A dynamic scene infrared image decapping method

Abstract

The present application belongs to the technical field of infrared image processing, and particularly relates to a dynamic scene infrared image cover removal method. An infrared thermal imaging device is by default closed to a cover removal algorithm, and after an infrared original image is input and a cover removal instruction is received, modeling is calculated to obtain a cover parameter table DistTable. The distance of each pixel point to the center of the image is calculated, the cover parameter table DistTable is inquired according to the distance, and a cover correction value is obtained. Each pixel point is subtracted by the cover correction value to obtain cover removal corrected image data. The dynamic scene infrared image cover removal method of the present application is real-time, and the infrared imaging system can provide more stable and higher quality image output under various environmental conditions.

Description

Technical Field

[0001] This invention belongs to the field of infrared image processing technology, specifically relating to a method for removing pot lids from infrared images of dynamic scenes. Background Technology

[0002] The "potlight effect" in infrared images is caused by uneven energy reception between the detector's center and surrounding areas due to thermal radiation from the lens structure and uneven optical energy, resulting in uneven image brightness. The characteristics of the potlight effect are: first, the potlight effect intensifies as the machine cold-starts and reaches thermal equilibrium; second, the potlight effect changes with ambient temperature. The potlight effect causes the image to appear washed out around the edges and dark in the center, resulting in loss of image depth and detail, affecting visual quality and subsequent processing and analysis. Solving the potlight effect requires first considering structural heat dissipation design; if heat dissipation cannot solve the problem, a potlight removal algorithm is needed.

[0003] There are two types of pot lid removal algorithms: calibration method and real-time calculation method. The calibration method involves acquiring a uniform surface image and calculating a compensation template, and then using the template to perform real-time pot lid removal compensation on the image; after calibration, the template image can be directly subtracted as a correction.

[0004] Mathematical modeling and real-time calculation and compensation can also be performed, which essentially solves the problem of the pot lid affecting the image. To adapt to environmental changes, multiple pot lid templates are collected and stored at different ambient temperatures. During use, the compensation coefficient of the pot lid can be calculated in real time through the temperature sensing module inside the mechanism, and the corresponding pot lid template can be selected, realizing the removal of the pot lid throughout the entire process from cold to hot. This consumes computing time and storage resources. The real-time calculation and compensation algorithm does not require data calibration. It generally uses distance function or block fitting for mathematical modeling, and then performs pot lid removal processing on the image.

[0005] Patent CN117911290A introduces an infrared image pot lid correction algorithm. The algorithm performs gridded processing on the infrared image of the pot lid and establishes a grid correction coefficient table. The infrared image containing the pot lid phenomenon is divided into blocks. The grid associated with the block to which the pixel to be corrected belongs is calculated, and the grid correction coefficient of the corresponding grid is obtained from the grid correction coefficient table. The actual correction coefficient of the pixel to be corrected is calculated through the grid correction coefficient of the pixel to be corrected. The pixel value of the pixel to be corrected is multiplied by the actual correction coefficient to obtain the correction value of the pixel to be corrected. The shortcomings of the method are: (1) Calibration and template data collection are required; (2) When the template collection environment is different from the usage environment, it will cause over-correction or under-correction; (3) From the analysis of the cause of pot lid formation, the pot lid effect is a low-frequency additive noise. After mathematical modeling, this patent uses multiplication operation for correction, which will cause over-correction and affect the scene effect.

[0006] Patent CN113421220A provides a method for removing the "pot lid" from infrared images, including: a calibration process and an actual application process; in the calibration process, a cubic polynomial function is obtained between the singular value of the ideal "pot lid-free" infrared image and the infrared focal plane temperature ti, and a quadratic function is obtained between the image pic1 and the infrared focal plane temperature ti; in the actual application process, the cubic and quadratic functions from the calibration process are used to remove the "pot lid" from the infrared image in the actual scene. The shortcomings of the method are: (1) The calibration process is cumbersome. (2) In the application process, it is still necessary to acquire a uniform surface infrared image, and then obtain an estimated background based on the fitting calculation, multiply it by the K coefficient, and obtain a "pot lid-free" image. This uniform surface infrared image is obtained through a shutter stop, which is generally installed between the lens and the detector, and cannot remove the "pot lid" caused by the temperature rise of the lens barrel.

[0007] Patent CN117197034A discloses an infrared image black pot suppression method, storage medium, and device. The method includes: calculating the distance from the original image pixel to the center of the black pot; the distance between the original image pixel and the center of the black pot is obtained by calculating the difference between the horizontal and vertical coordinates based on the pixel coordinates and the center coordinates of the black pot; calculating the edge coefficient of each pixel in the original image based on the distance between the pixel and the center of the black pot; the edge coefficient is obtained by mapping the distance between the pixel and the center of the black pot and multiplying it by an adjustment parameter; and calculating the pot lid suppression image based on the edge coefficient and the original image. This method fully utilizes the causes and effects of the pot lid effect, maps the distance to the edge coefficient for compensation, does not require pre-storing a large amount of data, and is more suitable for engineering applications. The shortcomings of the method are: (1) It does not require pre-calibration. The pot lid template correction value is obtained by calculating the distance function, which requires manual adjustment of parameters and does not automatically correct with the environment. (2) The edge coefficient matrix corrects each pixel value by point-to-point multiplication, which will lose scene details while removing the pot lid. Summary of the Invention

[0008] To address this, the present invention proposes a dynamic scene infrared image removal method for pot lids, which eliminates the need for multiple temperature point acquisitions of the pot lid template. The method performs modeling and correction based on the environmental conditions during use, adapting to various scenarios.

[0009] This invention proposes a dynamic scene infrared image removal method for removing pot lids, which can perform pot lid calibration, modeling, and correction on infrared thermal imaging equipment in use. It is immediately applicable, using currently calculated pot lid data to correct the current real-time video, removing pot lid noise as low-frequency pseudo-noise. The key is to remove high-frequency information in the scene without affecting the calculation of pot lid parameters, thus avoiding ghosting phenomena in scene imaging.

[0010] The steps of the method for removing the pot lid according to the present invention are described as follows:

[0011] (1) The infrared thermal imaging device has the pot lid removal algorithm turned off by default. After inputting the original infrared image and receiving the pot lid acquisition command, it calculates and models to obtain the pot lid parameter table DistTable.

[0012] (2) Calculate the distance from each pixel to the center of the image, and look up the pot lid parameter table DistTable according to the distance to obtain the pot lid correction value;

[0013] (3) Subtract the pot lid correction value from each pixel to obtain the image data after pot lid correction.

[0014] The infrared thermal imaging device receives the "get pot lid" command, calculates and models the parameters to obtain a pot lid parameter table; it resets the table when power is off. Then, it sends the "get pot lid" command again, calculates the pot lid parameter table using the current scene image, and replaces the previous table.

[0015] The steps for computational modeling and obtaining the pot lid parameter table are as follows:

[0016] (1) Input the current frame scene image SrcImage, perform Gaussian filtering, and obtain GaussFilterImage;

[0017] (2) Perform a logarithmic transformation on the Gaussian filtered image to convert it to the log domain to obtain GaussLogImage;

[0018] (3) Gradient calculation: extract the horizontal gradient matrix GradX and the vertical gradient matrix GradY of GaussLogImage;

[0019] (4) Convert matrices GradX and GradY into one-dimensional arrays respectively, and then concatenate and merge them to obtain a one-dimensional array Grad with twice the image resolution;

[0020] (5) Calculate the coordinate position correlation matrix for each pixel, with Q in the horizontal direction and R in the vertical direction. The parameter N controls the dimension nDim of the correlation matrix.

[0021] ;

[0022] (6) Convert the Q and R array matrices of each pixel into five-element one-dimensional arrays Mx and My, and concatenate and merge them to obtain the position matrix M;

[0023] (7) Calculate the Gamma parameter table

[0024] ;

[0025] Gamma expansion with zero padding is used for easier use in the next step.

[0026] ;

[0027] (8) Calculate the bias matrix Bias: Traverse each pixel of the image, calculate the bias parameter value Sum, and obtain the bias matrix Bias:

[0028] ;

[0029] (9) Bias correction: Subtraction is performed after converting the original image to the logarithmic domain.

[0030] ;

[0031] (10) Exponential transformation to obtain the corrected image:

[0032] ;

[0033] (11) Calculate the pot lid matrix template GGMask:

[0034] ;

[0035] (12) Divide the pot lid template into rings with the center pixel of the image as the center according to the distance difference;

[0036] (13) Calculate the pixel mean of each ring and generate an array of mean distances to the image center, RingGrayMean;

[0037] (14) Perform neighborhood smoothing correction on the distance list array RingGrayMean to generate a new mean list RingGrayMeanSmooth;

[0038] (15) Perform distance-weighted interpolation on the mean list RingGrayMeanSmooth, including the distance of each pixel in the image, to obtain the final pot lid distance parameter table DistTable.

[0039] As per steps 1-15 above, the pot lid modeling process is complete. Then, the DistTable can be used to calculate and correct the pot lid. The pot lid modeling process does not require real-time calculation; it executes once upon receiving the "Get Pot Lid" command. Real-time correction is then performed using the DistTable. The table is reset to zero upon power failure. A second "Get Pot Lid" command is then sent, and the pot lid parameter table is calculated using the current scene image, replacing the previous table.

[0040] The dynamic scene infrared image removal method of the present invention is implemented in real time, and the infrared imaging system can provide more stable and higher quality image output under various environmental conditions. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method of the present invention.

[0042] Figure 2 Flowchart for modeling the pot lid parameter table.

[0043] Figure 3 Input the original image of the pot lid scene.

[0044] Figure 4 The pot lid template GGMask is calculated and obtained.

[0045] Figure 5 Schematic diagram of the distance ring.

[0046] Figure 6 Distance ring mean table curve.

[0047] Figure 7 Image of pot lid calibration. Detailed Implementation

[0048] This invention introduces a method for removing pot lids from dynamic scene infrared images, which performs pot lid removal calibration, modeling, and correction on an uncooled infrared thermal imaging module device with a resolution of 1280*1024 that is currently in use.

[0049] The steps for removing the pot lid are as follows:

[0050] (1) Input the original infrared image and enable the pot lid removal algorithm; the pot lid removal algorithm is disabled by default.

[0051] (2) Calculate the distance from each pixel to the center of the image, and look up the pot lid parameter table DistTable according to the distance to obtain the pot lid correction value. The table length is 1 / 2 of the image diagonal, and the maximum value is no more than 820.

[0052] (3) Subtract the pot lid correction value from each pixel to obtain the image data after pot lid correction.

[0053] The infrared thermal imaging device receives the "get pot lid" command, calculates and models the parameters to obtain a pot lid parameter table; it resets the table when power is off. Then, it sends the "get pot lid" command again, calculates the pot lid parameter table using the current scene image, and replaces the previous table.

[0054] The steps for modeling the pot lid and obtaining its parameter table are as follows:

[0055] (1) Input the current frame scene image SrcImage, perform Gaussian filtering, and obtain GaussFilterImage.

[0056] (2) Perform a logarithmic transformation on the Gaussian filtered image to convert it to the log domain to obtain GaussLogImage.

[0057] (3) Gradient operation: extract the horizontal gradient matrix GradX and the vertical gradient matrix GradY of GaussLogImage.

[0058] (4) Convert matrices GradX and GradY into one-dimensional arrays respectively, and then concatenate and merge them to obtain an array Grad with twice the image resolution.

[0059] (5) Calculate the coordinate position correlation matrix for each pixel, with Q in the horizontal direction and R in the vertical direction. Parameter N is the dimension parameter, with a default value of 2, controlling the dimension nDim of the correlation matrix;

[0060] ;

[0061] i and j are dimensional loop variables, x represents the vertical coordinate, and y represents the horizontal coordinate.

[0062] (6) Convert the Q and R array matrices of each pixel into five-element one-dimensional arrays Mx and My, and then concatenate and merge them to obtain the position matrix M.

[0063] (7) Calculate the Gamma parameter table.

[0064] ;

[0065] Gamma is padded with zeros for easier use in the next step.

[0066] ;

[0067] (8) Calculate the bias matrix Bias. Traverse each pixel of the image, calculate the bias parameter value, and obtain the bias matrix Bias.

[0068] ;

[0069] (9) Bias correction. Subtraction is performed after converting the original image to the logarithmic domain.

[0070] ;

[0071] (10) Exponential transformation to obtain the corrected image.

[0072] ;

[0073] (11) Calculate the pot lid matrix template GGMask.

[0074] ;

[0075] (12) Divide the pot lid template into rings RingImg with the center pixel of the image as the center, according to the distance difference. The width of the ring is 20.

[0076] (13) Calculate the pixel mean of each ring and generate a list array of mean distances to the center of the image, RingGrayMean.

[0077] (14) Perform neighborhood smoothing correction on the distance list array RingGrayMean to generate a new mean list RingGrayMeanSmooth.

[0078] (15) Perform distance-weighted interpolation on the mean list RingGrayMeanSmooth, including the distance of each pixel in the image, to obtain the final pot lid distance parameter table DistTable.

[0079] Following steps 1-15 above, the pot lid modeling process is complete. Then, DistTable can be used to calculate and correct the pot lid. The pot lid modeling process does not require real-time calculation; it is executed once upon receiving the "Get Pot Lid" command. Real-time correction is then performed using DistTable.

[0080] By using the pot lid removal method of the present invention to achieve real-time implementation technology, the infrared imaging system can provide more stable and higher quality image output under various environmental conditions.

[0081] The infrared image removal method disclosed in this invention is based on real-time scene calibration, modeling, and correction. Specifically, when a pot-lid effect occurs during the operation of the infrared camera, a pot-lid model is modeled based on the current environment and scene, and pot-lid correction is performed on real-time video frames. This eliminates the need for multi-temperature point calibration, extensive storage of pot-lid models, and can be implemented on embedded chips, improving infrared image quality. No adjustment of the pot-lid removal method parameters is required. Before the infrared camera leaves the factory, there is no need to calibrate and store the pot-lid removal matrix template; during the use of the infrared camera, there is no need to calibrate the temperature uniformity surface and store the pot-lid removal matrix template.

Claims

1. A method for removing pot lids from dynamic scene infrared images, characterized in that, Includes the following steps: (1) The infrared thermal imaging device has the pot lid removal algorithm turned off by default. After inputting the original infrared image and receiving the pot lid acquisition command, it calculates and models to obtain the pot lid parameter table DistTable. (2) Calculate the distance from each pixel to the center of the image, and look up the pot lid parameter table DistTable according to the distance to obtain the pot lid correction value; (3) Subtract the pot lid correction value from each pixel to obtain the image data after pot lid correction; The steps for computational modeling and obtaining the pot lid parameter table are as follows: (1) Input the current frame scene image SrcImage, perform Gaussian filtering, and obtain GaussFilterImage; (2) Perform a logarithmic transformation on the Gaussian filtered image to convert it to the log domain to obtain GaussLogImage; (3) Gradient calculation: extract the horizontal gradient matrix GradX and the vertical gradient matrix GradY of GaussLogImage; (4) Convert matrices GradX and GradY into one-dimensional arrays respectively, and then concatenate and merge them to obtain a one-dimensional array Grad with twice the image resolution; (5) Calculate the coordinate position correlation matrix for each pixel, with Q in the horizontal direction and R in the vertical direction. The parameter N controls the dimension nDim of the correlation matrix. ； ； ； (6) Convert the Q and R array matrices of each pixel into five-element one-dimensional arrays Mx and My, and concatenate and merge them to obtain the position matrix M; (7) Calculate the Gamma parameter table ； Gamma expansion with zero padding is used for easier use in the next step. ； (8) Calculate the bias matrix Bias: Traverse each pixel of the image, calculate the bias parameter value Sum, and obtain the bias matrix Bias: ； (9) Bias correction: Subtraction is performed after converting the original image to the logarithmic domain. ； (10) Exponential transformation to obtain the corrected image: ； (11) Calculate the pot lid matrix template GGMask: ； (12) Divide the pot lid template into rings with the center pixel of the image as the center according to the distance difference; (13) Calculate the pixel mean of each ring and generate an array of mean distances to the image center, RingGrayMean; (14) Perform neighborhood smoothing correction on the distance list array RingGrayMean to generate a new mean list RingGrayMeanSmooth; (15) Perform distance-weighted interpolation on the mean list RingGrayMeanSmooth, including the distance of each pixel in the image, to obtain the final pot lid distance parameter table DistTable.

2. The method for removing pot lids from dynamic scene infrared images according to claim 1, characterized in that, When the power is off, the pot lid parameter table will be cleared to zero. Then, the command to retrieve the pot lid is sent again, and the pot lid parameter table is calculated again using the current scene image, replacing the previous pot lid parameter table.

Images