An adaptive super-large field of view infrared image non-uniformity correction method

Abstract

The application discloses a self-adaptive super-large field of view infrared image non-uniformity correction method, and belongs to the field of infrared image quality control. Firstly, the infrared camera laboratory radiation calibration coefficient is taken as a correction input, and the initial parameters of non-uniformity correction are obtained through multi-scene image statistics; then, the detector response of the super-large field of view center and stable response is selected as a standard reference benchmark, and the correction parameters of each pixel are corrected by using a specific sample matching technology from both sides of the standard reference benchmark to the detector, so that the non-uniformity correction of a single detector module is completed; then, a high-precision overlapping area correlation detection constraint model facing adjacent detectors is established, feature point matching correction is carried out by using the overlapping pixels of adjacent detectors, the correction parameters between adjacent detector modules are obtained, and the correction parameters are sequentially transmitted, so that the gray level balancing of multiple detector modules of the infrared camera is realized; finally, histogram statistics is used to realize self-adaptive dynamic stretching of the full field of view image, and non-uniformity correction of the whole image is realized.

Description

Technical Field

[0001] This invention belongs to the field of infrared image quality control, and particularly relates to an adaptive method for correcting non-uniformity of ultra-large field-of-view infrared images. Background Technology

[0002] Infrared imaging utilizes infrared camera detectors to collect the infrared radiation characteristics emitted or reflected by objects, converting the object's thermal radiation into electrical signals, and ultimately acquiring an infrared image that presents the object's temperature information. Because infrared imaging does not require an auxiliary light source, it can perform imaging anytime, anywhere. However, due to limitations in semiconductor materials and manufacturing processes, the responses of individual pixels in an infrared detector exhibit significant inconsistencies, resulting in noticeable texture in the infrared image output by the camera—that is, the presence of non-uniform noise between pixel responses, which directly affects image quality. For large field-of-view infrared cameras, the pixel size of the infrared detector is extremely large, making the non-uniform noise between pixel responses even more pronounced. Therefore, effectively correcting the non-uniformity of infrared images is one of the key technologies that infrared imaging technology needs to address.

[0003] Traditional methods for correcting non-uniformity in spaceborne infrared images mainly fall into two categories: blackbody calibration-based methods and large-sample statistical methods. For ultra-wide field-of-view infrared camera imaging, to balance the requirements of a large field of view and resolution, the camera's optical aperture is extremely large. Therefore, setting up a full-path calibration blackbody for blackbody calibration methods is not only very costly but also increases the camera's reliability risk. Furthermore, the effective application of traditional large-sample statistical correction methods relies on the stability of the background radiation in the camera's imaging. Typically, the temperature resolution of infrared imaging can reach the millikelvin level, making it highly sensitive to temperature changes. Additionally, ultra-wide field-of-view infrared cameras employ large-aperture off-axis three-mirror optical systems, which cannot guarantee the pupil matching requirements of traditional infrared systems. With changes in satellite attitude maneuvering, the camera's background radiation changes rapidly and becomes even more unstable. Therefore, these methods cannot achieve good non-uniformity correction results for infrared images, especially those with ultra-wide fields of view, which are sensitive to the background environment. Furthermore, ultra-wide field-of-view infrared cameras are composed of multiple detector modules. Due to limitations in semiconductor materials and manufacturing processes, the responses of individual pixels in the infrared detectors exhibit significant inconsistencies. This leads to different grayscale images acquired by the multiple detectors. Because of the limited number of feature points in the overlapping areas between adjacent detector modules, feature matching registration methods cannot be used. Moreover, the high-fidelity quantitative requirements of the data make it difficult to alter the original relative correspondence between infrared radiation and code values ​​using grayscale histogram matching. Therefore, traditional methods are insufficient to meet the requirements for non-uniformity correction in ultra-wide field-of-view infrared images. Summary of the Invention

[0004] The purpose of this invention is to propose an adaptive method for correcting non-uniformity in ultra-large field-of-view infrared images, which can effectively solve the problem of difficult non-uniformity correction in ultra-large field-of-view infrared images.

[0005] (II) Technical Solution

[0006] To achieve the above objectives and solve the above technical problems, the technical solution of the present invention is as follows:

[0007] Using the laboratory radiometric calibration coefficient of the infrared camera as the correction input, the initial parameters for non-uniformity correction were obtained through statistical analysis of multiple images.

[0008] The detector response with a large field of view and stable response is selected as the standard reference benchmark. The correction parameters of each pixel are corrected by using a specific sample matching technique on both sides of the detector with the standard reference benchmark to complete the non-uniformity correction of a single detector module.

[0009] A high-precision correlation detection constraint model for overlapping regions of adjacent detectors is established based on the optical system of an infrared camera. This model overcomes problems such as insufficient feature points in overlapping regions and changes in the quantitative relationship of code values. Feature point matching and correction are performed using overlapping pixels of adjacent detectors to obtain correction parameters between adjacent detector modules. These parameters are then passed sequentially to achieve grayscale balancing of multiple detector modules of the infrared camera.

[0010] Finally, based on specific application scenarios, histogram statistics are used to achieve adaptive dynamic stretching of the entire field of view image, thereby correcting the non-uniformity of the entire image.

[0011] (III) Effective Returns

[0012] 1. The present invention proposes an adaptive method for non-uniformity correction of ultra-large field-of-view infrared images, which can effectively solve the problem of difficulty in non-uniformity correction of ultra-large field-of-view infrared images.

[0013] 2. This invention breaks away from the previous reliance on on-board calibrators for infrared camera image correction, and at the same time adopts adaptive correction technology to reduce the drawback of infrared detectors being sensitive to changes in background temperature.

[0014] 3. The present invention is not limited by camera size, optical field of view and detector array length for infrared image correction. It reduces the complexity of hardware design and the design risk of spaceborne cameras, broadens the design ideas of spaceborne infrared cameras, and thus can achieve the correction effect of ultra-large field of view infrared images at a lower cost, effectively improving image quality.

[0015] 4. Since the non-uniformity of the entire image is largely affected by the differences in actual ground features, this invention fully considers these differences. Through sample selection rules, samples affected by ground features are effectively removed, thereby reducing the influence of ground feature factors in a single scene image. The sample selection rules and correction model fully consider the features in images of different scenes, making them more adaptable and facilitating the rapid selection of samples in a scene to achieve non-uniformity correction of a single scene image.

[0016] 5. Due to the limitations of insufficient features in overlapping regions and high requirements for code value quantification in ultra-large field-of-view images, it is difficult to perform grayscale balancing of images from multiple detector modules using conventional methods. This invention constrains the overlapping regions of different detectors in the optical system and performs correlation detection of these overlapping regions under these constraints to obtain the optimal overlapping region. This avoids problems such as insufficient feature points and changes in code value quantification relationships, and enables effective inter-module grayscale balancing for images of different complex terrain features across multiple modules, exhibiting better adaptability. Attached Figure Description

[0017] Figure 1 This is a flowchart of the ultra-large field-of-view infrared image non-uniformity correction method of the present invention;

[0018] Figure 2 This is a schematic diagram of grayscale balancing for different detector modules. Detailed Implementation

[0019] The present invention will now be explained and described in detail with reference to the accompanying drawings.

[0020] The main technical solutions of this invention are as follows: Figure 1 As shown, an adaptive method for correcting non-uniformity in ultra-large field-of-view infrared images mainly includes the following steps:

[0021] Step 1: Initialize and obtain the calibration model; obtain the initial calibration parameters and calibration model using radiation calibration data from a ground-based laboratory under low-temperature conditions;

[0022] Step 2, Initial calibration parameter acquisition: Using the statistics of multiple original images in orbit, new initial calibration parameters are obtained based on the initial calibration model in Step 1;

[0023] Step 3, Image Initialization and Correction: Using the initial correction parameters and initial correction model from Step 2, input the single-scene image to be corrected for initialization and correction;

[0024] Step 4: Standard reference element selection: Select multiple detector linear array centers with stable responses as standard reference elements by comparing calibration data and raw statistical data;

[0025] Step 5, Pixel-by-pixel statistical sample selection: For the image initialized and corrected in Step 3, samples of adjacent pixels are adaptively selected on both sides of the standard reference pixel.

[0026] Step 6: Generate new correction parameters: Perform statistical regression on the original sample data in the images corresponding to the adjacent pixels selected in Step 4 to achieve sample matching and obtain new correction coefficients.

[0027] Step 7: Generate a single detector module correction image: Use the new correction coefficients to perform non-uniformity correction on other images corresponding to a single detector, thus completing the non-uniformity correction of the single detector module image.

[0028] Step 8: Obtaining Pixel Registration Points from Multiple Detector Modules: Using the initial registration overlap region and its registration feature points obtained in the laboratory among multiple detector modules, for the image to be registered, a new high-precision registration overlap region and registration feature points are calculated based on the constraints of the camera optical system.

[0029] Step 9: Gray-scale balancing of multiple detector modules in ultra-wide field-of-view infrared images: Based on the registration overlap region pixels and registration feature points obtained in Step 8, the non-uniformity correction coefficients between multiple detector modules are calculated using the sample selection algorithm and matching method in Steps 5 and 6, achieving rapid balancing of ultra-wide field-of-view infrared images. Figure 2 As shown.

[0030] Step 10: Scene-based adaptive dynamic stretching of the entire image: For the balanced infrared image, dynamic stretching is performed based on histogram statistics of different scenes to achieve a better visual effect.

[0031] Furthermore, in step 1, using the calibration data, the least squares method is employed to obtain the correction model and correction coefficients. The correction model is as follows:

[0032] X′=k0×X+b0 (1)

[0033] Where X is the DN value of a certain pixel in the camera image, and X′ is the DN value of the corresponding pixel after non-uniformity correction. k0 is the correction slope coefficient (related to the response rate of the pixel), and b0 is the correction intercept coefficient.

[0034] In step 2, the histogram statistical algorithm for multiple images is used to calculate the new initial non-uniformity correction coefficients k1 and b1 in Formula 1.

[0035] For steps 3, 4, and 5, the initial correction coefficients are mainly used to select samples in the image that will be used to calculate the new correction coefficients.

[0036] Valid samples are selected in two steps based on adjacent pixel image data:

[0037] The first step is the formula as follows:

[0038]

[0039]

[0040] In Formula 2, x std_i std_i represents the grayscale value of the i-th reference pixel in the image, and there are n frames in total; x i dx represents the grayscale value of the i-th frame of adjacent pixels; i This represents the difference in grayscale value between the reference pixel and its neighboring pixels in the i-th frame; The standard deviation distribution of the gray value difference between the reference pixel and all imaging frames of neighboring pixels; This represents the average grayscale value of all imaging frames of adjacent pixels; This represents the average grayscale value of all imaging frames of the reference pixel; This represents the difference between the average grayscale value of a neighboring pixel and the reference pixel across all imaging frames. When the region is identified as a relevant region, it can be used as a sample point selected in the first step of screening.

[0041] The second step is as follows:

[0042]

[0043] Among them, (k 1,i b 1,i )and These represent the initial correction coefficients for neighboring pixel i and the reference pixel std_i, respectively. C is the difference in grayscale values ​​between neighboring pixel i and the reference pixel std_i after initial correction. Theoretically, for a uniform object, when C≈0, it indicates that the correction coefficients have high accuracy. Assuming that the object information obtained by adjacent pixels in the image is not significantly different, based on this assumption, samples with small absolute differences in the C values ​​of adjacent pixels after initial correction are suitable for calculating the new coefficients, while samples with large absolute differences in the C values ​​need to be discarded.

[0044] Furthermore, the histogram distribution of the difference C is obtained. The point with the highest probability of occurrence in the histogram corresponds to a value of C0, and its probability of occurrence is taken as P0. Samples with a probability greater than 0.5*P0 are selected from the image as sample points filtered in the second step.

[0045] The intersection of the sample points selected in the above two steps is taken as the actual effective sample points.

[0046] In steps 5 and 6, based on the correction model, new correction coefficients (k2, b2) are calculated by matching the selected specific samples using statistical methods. The calculation formulas are shown in formulas 4 and 5, completing the non-uniformity correction of the image corresponding to a single detector module.

[0047]

[0048] X′=k2×X+b2 (5)

[0049] Among them, yy i , Here, xx represents the grayscale value of the i-th valid sample selected from the standard reference element, and the mean grayscale value of all valid samples. i , is the gray value of the i-th valid sample selected from adjacent pixels, and the mean gray value of all valid samples; k2 is the slope coefficient of the new non-uniformity correction, and b2 is the intercept coefficient of the new non-uniformity correction.

[0050] In steps 8 and 9, initial registration points are selected based on the constraints of the camera optical system. The optimal correlation region is then calculated at these initial registration points using correlation calculation formulas to determine the new registration pixels, as shown in formulas 6 and 7. After obtaining the latest registration pixels for the image, the two modules to be registered are treated as equivalent to individual pixels in steps 5 and 6. The same sample selection and statistical regression algorithms are used to perform statistical calculations on the modules as a whole, obtaining the non-uniformity correction coefficients between adjacent detector modules. Analogous to the pixel-by-pixel infrared non-uniformity correction method, the non-uniformity correction coefficients generated by the inter-module balancing are applied to the pixels of other detector modules to be registered, thus completing the grayscale balancing between multiple detector modules.

[0051]

[0052] Where, x n With y n represents the pixel values ​​of the overlapping region in the module to be registered and the pixel values ​​of the overlapping region in the reference module, respectively; N is the total number of pixels in the overlapping region; p0 is the intercept value of the image to be registered after the overlapping region is balanced, and p1 is the slope value of the image to be registered after the overlapping region is balanced. and This refers to the intercept and slope in the two-point correction of the image to be registered, and R is the similarity of the overlapping region. By fine-tuning the calibration region, the region with the highest similarity (maximum R) can be found, and the algorithms in steps 5 and 6 can be performed analogously.

[0053] After non-uniformity correction and inter-module balancing, to improve the visibility of the region of interest (ROI), linear stretching is first performed based on Equation 8, where γ = 1; then, adaptive correction is performed on the actual scene based on Equation 9, ultimately enhancing the features of the ROI. The dynamic stretching model is as follows:

[0054]

[0055] γ=-0.3×(I AVERAGE (9)

[0056] In formulas 8 and 9, I represents the pixel value of the original image. min I is the minimum value among all pixels. max H represents the maximum value among all pixels, and L represents the upper and lower limits of the specified stretched grayscale value, respectively. This grayscale value is obtained by statistically analyzing the distribution of grayscale values ​​in the entire image and adaptively performing the calculation. AVERAGE γ is the average pixel value of the image after linear stretching, and γ is the correction parameter for adaptive Gamma correction.

[0057] Since the non-uniformity of the entire image is largely affected by the differences in actual ground features, this invention fully considers these differences. Through sample selection rules, samples affected by ground features are effectively removed, thereby reducing the influence of ground feature factors in a single scene image. The sample selection rules and correction model fully consider the features in images of different scenes, making them more adaptable and facilitating the rapid selection of samples in a scene to achieve non-uniformity correction of a single scene image.

[0058] Due to the limitations of insufficient features in overlapping regions and high requirements for code value quantification in ultra-large field-of-view images, it is difficult to perform grayscale balancing of images from multiple detector modules using conventional methods. This invention constrains the overlapping regions of different detectors in the optical system and performs correlation detection of these overlapping regions under these constraints to obtain the optimal overlapping region. This avoids problems such as insufficient feature points and changes in code value quantification relationships, and enables effective inter-module grayscale balancing for images of different complex terrain features across multiple modules, exhibiting better adaptability.

[0059] This invention proposes an adaptive method for correcting non-uniformity in ultra-large field-of-view infrared images. Addressing the challenge of correcting non-uniformity in ultra-large field-of-view infrared images sensitive to changes in ground object temperature and background, the method employs a specific sample matching technique to correct the correction parameters of each pixel in the detector, thus completing the non-uniformity correction of a single detector module. Then, based on the infrared camera's optical system, a high-precision correlation detection constraint model for overlapping regions of adjacent detectors is established. Feature point matching correction is performed using overlapping pixels from adjacent detectors to obtain the correction parameters between adjacent detector modules, which are then sequentially passed to achieve grayscale balancing of multiple detector modules in the infrared camera. Finally, based on specific application scenarios, histogram statistics are used to achieve adaptive dynamic stretching of the entire field-of-view image, realizing the non-uniformity correction of the entire image.

[0060] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. An adaptive method for correcting non-uniformity in ultra-large field-of-view infrared images, characterized in that, Includes the following steps: Step 1: Initialize and obtain the calibration model; obtain the initial calibration parameters and calibration model using radiation calibration data from a ground-based laboratory under low-temperature conditions; Step 2, Initial calibration parameter acquisition: Using the statistics of multiple original images in orbit, new initial calibration parameters are obtained based on the initial calibration model in Step 1; Step 3, Image Initialization and Correction: Using the initial correction parameters and initial correction model from Step 2, input the single-scene image to be corrected for initialization and correction; Step 4: Standard reference element selection: Select multiple detector linear array centers with stable responses as standard reference elements by comparing calibration data and raw statistical data; Step 5, Pixel-by-pixel statistical sample selection: For the image initialized and corrected in Step 3, samples of adjacent pixels are adaptively selected on both sides of the standard reference pixel. For steps 3, 4, and 5, the initial correction coefficients are used to filter the samples in the image that will participate in the calculation of the new correction coefficients. Valid samples are selected in two steps based on adjacent pixel image data: The first step is the formula as follows: （2） In formula 2, For the reference pixel, the grayscale value of the i-th frame, The reference element is represented by n frames in the image. The grayscale value of the i-th frame of the adjacent pixels; This represents the difference in grayscale value between the reference pixel and its neighboring pixels in the i-th frame; The standard deviation distribution of the gray value difference between the reference pixel and all imaging frames of neighboring pixels; This represents the average grayscale value of all imaging frames of adjacent pixels; This represents the average grayscale value of all imaging frames of the reference pixel; This represents the difference between the average grayscale value of a neighboring pixel and the reference pixel across all imaging frames. When the region is identified as a relevant region, it can be used as the sample point selected in the first step. The second step is as follows: （3） in, and Corresponding to neighboring i and reference pixels respectively The initial correction coefficients; For neighboring i and reference pixels The difference in grayscale values ​​obtained after initial correction; theoretically, for a uniform object, when This indicates that the correction coefficient has high accuracy. Assuming that adjacent pixels acquire little difference in object information in the image, based on this assumption, after initial correction, adjacent pixels... Samples with smaller absolute value differences are suitable for calculating the new coefficients. Samples with significantly different absolute values ​​need to be removed. Obtain the histogram distribution of the difference C, and assign the value of the point with the highest probability in the histogram to the value. Its probability of occurrence is used as Select an image with a value greater than 0.5*. The samples were used as the sample points selected in the second step. The intersection of the sample points selected in the above two steps is taken as the actual effective sample points selected. Step 6: Generate new correction parameters: Perform statistical regression on the original sample data in the images corresponding to the adjacent pixels selected in Step 4 to achieve sample matching and obtain new correction coefficients. Step 7: Generate a single detector module correction image: Use the new correction coefficients to perform non-uniformity correction on other images corresponding to a single detector, thus completing the non-uniformity correction of the single detector module image. Step 8: Obtaining Pixel Registration Points from Multiple Detector Modules: Using the initial registration overlap region and its registration feature points obtained in the laboratory among multiple detector modules, for the image to be registered, a new high-precision registration overlap region and registration feature points are calculated based on the constraints of the camera optical system. Step 9: Gray-scale balancing between multiple detector modules in ultra-large field-of-view infrared images: Based on the registration overlap region pixels and registration feature points between modules obtained in Step 8, the non-uniformity correction coefficients between multiple detector modules are calculated using the sample screening algorithm and matching method in Steps 5 and 6, thereby achieving rapid balancing of ultra-large field-of-view infrared images. Step 10: Scene-based adaptive dynamic stretching of the entire image: For the balanced infrared image, dynamic stretching is performed based on histogram statistics of different scenes to achieve visual effect.

2. The adaptive ultra-large field-of-view infrared image non-uniformity correction method according to claim 1, characterized in that, In step 1, using the calibration data, the least squares method is employed to obtain the calibration model and calibration coefficients. The calibration model is as follows: （1） in, This refers to the DN value of a specific pixel in the camera image. This represents the DN value of the corresponding pixel after non-uniformity correction. The slope coefficient, which is related to the responsivity of the pixel, is used for correction. To correct the intercept coefficient.

3. The adaptive ultra-large field-of-view infrared image non-uniformity correction method according to claim 1, characterized in that, In step 2, a histogram statistical algorithm for multiple images is used to calculate the new initial non-uniformity correction coefficients in Formula 1. and .

4. The adaptive ultra-large field-of-view infrared image non-uniformity correction method according to claim 1, characterized in that, In steps 5 and 6, based on the correction model, statistical methods are used to match the selected specific samples and calculate new correction coefficients. ; The calculation formulas are shown in formulas (4) and (5) to complete the non-uniformity correction of the image corresponding to a single detector module; （5） in, , The grayscale value of the i-th valid sample selected from the standard reference element, and the mean grayscale value of all valid samples; , The gray value of the i-th valid sample selected from adjacent pixels, and the mean gray value of all valid samples; The slope coefficients for the new non-uniformity correction. The intercept coefficient is used for the new non-uniformity correction.

5. The adaptive ultra-large field-of-view infrared image non-uniformity correction method according to claim 1, characterized in that, In steps 8 and 9, an initial registration point is selected based on the constraints of the camera optical system, and the optimal correlation region is calculated at the initial registration point using the correlation calculation formula to determine the new registration pixel. Using formulas (6) and (7), after obtaining the latest registration pixel of the image, the two modules to be registered are equivalent to a single pixel in steps 5 and 6. The same sample screening and statistical regression algorithm is used to perform statistical calculations on the module as a whole to obtain the non-uniformity correction coefficient between adjacent detector modules. Analogous to the infrared non-uniformity correction method of pixel-by-pixel transmission, the non-uniformity correction coefficient generated by the balancing between modules is applied to the pixels of other detector modules to be registered, thus completing the gray-scale balancing between multiple detector modules. （6） （7） in, and These represent the pixel values ​​of the overlapping region in the module to be registered and the pixel values ​​of the overlapping region in the reference module, respectively. This represents the total number of pixels in the overlapping region. The image to be registered after balancing the calculated overlapping area. The intercept characterization value, The image to be registered after balancing the calculated overlapping area The slope characterization value, and That is, the image to be registered Intercept and slope in two-point correction This refers to the similarity of the overlapping regions; by fine-tuning the calibrated regions, the regions with the highest similarity can be found, which can be analogous to the algorithms in steps 5 and 6.

Images