Infrared image column stripe non-uniformity correction method based on improved median histogram

Abstract

The invention discloses an infrared image column stripe heterogeneity correction method based on an improved median histogram, and belongs to the field of infrared image heterogeneity correction. The method comprises the following steps: calculating a column histogram of a single-frame infrared image and a column mean value curve subjected to wavelet denoising according to an input infrared image; performing inversion operation according to the obtained column histogram to obtain a column inverse histogram; generating an adaptive window of the bilateral filter according to the column mean value curve; generating a bilateral filter according to the spatial domain distance of the column and the gray domain difference of the column mean value curve; filtering the column inverse histogram according to the obtained bilateral filter to obtain a median histogram; and an input infrared image is mapped according to the median histogram to obtain an infrared image after stripe removal. According to the method, a better fringe removal effect is achieved, blind pixel points are avoided, and the step of repeatedly calculating a comparison result to solve an optimal parameter value in an original algorithm is omitted.

Description

Technical Field

[0001] The invention belongs to the field of infrared image non-uniformity correction, and in particular relates to an infrared image column stripe non-uniformity correction method based on an improved median histogram. Background Art

[0002] Ideally, the response characteristic curve of each detection unit in the infrared imaging system and the readout amplification characteristics of its readout circuit should be exactly the same, that is, when irradiated by a uniform heat source, the output signal intensity of each detection unit should be exactly the same. However, due to factors such as the material and manufacturing process of the device, the ambient temperature in the application scenario, noise and messy radiation, etc., the response rate of each detection unit cannot be completely consistent, which ultimately leads to non-uniform noise or fixed pattern noise in the generated infrared image. Especially when different channels of the readout circuit are non-uniform, this leads to obvious brightness differences between two adjacent rows or columns of the output image, resulting in stripe non-uniform noise. The direction of the stripe noise is related to the readout method of the pixel. If the readout circuit of the same row of pixels shares the same channel, stripes in the horizontal direction will be generated. If the readout circuit of the same column of pixels shares the same channel, stripes in the vertical direction will be generated. The existence of stripe non-uniform noise greatly reduces the quality of the infrared image, which not only reduces the visibility of the image, but also has an adverse effect on subsequent processing tasks such as target detection and recognition. In order to improve the quality of infrared images and eliminate these stripe noises, non-uniformity correction is usually required.

[0003] The non-uniformity correction algorithm can be divided into two categories according to whether a calibration radiation source such as a black body is required: one is the calibration method, which can be divided into single-point correction, two-point correction, and multi-point correction based on multiple calibration temperature points. The calibration method requires the use of a uniformly radiating black body during the correction process to collect image data of the black body at different temperatures. Since the black body radiation is uniform, the global mean of the black body image is used as the theoretical response true value. Therefore, the response curve parameters of each detection element can be obtained by fitting these calibration data. These parameters can be directly stored in the imaging system and directly called when imaging. The calibration method has a good correction effect and is very simple to use and can meet real-time requirements. Therefore, many real-time correction imaging systems in practical applications use calibration methods for non-uniformity correction. However, when the use environment of the infrared imaging system changes, the pixel response characteristics on the infrared focal plane array will drift over time. At this time, the non-uniformity correction coefficient obtained by the calibration method under laboratory conditions is no longer applicable to the current situation, and the image cannot be effectively corrected for non-uniformity. If you want to obtain a better non-uniformity correction effect, you can only pause the imaging process and recalculate the correction coefficient through calibration. However, when the detector is performing target detection, tracking or guidance tasks, pausing the imaging process is likely to cause the loss of the target, seriously affecting its use. The other type is scene-based methods, which are mainly divided into four categories: statistical non-uniformity correction methods, filtering-based non-uniformity correction methods, optimization-based non-uniformity correction methods, and learning-based non-uniformity correction methods. The scene method obtains the non-uniformity correction coefficient of each detection pixel according to the captured image during the imaging process. This method gets rid of calibration radiation sources such as black bodies, and can adaptively track changes in the response characteristics of pixels, so it has broader application prospects.

[0004] According to the number of image frames used in the correction process, it can also be divided into: a single-frame based non-uniformity correction method and a multi-frame based non-uniformity correction method. In the application process, since the multi-frame based non-uniformity correction method needs to use the spatiotemporal information in the continuous frame images, it requires certain assumptions on the scene motion, such as the randomness of the scene motion. When the imaging scene does not meet these assumptions, ghosting and artifacts will appear in the corrected image, which greatly affects the image quality. The single-frame based non-uniformity correction method can generally avoid ghosting and artifacts caused by scene changes because it only uses a single frame image to complete the non-uniformity correction. The infrared image non-uniformity correction algorithm based on the median histogram is based on the fact that a single column (or row, depending on the readout system) of the image itself carries enough information for histogram equalization. Because the image is continuous, the difference between two adjacent columns is statistically small, which means that the histograms of the two adjacent columns are almost equal, and because the stripe non-uniformity is usually independent. Therefore, by transforming the histogram of each column (or row) to the median histogram of the neighborhood column (or row) histogram, the stripe non-uniformity noise can be eliminated. The infrared image non-uniformity correction algorithm based on the median histogram has the advantages of simple principle, low complexity, and strong ability to retain image details. However, in practice, since the traditional infrared image stripe non-uniformity correction algorithm based on the median histogram uses a fixed window determined by the standard deviation in Gaussian filtering, considering that the histogram characteristics of the neighborhood columns are not similar in the details and edges of the infrared image, the fixed window will reduce the accuracy of the median histogram, resulting in an increase in correction residuals and blind pixels, and the traditional algorithm needs to traverse the parameters within a parameter range according to the minimum variation principle to determine the parameters, which causes a waste of computing resources. Summary of the invention

[0005] Aiming at the problem that the original infrared image non-uniformity correction algorithm based on median histogram easily leads to correction residuals in scene details and edges, blind pixels are generated due to mapping, and computing resources are wasted, the present invention proposes an infrared image column stripe non-uniformity correction method based on improved median histogram, which optimizes the infrared image non-uniformity correction algorithm based on median histogram by introducing column mean curve denoised by one-dimensional wavelet, adaptive filtering window and bilateral filter that considers grayscale domain weight and distance domain weight at the same time, effectively improves the correction effect, and shows significant advantages in scene details and edges in particular.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] The method for correcting infrared image non-uniformity based on improved median histogram includes the following steps:

[0008] Step (1), based on the input infrared image Calculate the column histogram of a single frame infrared image And the column mean curve after wavelet denoising ;

[0009] Step (2), according to the correspondence between the histogram and the inverse histogram, the generated column histogram Get the column inverse histogram ;

[0010] Step (3): Set the maximum window size of the adaptive window and minimum window And according to the column mean curve Generate adaptive window for bilateral filter ;

[0011] Step (4), based on the obtained adaptive window The difference between the spatial domain distance of the column and the grayscale domain of the column mean curve generates the bilateral filter kernel matrix ;

[0012] Step (5), according to the obtained bilateral filter kernel matrix Inverse histogram Filter to get the median histogram of the column ;

[0013] Step (6) According to the median histogram Mapping is performed to obtain the infrared image after striping .

[0014] The beneficial effects of the present invention are:

[0015] When calculating the column mean curve of the infrared image, it is considered that the stripe noise changes the column mean characteristics of the real infrared image, so the column mean curve after one-dimensional wavelet denoising can better reflect the correlation between the columns of the real infrared image.

[0016] Considering the gradient of the column mean curve, the columns with drastic changes in the column mean correspond to the details and edges in the image. At this time, the correlation between the columns is poor, and a smaller window is used. On the contrary, the columns with gentle changes in the column mean correspond to the uniform area in the image. At this time, the correlation between the columns is strong, and a larger window is used. Compared with the fixed-size Gaussian filter adopted by the original algorithm, the improved algorithm generates a bilateral filter kernel matrix based on the adaptive window size, the spatial domain distance of the column number, and the gray domain difference of the column mean curve. The correction effect is better and the repeated steps of iterative calculation of the original algorithm to determine the parameters are avoided.

[0017] The present invention effectively solves the problem that the original infrared image non-uniformity correction algorithm based on the median histogram easily causes correction residuals at scene details and edges, generates blind pixels due to mapping, and easily causes waste of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of a method for correcting non-uniformity of infrared image stripes based on an improved median histogram according to the present invention;

[0019] Figure 2 Figure 1 is a correction result diagram based on the method of the present invention, wherein (a) is the original image, (b) is the correction result of the original algorithm, and (c) is the correction result of the improved algorithm;

[0020] Figure 3 Figure 3 is a corrected stripe non-uniform noise diagram based on the method of the present invention, wherein (a) is the correction result diagram of the original algorithm, (b) is the correction result diagram of the improved algorithm, (c) is the enlarged diagram of the correction result of the original algorithm, and (d) is the enlarged diagram of the correction result of the improved algorithm. DETAILED DESCRIPTION

[0021] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0022] like Figure 1 As shown, the infrared image column stripe non-uniformity correction method based on improved median histogram provided by the present invention comprises the following steps:

[0023] Step 1: Based on the input infrared image Computed column histogram and the column mean curve after wavelet denoising of the one-dimensional curve The method is as follows:

[0024] Infrared image The row number of any row is denoted as m, m [1,M],M 2;

[0025] Infrared image The column number of any column is n, n [1, N], N 2;

[0026] Infrared image The pixel value of any pixel point (m, n) is recorded as ;

[0027] Column Histogram The calculation method is as follows:

[0028] ,

[0029] in, It represents the nth column of the column histogram H, that is, the histogram generated by the nth column of the infrared image. l represents the grayscale level, which defaults to 0-255.

[0030] Calculate the mean of each column and perform one-dimensional wavelet denoising to remove the fluctuation of the column mean caused by the column stripes of the infrared image, so as to better restore the fluctuation of the pixel value column mean itself in the real image. The formula is as follows:

[0031] ,

[0032] For one-dimensional signals Perform wavelet transform to obtain wavelet coefficients , where a is the scale factor and b is the translation factor, the formula is as follows:

[0033] ,

[0034] in is the wavelet basis function, and t represents the integral variable.

[0035] For wavelet coefficients Apply a threshold to remove noise. Common threshold methods include hard threshold and soft threshold. Here, soft threshold denoising is used to achieve better denoising effect. The threshold is set to λ, and the coefficient after processing is for:

[0036] ,

[0037] The processed coefficients are inversely transformed to obtain the denoised column mean curve , the formula is as follows:

[0038] .

[0039] Step 2: According to the correspondence between the histogram and the inverse histogram, the column histogram generated Get the column inverse histogram The specific steps to calculate the inverse histogram are as follows:

[0040] ,

[0041] Where z represents the normalized frequency of gray level accumulation.

[0042] Step 3: Based on the column mean curve after wavelet denoising , and set the maximum window of the adaptive window and minimum window Further generate the adaptive window of the bilateral filter The specific steps to generate an adaptive window are as follows:

[0043] The maximum value of the adaptive window is recorded as: max_window,

[0044] The minimum value of the adaptive window is recorded as: min_window,

[0045] Calculate column mean curve The gradient values ​​are sorted in order of size, and the adaptive windows of the columns corresponding to the maximum 10% and the minimum 10% are set to min_window and max_window respectively.

[0046] The windows of the remaining columns are linearly stretched according to the maximum and minimum values ​​to obtain adaptive windows. .

[0047] Step 4: Adaptive window size based on the obtained The difference between the spatial domain distance of the column and the grayscale domain of the column mean curve generates the bilateral filter kernel matrix The specific steps to generate the bilateral filter kernel matrix are as follows:

[0048] Spatial domain weights:

[0049] ,

[0050] in is the variance in the spatial domain, Indicates the column number currently being processed.

[0051] Intensity domain weight:

[0052] ,

[0053] in is the variance of the intensity domain.

[0054] According to the adaptive window size , the kernel of the bilateral filter of each column is set to an adaptive window size and normalized:

[0055] ,

[0056] in For Column Neighborhood, For Column The weight value of the bilateral filter in the nth column in the neighborhood is The weight values ​​of the bilateral filter are calculated and normalized for each column, and finally the bilateral filter kernel matrix is ​​formed. .

[0057] Step 5: According to the obtained bilateral filter kernel matrix Inverse histogram Filter to get the median histogram of the column The specific steps to generate a median histogram of a column are as follows:

[0058] ,

[0059] Step 6: The original noisy image is extracted according to the median histogram Mapping is performed to obtain the infrared image after striping The specific steps to generate a de-streaked infrared image are as follows:

[0060] in Any pixel value in can be expressed as:

[0061] ,

[0062] At this point, the output de-striated infrared image is obtained.

[0063] In order to verify the effectiveness of the method, the present invention tests the real infrared image containing stripe non-uniformity. Figure 2 The original image shown in (a) is the image corrected by the original algorithm. Figure 2 As shown in (b), the image corrected by the improved algorithm is as follows Figure 2 As shown in (c), it can be seen that after the processing of the improved algorithm, the column stripes are greatly removed and the image quality is improved. Figure 3 From the correction effect of another infrared image shown, it can be seen that the improved method of the present invention overcomes the phenomenon of blind pixels generated by mapping in the original algorithm and achieves a better stripe removal effect, wherein (a) is the correction result of the original algorithm, (b) is the correction result of the improved algorithm, (c) is the enlarged image of the correction result of the original algorithm, and (d) is the enlarged image of the correction result of the improved algorithm.

[0064] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for correcting infrared image row fringe non-uniformity by improving the median histogram, characterized in that: The following steps are involved: Step (1), based on the input infrared image Calculate the column histogram of a single frame infrared image And the column mean curve after wavelet denoising ; Step (2), according to the correspondence between the histogram and the inverse histogram, the generated column histogram Get the column inverse histogram ; Step (3): Set the maximum window size of the adaptive window and minimum window , and according to the column mean curve Generate adaptive window for bilateral filter ; Step (4), based on the obtained adaptive window The difference between the spatial domain distance of the column and the grayscale domain of the column mean curve generates the bilateral filter kernel matrix ; Step (5), according to the obtained bilateral filter kernel matrix Inverse histogram Filter to get the median histogram of the column ; Step (6), input infrared image According to the median histogram Mapping is performed to obtain the infrared image after striping .

2. The infrared image column fringe non-uniformity correction method based on improved median histogram according to claim 1, characterized in that: The step (1) comprises: Infrared image The row number of any row is denoted as m, m [1,M],M 2; Infrared image The column number of any column is n, n [1, N], N 2; Infrared image The pixel value of any pixel point (m, n) is recorded as ; Column Histogram The calculation method is as follows: ， in, represents the nth column of the column histogram H, represents the gray level, [0, 255]; Compute the mean of each column: ， For one-dimensional signals Perform wavelet transform to obtain wavelet coefficients ,in is the scale factor, is the translation factor: ， in is the wavelet basis function, t is the integral variable; For wavelet coefficients Apply soft threshold denoising and set the threshold to , the coefficient after processing for: ， The processed coefficients are inversely transformed to obtain the denoised column mean curve : 。 3. The infrared image row fringe non-uniformity correction method based on improved median histogram according to claim 2, characterized in that: The column inverse histogram is calculated in step (2) The method is as follows: ， Where z represents the normalized frequency of gray level accumulation.

4. The infrared image row fringe non-uniformity correction method based on improved median histogram according to claim 3, characterized in that: The step (3) comprises: Compute the denoised column mean curve The gradient values ​​are sorted in order of size, and the adaptive windows of the columns corresponding to the largest 10% and the smallest 10% are set to and ; The windows of the remaining columns are linearly stretched according to the maximum and minimum values ​​to obtain adaptive windows. .

5. The infrared image row fringe non-uniformity correction method based on improved median histogram according to claim 4, characterized in that: The step (4) comprises: Calculate spatial domain weights: ， in is the variance in the spatial domain, Indicates the column number currently being processed; Calculate the intensity domain weight: ， in is the variance of the intensity domain; Set the kernel of the bilateral filter of each column to the adaptive window size and perform normalization: ， in For Column Neighborhood, For Column The weight value of the bilateral filter used in the nth column of the neighborhood, a total of Calculate the weight value of the bilateral filter for each column and normalize it to form the bilateral filter kernel matrix .

6. The infrared image row stripe non-uniformity correction method based on improved median histogram according to claim 5, characterized in that: In step (5), the median histogram of the column The Any pixel value in the column is expressed as: 。 7. The infrared image row fringe non-uniformity correction method based on improved median histogram according to claim 1, characterized in that: In step (6), the infrared image after stripe removal Any pixel value in is represented as: 。

Images