A single-frame infrared image stripe non-uniformity correction method based on gradient domain weighted guided filtering

Abstract

The application discloses a single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering, and comprises the following steps: one-dimensional row guided filtering is used to separate an original infrared image in a horizontal direction, and the original infrared image is decomposed into a low-frequency smooth component and a high-frequency component; the original infrared image is quantized, and an adaptive model is used to determine the optimal filter window size; gradient domain weighted guided filtering is used to extract the edge structure of the infrared image, and the edge structure is superimposed on the smooth component to obtain a corrected infrared image. The application can realize stripe non-uniformity correction in a single-frame range for different infrared scenes, effectively ensure the image texture details, better remove the stripe non-uniformity, and significantly improve the infrared image quality and correction performance.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of infrared focal plane array imaging, and relates to a single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering. BACKGROUND

[0002] As a key technology in the actual application process of infrared radiation, infrared imaging technology is different from visible light imaging. Infrared imaging belongs to passive imaging, does not need to rely on other radiation sources, has strong concealment, and can still normally detect targets even in harsh environments due to less absorption of infrared radiation by the atmosphere. Therefore, infrared imaging technology is less affected by external factors, and is widely used in military and civilian fields, including infrared reconnaissance, resource exploration, petroleum and chemical industry, metallurgical manufacturing, fire fighting, temperature measurement and medical treatment and the like.

[0003] As the core of infrared imaging technology, infrared focal plane array (IRFPA) produces non-uniformity in the output infrared image due to the limitations of production process and manufacturing materials and the influence of thermal environment changes, which seriously affects the imaging quality and the extraction of useful information. In the infrared image produced by the staring infrared focal plane array, stripe non-uniformity is generally present. Bias voltage noise in the IRFPA amplification circuit can affect the response of the entire column of pixels, thereby causing stripe-shaped fixed pattern noise (FPN), which is usually referred to as stripe non-uniformity.

[0004] In order to solve this problem, domestic and foreign scholars have carried out extensive research on the image processing algorithm of the infrared imaging system, and have proposed many algorithms with remarkable effects. However, there are still some problems in the non-uniformity correction algorithm of the infrared image at present.

[0005] (1) The non-uniformity correction algorithm based on radiation source calibration needs to repeatedly calibrate the correction parameters when the external environment and the temperature of the infrared imaging system itself change, which affects the continuity of work, and therefore has defects in the fields of military reconnaissance, monitoring and detection and the like. The non-uniformity correction algorithm based on radiation source calibration without a shutter needs to go through a complex calibration, has poor real-time performance, and is difficult to achieve ideal correction results.

[0006] (2) The non-uniformity correction algorithm based on a scene has self-adaptive correction capability, but excessively relies on the randomness of moving sequence images and target scenes. For a stationary target scene, this kind of algorithm will mistakenly compensate the scene information as non-uniformity, causing a "ghost" phenomenon. At the same time, the convergence speed is slow, and the real-time performance and non-uniformity correction efficiency are insufficient.

[0007] Therefore, a single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering is needed to solve the above problems. SUMMARY

[0008] The application aims to provide a single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering, which can realize stripe non-uniformity correction in a single frame for different infrared scenes.

[0009] Technical scheme: To achieve the above-mentioned purpose, the single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering comprises the following steps:

[0010] Step one: one-dimensional row guided filtering is used to separate the original infrared image in the horizontal direction, and the original infrared image is decomposed into a low-frequency smooth component and a high-frequency component;

[0011] Step two: the original infrared image is quantized, and based on the quantized parameters, an adaptive model is constructed to determine the optimal filter window size;

[0012] Step three: based on the filter window size provided in step two, the low-frequency smooth component and the high-frequency component obtained in step one are taken as the input of the gradient domain weighted guided filtering, the edge structure of the infrared image is extracted, and the extracted image edge texture information is superimposed on the low-frequency smooth component to obtain the corrected infrared image.

[0013] Further, the original infrared image in step one is an infrared image generated by a refrigeration type infrared focal plane array.

[0014] Further, the original infrared image in step one is an infrared image with stripe non-uniformity.

[0015] Further, the original infrared image in step one is separated in the horizontal direction, comprising the following steps:

[0016] S11, an infrared image model with stripe non-uniformity is established, which is represented as follows:

[0017] p i =u i +s i +t i ;

[0018] In the formula, p i represents the original infrared image, u i represents the low-frequency smooth component, s i represents the stripe non-uniformity, and t i represents the high-frequency component of the image texture information.

[0019] S12, the original infrared image p i is taken as the input image and the guide image of one-dimensional row guided filtering, and the regularization parameter ∈ is set to 0.4 2With a window size of 1×9, the original infrared image is horizontally separated into low-frequency smoothing components and high-frequency components, as shown in the following formula:

[0020] p i =u i +n i ;

[0021] In the formula, n i It includes high-frequency components such as stripe noise and image texture information.

[0022] Furthermore, in step two, the stripe non-uniformity and image texture information of the original infrared image are quantified using the root mean square difference (RMSE) and the local contrast evaluation factor (EME), respectively, as expressed by the following formula:

[0023]

[0024]

[0025] In the formula, the original infrared image size is M×N, f(x,y) represents the pixel value of the pixel at coordinates (x,y), and k1 and k2 represent the number of blocks in the vertical and horizontal directions of the original infrared image, respectively, and are set as k1=floor(M / 8) and k2=floor(N / 8) respectively. These represent the maximum and minimum grayscale values ​​of pixels within the image block, respectively.

[0026] Furthermore, the energy function of the gradient-domain weighted guided filter described in step three is as follows:

[0027]

[0028] In the formula, (a k b k ω represents the linear coefficients within the filter window when the center of the local window is located at k. k Represents the local window of the filter, u i Represents the guiding image, n i Indicates the input image. Let g(k) represent the edge-aware constraint, g(k) represent the edge-protection constraint, and ∈ represent the regularization parameter.

[0029] Furthermore, the linear coefficients of the gradient-domain weighted guided filter are as follows:

[0030]

[0031] In the formula, μ k and Representing the guiding image u i In the local filter window ωk The mean and variance in the data. Indicates the input image n i With guide image u i In the local filter window ω k covariance in Indicates the input image n i In the local filter window ω k The mean value in the filter, |ω| represents the local window value ω. k The total number of pixels in the image.

[0032] Furthermore, the gradient domain weighted guided filter's edge-aware constraint As shown below:

[0033]

[0034] in

[0035]

[0036] In the formula, E(k) represents the normalization term. Represents the horizontal gradient of an infrared image. γ represents the absolute difference between the horizontal gradient information of a pixel and its neighboring pixels, γ is the regularization parameter, ω1 and ω2 represent different window radii, which are set to 2 and 20 respectively, and N0 represents the total number of pixels in the infrared image.

[0037] Furthermore, the edge protection constraint g(k) of the gradient domain weighted guided filter is as follows:

[0038]

[0039] in:

[0040]

[0041] In the formula, This represents the edge-aware constraint, and N0 represents the total number of pixels in the infrared image.

[0042] Furthermore, the low-frequency smoothing component and high-frequency component obtained in step one are used as the guide image and input image for gradient domain weighted guided filtering in step three, respectively, and the filter window size is given in step two.

[0043] Beneficial effects: The present invention has the following advantages: The single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering of the present invention can achieve stripe non-uniformity correction within a single frame range for different infrared scenes, effectively ensuring image texture details while better removing stripe non-uniformity, and significantly improving infrared image quality and correction performance. Attached Figure Description

[0044] Figure 1 This is a flowchart of the single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering according to the present invention;

[0045] Figure 2 This is a flowchart illustrating the specific algorithm for correcting stripe non-uniformity in single-frame infrared images based on gradient domain weighted guided filtering, as described in this invention.

[0046] Figure 3 This invention provides a window adaptive model for the single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering.

[0047] Figure 4 This is the window adaptive model fitting function expression for the single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering in this invention. Detailed Implementation

[0048] The technical solution of the present invention will be described in detail below with reference to the embodiments and accompanying drawings.

[0049] Please see Figure 1 , Figure 2 , Figure 3 and Figure 4 As shown, the single-frame infrared image stripe non-uniformity correction algorithm based on gradient domain weighted guided filtering of the present invention includes the following steps:

[0050] S1. An infrared image with striped non-uniformity is generated using a cooled infrared focal plane array. A one-dimensional row-guided filter is used to separate the original infrared image in the horizontal direction, decomposing the original infrared image into low-frequency smooth components and high-frequency components.

[0051] S11. Establish an infrared image model with striped non-uniformity, which is represented by the following formula:

[0052] p i =u i +s i +t i

[0053] S12, using the original infrared image p i The input and guide images for the one-dimensional row guided filter are set with a regularization parameter of ∈ = 0.4. 2 With a window size of 1×9, the original infrared image is horizontally separated into low-frequency smoothing components and high-frequency components, as shown below:

[0054] p i =u i +n i

[0055] S2. Quantize the original infrared image and use the constructed adaptive model to determine the optimal filter window size;

[0056] S21. The stripe non-uniformity and image texture information of the original infrared image are quantified by the root mean square error (RMSE) and the local contrast evaluation factor (EME), respectively, and expressed by the following formula:

[0057]

[0058]

[0059] S22, Input the quantized parameters into Figure 3 , Figure 4 In the window adaptive model shown, the optimal window size is obtained.

[0060] S3. Use gradient domain weighted guided filtering to extract the edge structure of the infrared image and superimpose it onto the smoothing component to obtain the corrected infrared image.

[0061] S31. The low-frequency smoothing component and high-frequency component obtained in step one are used as the guide image and input image of the gradient domain weighted guided filtering in step three, respectively. The size of the filtering window is given in step two.

[0062] S32. The energy function of gradient-domain weighted guided filtering is shown below:

[0063]

[0064] To better preserve image information, the energy function E(a) is made more efficient. k b k By obtaining the minimum value, the linear coefficients can be obtained as follows:

[0065]

[0066] in, g(k) and g(k) represent edge-aware constraints and edge-preserving constraints, respectively, with the aim of better distinguishing edge texture information from stripe non-uniformity in an image and improving correction performance, as shown below.

[0067]

[0068] in:

[0069]

[0070]

[0071] in:

[0072]

[0073] S33. The image edge texture information extracted by gradient domain weighted guided filtering is superimposed on the smoothing component to obtain the corrected infrared image.

Claims

1. A method for correcting stripe non-uniformity in single-frame infrared images based on gradient domain weighted guided filtering, characterized in that, Includes the following steps: Step 1: Use one-dimensional row guided filtering to perform horizontal separation on the original infrared image, decomposing the original infrared image into low-frequency smooth components and high-frequency components. Step 2: Quantize the original infrared image, and determine the optimal filter window size based on the quantized parameters using the constructed adaptive model; Step 3: Based on the filtering window size provided in Step 2, the low-frequency smoothing component and high-frequency component obtained in Step 1 are used as inputs to gradient domain weighted guided filtering to extract the edge structure of the infrared image. The extracted image edge texture information is then superimposed on the low-frequency smoothing component to obtain the corrected infrared image. The energy function of the gradient-domain weighted guided filter is shown below: ； In the formula, This represents the linear coefficients within the filter window when the center of the local window is located at k. Indicates the local window of the filter. Indicates a guide image. Indicates the input image. This represents edge-aware constraints. Indicates edge protection constraints, Represents the regularization parameter; The linear coefficients of the gradient-domain weighted guided filter are shown below: ， ； In the formula, and These represent the guide images. In the local filtering window The mean and variance in the data. Indicates the input image With guide image In the local filtering window covariance in Indicates the input image In the local filtering window The mean of the middle, Represents the local window of the filter The total number of pixels in; The edge-aware constraint of the gradient-domain weighted guided filter As shown below: ， in ； In the formula, Represents the normalization term. Represents the horizontal gradient of an infrared image. This represents the absolute difference in horizontal gradient information between a pixel and its neighboring pixels. For regularization parameters, , These represent different window radius sizes, set to 2 and 20 respectively. This represents the total number of pixels in an infrared image. The edge protection constraint of the gradient domain weighted guided filter As shown below: ， in: ； In the formula, This represents edge-aware constraints. This represents the total number of pixels in an infrared image.

2. The method for correcting stripe non-uniformity in single-frame infrared images based on gradient domain weighted guided filtering according to claim 1, characterized in that, The original infrared image mentioned in step one is an infrared image generated by a cooled infrared focal plane array.

3. The method for correcting stripe non-uniformity in a single-frame infrared image based on gradient domain weighted guided filtering according to claim 1, characterized in that: The original infrared image mentioned in step one is an infrared image with non-uniform stripes.

4. The method for correcting stripe non-uniformity in single-frame infrared images based on gradient domain weighted guided filtering according to claim 1, characterized in that, The horizontal separation of the original infrared image described in step one includes the following steps: S11. Establish an infrared image model with striped non-uniformity, which is represented by the following formula: ； In the formula, Represents the original infrared image. Indicates the low-frequency smoothing component. Indicates non-uniformity of stripes. High-frequency components representing image texture information; S12, using the original infrared image The input and guide images are used as the input and guide images for one-dimensional row guided filtering, and regularization parameters are set. Window size is The original infrared image is separated into low-frequency smoothing components and high-frequency components in the horizontal direction, as shown below: ； In the formula, It represents the high-frequency components that include stripe noise and image texture information.

5. The method for correcting stripe non-uniformity in single-frame infrared images based on gradient domain weighted guided filtering according to claim 1, characterized in that, The method for quantizing the original infrared image in step two is as follows: the stripe non-uniformity and image texture information of the original infrared image are quantified by the root mean square error (RMSE) and the local contrast evaluation factor (EME), respectively, as expressed by the following formula: ， ； In the formula, the original infrared image size , Indicates coordinates as The size of the pixel value of the pixel. , These represent the number of blocks in the vertical and horizontal directions of the original infrared image, respectively, and are set sequentially as follows: , , , These represent the maximum and minimum grayscale values ​​of pixels within the image block, respectively.

6. The method for correcting stripe non-uniformity in a single-frame infrared image based on gradient domain weighted guided filtering according to claim 1, characterized in that, The low-frequency smoothing component and high-frequency component obtained in step one are used as the guide image and input image for gradient domain weighted guided filtering in step three, respectively.

Images