calculate the true value .

[0024] Furthermore, an infrared image denoising method based on scene correction, where S21 includes the following sub-steps:

[0025] S211: Accumulate the matching point information ysum and wsum:

[0026] S2111: Search within a radius of nR at the coordinate (i, j) of the compensated historical frame image. The vertical search range is p ∈ [i - nR, i + nR], and the horizontal search range is q ∈ [j - nR, j + nR]. When the pixel value at the point (p, q) meets the search condition, perform weight accumulation. The formula is: ;

[0027] ;

[0028] Among them, represents the cumulative sum after multiplying the pixel value by the weight, wsum represents the cumulative weight sum, and ysum and wsum are both cumulative matching point information, initialized to 0. represents the pixel value at the coordinate (p, q) on the compensated historical frame image.

[0029] Among them, ;

[0030] represents the weight. represents the bilateral filtering coefficient. represents the degree of motion. represents the image difference information, and Thr represents the threshold.

[0031] Among them,

[0032] ;

[0033] Among them,

[0034] ;

[0035] represents the coefficient inversely proportional to the local variance of the image ; represents the threshold level;

[0036] Among them,

[0037] ;

[0038] represents the natural exponential function, , represents the bilateral filtering parameter;

[0039] Among them,

[0040] ;

[0041] S212: When traversing p and q pixels within the search range, make the image difference information ydiff the smallest is ymatch, and the most matching pixel value is statistically calculated. The formula is:

[0042] ;

[0043] Among them, represents the most matching pixel value, represents the minimum value.

[0044] Furthermore, an infrared image denoising method based on scene correction, the S22 includes the following sub-steps:

[0045] S221: Continue to accumulate the matching point information ysum and wsum;

[0046] S2211: Search within a radius of nR at the coordinate (i, j) of the compensated current frame image The vertical search range is p ∈ [i - nR, i + nR], and the horizontal search range is q ∈ [j - nR, j + nR]. When p and q meet the search conditions, weight accumulation is performed. The formula is:

[0047] ;

[0048] Among them, represents the cumulative sum after multiplying the pixel value by the weight, and wsum represents the weight cumulative sum;

[0049] Among them, ;

[0050] represents the weight, represents the bilateral filtering coefficient, represents the degree of motion. At this time is a fixed value, , thr represents the threshold level, represents the image difference information;

[0051] Among them,

[0052] ;

[0053] Among them,

[0054] ;

[0055] represents the natural exponential function, , represent bilateral filtering parameters;

[0056] S222: When traversing p and q pixels within the search range, update the most matching pixel value, and the formula is:

[0057] ;

[0058] Among them, represents the updated most matching pixel value.

[0059] Furthermore, a method for denoising infrared images based on scene correction, where S23 includes the following sub-steps:

[0060] S231: Calculate the true value , and the formula is:

[0061] ;

[0062] Among them, both a and b represent configurable coefficients, ysum and wsum both represent matching point information, represents the updated most matching pixel value.

[0063] Furthermore, a method for denoising infrared images based on scene correction, where S3 includes the following sub-steps:

[0064] S31: Calculate the learning factor at the coordinates (i, j), and the formula is:

[0065] ;

[0066] Among them, represents the learning factor at the coordinates (i, j), represents the configuration coefficient, represents the degree of motion factor;

[0067] Among them,

[0068] ;

[0069] all represent configurable coefficients, and else represents otherwise, Representing the degree of motion;

[0070] Among them,

[0071] .

[0072] Furthermore, an infrared image denoising method based on scene correction, where S4 includes the following sub-steps:

[0073] S41: Update the data of the gain image G and the bias image O, and the formula is:

[0074] ;

[0075] Among them, represents the learning factor, represents the true value, represents the pixel value of the compensated current frame image at coordinates (i,j), represents the image value at coordinates (i,j) of the current frame image, represents the updated data of the gain image G, represents the updated data of the bias image O.

[0076] Furthermore, an infrared image denoising method based on scene correction, where S5 includes the following sub-steps:

[0077] S51: Update the threshold level thr, and the formula is:

[0078] ;

[0079] Among them, represents the image noise level of the current frame after compensation, calculated by calculation, represents the image noise level of the current frame before compensation, calculated by calculation;

[0080] Among them,

[0081] ;

[0082] H represents the image height, W represents the image width, and var1(i,j) represents the local variance at coordinates (i,j);

[0083] Among them,

[0084] ;

[0085] represents the coordinate coordinates of points within the neighborhood range, represents coordinates on the image The mean value at

[0086] ;

[0087] represents The pixel value at coordinates (p, q) on the image;

[0088] where

[0089] ;

[0090] H represents the height of the image, W represents the width of the image, and var0(i, j) represents the local variance at coordinates (i, j);

[0091] where

[0092] ;

[0093] represents the coordinates within the neighborhood range of the point coordinates, represents the coordinates on the image at the mean value;

[0094] ;

[0095] represents The pixel value at coordinates (p, q) on the image.

[0096] The beneficial effects of the present invention are as follows: Through an infrared image denoising method based on scene correction, based on the idea of neural network scene correction, improvements are proposed for the acquisition of the true value. The true value is calculated using a combination of the spatio-temporal domain, and the weights of the true value obtained in the time domain and the true value obtained in the spatial domain are adjusted according to the judgment of the degree of motion to adapt to the moving scene and the static scene; improvements are proposed for the artifact effect, and the degree of participation of edge pixels in the calculation of the true value is controlled using local variance constraint and denoising degree judgment; the learning rate is adaptively adjusted through the judgment of the degree of motion, so as to effectively denoise the infrared data. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] Figure 1 is the flowchart of this method.

[0098] Figure 2 is the flowchart for calculating the straight value.

[0099] Figure 3 is the flowchart for statistically analyzing the true value information on the historical compensation image.

[0100] Figure 4It is a flowchart for counting the true value information on the current compensated image. Detailed implementation manner

[0101] The present invention will be further described, but the protection scope of the present invention is not limited to the following.

[0102] In two infrared images with a height of H and a width of W , the two images are 14-bit raw image data after two-point correction, and the front and rear frame data adjacent in time. is the historical frame image, is the current frame image. Use the following method to denoise each pixel point on the current frame image, and the obtained result is used as the denoised result image of the current frame.

[0103] Initialize the gain image G and the offset image O, that is, set each element of G to 1 and each element of O to 0. The width and height of the G and O images are the same as those of the current frame image.

[0104] Taking the image value at the coordinate (i, j) as an example, where i and j are dummy variables, and the coordinate positions (i, j) represent different pixel positions respectively, the denoising process of the present invention is described as follows:

[0105] As shown in the appendix Figure 1 A method for denoising infrared images based on scene correction includes the following steps:

[0106] S1: Compensate the current frame image and the historical frame image using the data of the gain image G and the offset image O;

[0107] S2: Calculate the true value at each pixel position of the compensated current frame image;

[0108] S3: Calculate the learning factor at each pixel position;

[0109] S4: Update the data of the gain image G and the offset image O;

[0110] S5: Update the threshold;

[0111] S6: Output the compensated image of the current frame as the denoising result, and collect the next frame of image data to restart the steps.

[0112] As shown in the appendix Figure 2 A method for denoising infrared images based on scene correction, where the S1 includes the following sub-steps:

[0113] S11: Compensate the current frame image and the historical frame image at the coordinate (i, j) using the data of the gain image G and the offset image O, and the compensation formula is:

[0114] ;

[0115] Among them, represents the image value at the image coordinates (i, j) of the current frame, represents the image value at the image coordinates (i, j) of the historical frame, represents the current frame image data with height H and width W, represents the historical frame image data with height H and width W;

[0116] Among them, represents the offset image value at the coordinates (i, j), represents the gain image value at the coordinates (i, j); the gain image G is initialized, and each element of the gain image G is set to 1; the offset image O is initialized, and each element is set to 0; the width and height of the gain image G and the offset image O are the same as those of the current frame image;

[0117] Among them, represents the pixel value at the coordinates (i, j) of the compensated current frame image, represents the pixel value at the coordinates (i, j) of the compensated historical frame image.

[0118] As shown in the appendix Figure 3 shown, a method for denoising infrared images based on scene correction, the S2 includes the following sub-steps:

[0119] S21: Calculate the image value at the coordinates (i, j) of the compensated current frame image The true value statistical information on the compensated image of the historical frame ;

[0120] S22: Calculate the image value at the coordinates (i, j) of the compensated current frame image The true value statistical information on the compensated image of the current frame ;

[0121] S23: Calculate the true value .

[0122] As shown in the appendix Figure 4 shown, a method for denoising infrared images based on scene correction, the S21 includes the following sub-steps:

[0123] S211: Accumulate the matching point information ysum, wsum:

[0124] S2111: Search within a radius of nR at the coordinates (i, j) of the compensated historical frame image The vertical search range p ∈ [i - nR, i + nR], and the horizontal search range q ∈ [j - nR, j + nR]. When the pixel value at the point (p, q) meets the search condition, perform weight accumulation. The formula is:

[0125] ;

[0126] Among them, represents the cumulative sum after multiplying the pixel value by the weight. wsum represents the cumulative weight sum. ysum and wsum are both cumulative matching point information, and both are initialized to 0. represents the pixel value at the coordinate (p, q) on the compensated historical frame image.

[0127] Among them, ;

[0128] represents the weight. represents the bilateral filtering coefficient. represents the degree of motion. represents the image difference information, and Thr represents the threshold.

[0129] Among them,

[0130] ;

[0131] Among them,

[0132] ;

[0133] represents the coefficient inversely proportional to the local variance of the image ; represents the threshold level.

[0134] Among them,

[0135] ;

[0136] represents the natural exponential function. , represent the bilateral filtering parameters.

[0137] Among them,

[0138] ;

[0139] S212: When traversing the p and q pixels within the search range, the that minimizes the image difference information ydiff is ymatch, and the most matching pixel value is statistically calculated. The formula is:

[0140] ;

[0141] Among them, represents the most matching pixel value. represents the minimum value.

[0142] Further, an infrared image denoising method based on scene correction, where S22 includes the following sub-steps:

[0143] S221: Continue to accumulate the matching point information ysum and wsum;

[0144] S2211: Search within a radius of nR at the coordinate (i, j) in the compensated current frame image The vertical search range is p ∈ [i - nR, i + nR], and the horizontal search range is q ∈ [j - nR, j + nR]. When p and q meet the search conditions, perform weight accumulation. The formula is:

[0145] ;

[0146] Among them, represents the sum of the products of the pixel values and the weights, and wsum represents the sum of the weight accumulations;

[0147] Among them, ;

[0148] represents the weight, represents the bilateral filtering coefficient, represents the degree of motion. At this time, is a fixed value, , thr represents the threshold level, represents the image difference information;

[0149] Among them,

[0150] ;

[0151] Among them,

[0152] ;

[0153] represents the natural exponential function, 、 represent the bilateral filtering parameters;

[0154] S222: When traversing the p and q pixels within the search range, update the most matching pixel value. The formula is:

[0155] ;

[0156] Among them, represents the updated most matching pixel value.

[0157] Further, an infrared image denoising method based on scene correction, where S23 includes the following sub-steps:

[0158] S231: Calculate the true value , the formula is:

[0159] ;

[0160] Among them, both a and b represent configurable coefficients, ysum and wsum both represent matching point information, represents the updated most matching pixel value.

[0161] Further, an infrared image denoising method based on scene correction, the S3 includes the following sub-steps:

[0162] S31: Calculate the learning factor at coordinates (i, j) , the formula is:

[0163] ;

[0164] Among them, represents the learning factor at coordinates (i, j), represents the configuration coefficient, represents the degree of motion factor;

[0165] Among them,

[0166] ;

[0167] all represent configurable coefficients, else represents otherwise, represents the degree of motion;

[0168] Among them,

[0169] .

[0170] Further, an infrared image denoising method based on scene correction, the S4 includes the following sub-steps:

[0171] S41: Update the gain image G and bias image O data, the formula is:

[0172] ;

[0173] Among them, represents the learning factor, represents the true value, represents the pixel value of the compensated current frame image at coordinates (i, j), represents the image value at coordinates (i, j) of the current frame image, represents the updated gain image G data, represents the updated bias image O data.

[0174] Further, an infrared image denoising method based on scene correction, where S5 includes the following sub-steps:

[0175] S51: Update the threshold level thr, with the formula:

[0176] ;

[0177] Wherein, represents the image noise level after compensation for the current frame, calculated by ; represents the image noise level before compensation for the current frame, calculated by ;

[0178] Wherein,

[0179] ;

[0180] H represents the height of the image, W represents the width of the image, and var1(i,j) represents the local variance at the coordinate (i,j);

[0181] Wherein,

[0182] ;

[0183] represents the point coordinates within the neighborhood of the coordinate , represents the mean value at the coordinate on the image;

[0184] ;

[0185] represents the pixel value at the coordinate (p,q) on the image;

[0186] Wherein,

[0187] ;

[0188] H represents the height of the image, W represents the width of the image, and var0(i,j) represents the local variance at the coordinate (i,j);

[0189] Wherein,

[0190] ;

[0191] represents the point coordinates within the neighborhood of the coordinate , represents the mean value at the coordinate on the image;

[0192] ;

[0193] represents the pixel value at the coordinate (p, q) on the image.

[0194] It should be noted that: for the pixels at the boundary of the infrared image, such as when i < nR, the vertical coordinate of the upper-left vertex of the neighborhood is 1, and the vertical coordinate of the lower-right vertex of the neighborhood is nD (nD is the diameter of the neighborhood, nD = 2×nR + 1); similarly, when i > H - nR, the vertical coordinate of the upper-left vertex of the neighborhood is H - nD + 1, and the vertical coordinate of the lower-right vertex of the neighborhood is H;

[0195] It should be noted that: for the pixels at the boundary of the infrared image, such as when j < nR, the horizontal coordinate of the upper-left vertex of the neighborhood is 1, and the horizontal coordinate of the lower-right vertex of the neighborhood is D; similarly, when i > W - nR, the horizontal coordinate of the upper-left vertex of the neighborhood is W - nD + 1, and the horizontal coordinate of the lower-right vertex of the neighborhood is W.

[0196] Specific Embodiment 1 Specific Setting of Parameters

[0197] S2111: Search within a radius of nR at the coordinate (i, j) in the compensated historical frame image with nR taking the value of 2, the vertical coordinate search range p ∈ [i - nR, i + nR], and the horizontal coordinate search range q ∈ [j - nR, j + nR]. When p and q meet the search conditions, perform weight accumulation, and the formula is:

[0198] Search within a radius of nR at the coordinate (i, j) in the compensated current frame image with the vertical coordinate search range p ∈ [i - nR, i + nR] and the horizontal coordinate search range q ∈ [j - nR, j + nR]. When p and q meet the search conditions, perform weight accumulation, and the formula is:

[0199] ;

[0200] Among them, represents the cumulative sum after multiplying the pixel value by the weight, wsum represents the weight cumulative sum, and ysum and wsum are both cumulative matching point information;

[0201] Among them, ;

[0202] represents the weight, represents the bilateral filtering coefficient, represents the degree of motion, represents the image difference information, and Thr represents the threshold;

[0203] Among them,

[0204] ;

[0205] Among them,

[0206] ;

[0207] represents a calculation coefficient inversely proportional to the local variance var;

[0208] Among them,

[0209] ;

[0210] is a configurable coefficient with a value of 128;

[0211] represents the threshold level, which is related to the image noise level, and the calculation method has been described in S51;

[0212] Among them,

[0213] ;

[0214] Among them,

[0215] ;

[0216] represents the natural exponential function, , represents the bilateral filtering parameter, takes the value 30, takes the value nR * 3;

[0217] Among them,

[0218] ;

[0219] S212: When traversing p and q pixels within the search range, the that minimizes the image difference information ydiff is ymatch, and the most matching pixel value is statistically calculated. The formula is:

[0220] ;

[0221] Among them, represents the most matching pixel value, represents the minimum value.

[0222] The said S23 includes the following sub - steps:

[0223] S231: Calculate the true value , and the formula is:

[0224] ;

[0225] Among them, both a and b represent configurable coefficients, both taking the value of 4, ysum and wsum both represent matching point information, represents the updated most matching pixel value.

[0226] S3 includes the following sub-steps:

[0227] S31: Calculate the learning factor at coordinates (i, j) , and the formula is:

[0228]

[0229] Among them, represents the learning factor at coordinates (i, j), represents the configuration coefficient, taking the value of 1e-2, represents the degree of motion factor;

[0230] Among them,

[0231] ;

[0232] all represent configurable coefficients, taking the values of fm0 > 1, fm1 < 1, fm2 = 1, and else represents otherwise, represents the degree of motion, and Thr represents the threshold.

[0233] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An infrared image denoising method based on scene correction, characterized in that, It includes the following steps: S1: Compensate the current frame image and the historical frame image using the gain image G and the offset image O data; S2: Calculate the true value of the compensated current frame image at each pixel position; S3: Calculate the learning factor at each pixel position; S4: Update the gain image G and the offset image O data; S5: Update the threshold; S6: Output the compensated image of the current frame as the denoising result, collect the next frame of image data, and restart the steps; The S2 includes the following sub-steps: S21: Calculate the image value at the compensated current frame image coordinates (i, j). On the compensated map of the historical frame image The true value statistical information; S22: Calculate the image value at the compensated current frame image coordinates (i, j). True value statistical information on the compensated image of the current frame image ; S23: Calculate the true value ; The S21 includes the following sub-steps: S211: Accumulate the matching point information ysum and wsum: S2111: Search within a radius of nR at the coordinate (i, j) in the compensated historical frame image Search within a range of p ∈ [i - nR, i + nR] for the vertical coordinate and q ∈ [j - nR, j + nR] for the horizontal coordinate within a radius of nR at the coordinate (i, j). When the pixel value at the point (p, q) meets the search condition, perform weighted accumulation. The formula is as follows: ； Among them, represents the cumulative sum after multiplying the pixel value by the weight. wsum represents the cumulative weight sum. Both ysum and wsum are cumulative matching point information, and they are both initialized to 0. represents the pixel value at the coordinate (p, q) on the compensated historical frame image. Among them, ; represents the weight, represents the bilateral filtering coefficient, represents the degree of motion, represents the image difference information, and Thr represents the threshold; Wherein, ； Wherein, ； A coefficient inversely proportional to the local variance of the image ; and a threshold level. Wherein, ； represents the natural exponential function, , represents the bilateral filtering parameter; Wherein, ； S212: When traversing pixels p and q within the search range, the one that minimizes the image difference information ydiff is ymatch, and the most matching pixel value is counted. The formula is: ； Among them, represents the most matching pixel value, represents the minimum value.

2. The infrared image denoising method based on scene correction according to claim 1, characterized in that, The S1 includes the following sub-steps: S11: Compensate the current frame image and the historical frame image at the coordinate (i, j) using the gain image G and the offset image O data, and the compensation formula is: ； Among them, represents the image value at the image coordinates (i, j) of the current frame image, represents the image value at the image coordinates (i, j) of the historical frame image, represents the current frame image data with height H and width W, represents the historical frame image data with height H and width W; Among them, represents the offset image value at the coordinate (i, j), represents the gain image value at the coordinate (i, j); the gain image G is initialized, and each element of the gain image G is set to 1; the offset image O is initialized, and each element is set to 0; the width and height of the gain image G and the offset image O are the same as those of the current frame image; Among them, represents the pixel value of the compensated current frame image at coordinates (i, j), represents the pixel value of the compensated historical frame image at coordinates (i, j).

3. A method for denoising infrared images based on scene correction according to claim 1, characterized in that The S22 includes the following sub-steps: S221: Continue to accumulate the matching point information ysum and wsum; S2211: Search within a radius of nR at the coordinate (i, j) in the compensated current frame image Search within the range of p ∈ [i - nR, i + nR] for the vertical coordinate and q ∈ [j - nR, j + nR] for the horizontal coordinate within a radius of nR at the coordinate (i, j). When p and q meet the search conditions, perform weighted accumulation. The formula is as follows: ； Among them, represents the sum of the products of pixel values and weights, and wsum represents the sum of weights; Among them, ; represents the weight, represents the bilateral filtering coefficient, represents the degree of motion, at this time is a fixed value, , thr represents the threshold level, represents the image difference information; Wherein, ； Wherein, ； represents the natural exponential function, , represents the bilateral filtering parameter; S222: When traversing the p and q pixels within the search range, update the most matching pixel value, and the formula is: ； Among them, represents the updated most matching pixel value.

4. A method for denoising infrared images based on scene correction according to claim 1, characterized in that The S23 includes the following sub-steps: S231: Calculate the true value , and the formula is: ； Among them, both a and b represent configurable coefficients, and both ysum and wsum represent matching point information, representing the updated most matching pixel value.

5. A method for denoising infrared images based on scene correction according to claim 1, characterized in that, The S3 includes the following sub-steps: S31: Calculate the learning factor at coordinates (i, j) , and the formula is: ； Among them, represents the learning factor at the coordinate (i, j), represents the configuration coefficient, represents the degree of motion factor; Wherein, ； All represent configurable coefficients, and else represents otherwise. represents the degree of movement; Wherein, 。 6. A method for denoising infrared images based on scene correction according to claim 1, characterized in that, The S4 includes the following sub-steps: S41: Update the gain image G and the offset image O data, and the formula is: ； Among them, represents the learning factor, represents the true value, represents the pixel value of the compensated current frame image at coordinates (i, j), represents the image value of the current frame image at coordinates (i, j), represents the updated gain image G data, represents the updated offset image O data.

7. A method for denoising infrared images based on scene correction according to claim 1, characterized in that, The S5 includes the following sub-steps: S51: Update the threshold level thr, and the formula is: ； Among them, represents the image noise level after compensation for the current frame, which is calculated by . represents the image noise level before compensation for the current frame, which is calculated by . Wherein, ； H represents the image height, W represents the image width, and var1(i, j) represents the local variance at the coordinate (i, j); Wherein, ； Representative coordinates Point coordinates within the neighborhood range, represent the coordinates on the image and the mean value at that position; ； representative the pixel value at the coordinate (p, q) on the image; Wherein, ； H represents the image height, W represents the image width, and var0(i, j) represents the local variance at the coordinate (i, j); Wherein, ； Representative coordinates Point coordinates within the neighborhood range, represent the coordinates on the image and the mean value at that location; ； represent the pixel value at coordinates (p, q) on the image

Images