Infrared image cross grain removing method

Abstract

The invention belongs to the technical field of image processing, and particularly relates to an infrared image cross grain removal method. According to the method, a correction method based on intra-frame statistics is adopted, an even-numbered line is used as a brightness reference, correction is carried out on an odd-numbered line, the gray value difference between the odd-numbered line and the even-numbered line is calculated, and corresponding compensation is carried out on the odd-numbered line, so that the brightness of the odd-numbered line is kept consistent with that of the even-numbered line; therefore, the cross grain phenomenon caused by the difference between the odd and even row reading circuits is eliminated; according to the method, the column mean value array in the vertical direction is adopted to adjust the local difference image, the problem of sawtooth dislocation of the target edge after the image is compensated is avoided, and it is guaranteed that the image quality is not affected while cross grains are removed; according to the method, each frame of image is independently compensated, the method adapts to dynamic scenes, meanwhile, the calculated amount is small, and the method is suitable for real-time processing.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and specifically relates to a method for removing horizontal stripes from infrared images. Background Technology

[0002] To improve data readout speed, optimize circuit timing control, and reduce signal interference risks, high-resolution uncooled infrared detectors typically employ parallel output for odd and even rows in their readout circuits. This means two independent readout circuits are used to handle the output of odd and even rows of data, respectively. However, due to manufacturing limitations, these two readout circuits may differ in their device characteristics. These differences may include variations in circuit gain, noise levels, and signal processing characteristics. Consequently, when imaging the same scene, the odd and even rows, processed by readout circuits with different characteristics, will exhibit significant differences in output grayscale values. These differences can reach 200-400 grayscale levels, far exceeding the normal grayscale variation between adjacent rows (a grayscale difference of <10), resulting in noticeable horizontal stripes in the image and severely impacting image quality and readability. Especially for a scene with a uniform temperature, theoretically, the infrared radiation should be uniformly distributed across the entire surface. This translates to consistent or minimally variable grayscale values ​​for each pixel in the image, exhibiting a certain regularity. However, due to the potential nonlinearity of certain components in the readout circuit, the gain or bias parameters of the two readout circuits change inconsistently with the horizontal direction of the input signal (the electrical signal corresponding to the infrared radiation). This results in different processing effects for odd and even rows of pixels at different locations, causing inconsistent grayscale differences between adjacent odd and even rows of pixels in the same uniform temperature scene from left to right. Furthermore, there is a complex interaction between the differences in the readout circuit's device characteristics and scene variations. Since the gain, noise, and other characteristics of the readout circuit are not constant, they may be affected by the input signal (the infrared radiation signal corresponding to the scene). This complex influence means that the grayscale difference between odd and even rows at the same location is not constant, and there is no simple fixed difference relationship or a relationship that can be accurately described by weighting coefficients. Therefore, the horizontal stripe problem caused by readout circuit differences cannot be eliminated by simple calibration or compensation methods; more complex algorithms and techniques are needed to process and correct the image.

[0003] Chinese patent CN115100070A discloses a method for removing shadows from an image. This method converts an original image with horizontal stripe shadows into a grayscale image and performs low-pass filtering on the grayscale image to determine a first light field image and a second light field image. The first light field image characterizes the brightness information of each pixel in the original image after removing the horizontal stripe shadow interference, while the second light field image characterizes the brightness information of each pixel in the original image while retaining the horizontal stripe shadow interference. Based on the first and second light field images, a shadow compensation matrix is ​​determined, which records the shadow compensation coefficients corresponding to each pixel in the original image. Based on the shadow compensation matrix, shadow correction processing is performed on the original image to obtain an image with the horizontal stripe shadows removed. Chinese patent CN118552436A discloses a method and apparatus for removing stripe noise from infrared images. The method generates a first smoothed image based on the infrared image; generates first pixel mean data based on the infrared image; generates second pixel mean data based on the first smoothed image; generates vertical stripe noise data based on the first and second pixel mean data; generates a first processed image based on the infrared image and the vertical stripe noise data; generates a second smoothed image based on the first processed image; generates third pixel mean data based on the first processed image; generates fourth pixel mean data based on the second smoothed image; generates horizontal stripe noise data based on the third and fourth pixel mean images; and generates a stripe-free image based on the first processed image and the horizontal stripe noise data. In the above-mentioned prior art solutions, a light field map or smoothed map without horizontal stripe shadow interference is first generated as a statistical analysis reference image. Then, a shadow compensation matrix or stripe noise image is obtained by subtracting the original image from the reference image. Finally, shadow correction processing or stripe removal processing is performed on the original image. These technical solutions obtain statistical analysis reference images through guided filtering or other methods, representing low-noise, stripe-free images for a specific scene. However, if used as a general reference for different scenes, the obtained shadow compensation matrix coefficients or stripe noise data may show significant deviations, inevitably affecting the correction effect. Therefore, they cannot adapt to dynamic scene changes. Chinese patent CN119399057A discloses a method, device, and medium for stripe correction using an uncooled detector. First, a 3*5 convolution template is used to traverse the image, obtaining the position of the first appearance of a stripe. Then, a 3*3 convolution template is used to distinguish whether the stripes are continuous. Different weight coefficients are assigned to different rows during the convolution process to reflect the different effects caused by different positions. By setting different convolution operators to process the image, it can be applied to the needs of different scenes. However, in this type of solution, the convolution operators and weight coefficients need to be adjusted according to scene changes, requiring long correction times and large computational resources, making it unsuitable for real-time applications and unable to meet the needs of mass production scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the technical problem to be solved by this invention is to provide a method for removing horizontal stripes in infrared images. This method effectively eliminates the horizontal stripe effect in infrared images without weakening local image details and global smoothness. By employing low-complexity filtering and calculation, it meets the real-time correction requirements for dynamic scenes in high-resolution infrared images or videos.

[0005] The technical solution of this invention employs a correction method based on intra-frame statistics, which eliminates horizontal stripe effects by analyzing and dynamically compensating for the grayscale differences between odd and even rows. The core idea of ​​this method is to use even-numbered rows as the brightness benchmark and correct odd-numbered rows. By calculating the difference between odd and even rows, the odd-numbered rows are adjusted accordingly to ensure their brightness is consistent with that of even-numbered rows, thereby eliminating the horizontal stripe phenomenon caused by the difference between odd and even rows.

[0006] Specifically, the steps include the following:

[0007] Step S1: For the original image containing horizontal stripes, calculate the difference in pixel grayscale between the odd-numbered rows and the adjacent even-numbered rows, and record it as the difference between the odd-numbered rows.

[0008] Step S2: Generate an initial odd-row difference image from all odd-row differences in the original image;

[0009] Step S3: For the initial odd-numbered row difference image, calculate and obtain its column mean array;

[0010] Step S4: Adjust the initial odd row difference image generated in step S2 using the column mean array obtained in step S3 to obtain a new odd row difference image;

[0011] Step S5: Based on the new odd-numbered row difference image obtained in step S4, correct the brightness of the odd-numbered row pixels in the original image to obtain the image result with horizontal stripes removed.

[0012] Preferably, in step S1, to calculate the grayscale difference between the pixels in the odd-numbered rows and the adjacent even-numbered rows, each odd-numbered row is processed pixel by pixel, with the current pixel as the center, using... The mean filter is applied to the horizontal window, and the resulting low-frequency grayscale value is denoted as nCurGrayMean. The same window is then used to apply the mean filter to the corresponding pixels in the previous even-numbered row of the current odd-numbered row, and the resulting low-frequency grayscale value is denoted as nLastGrayMean. The difference between these two corresponding low-frequency grayscale values ​​is calculated and denoted as the odd-numbered row difference. The calculation formula is:

[0013]

[0014] Preferably, in step S2, the resolution is The differences of all odd-numbered rows of the original image are used to form a matrix to generate the initial odd-numbered row difference image, denoted as iGrayMeanDiff. The resolution of iGrayMeanDiff is 1280 pixels wide and 512 pixels high.

[0015] Preferably, in step S3, the mean of each column of the initial odd-row difference image iGrayMeanDiff is calculated to obtain a column mean array, denoted as aMeanDiff, with a size of [missing information]. , representing the average compensated grayscale value for each column.

[0016] Preferably, in step S4, to avoid the problem of jagged edges appearing on the target edge due to directly using the initial odd-row difference image iGrayMeanDiff to compensate the original image, the column mean array aMeanDiff is used to adjust the initial odd-row difference image iGrayMeanDiff to obtain a new odd-row difference image MnGrayMeanDiff. The calculation formula is as follows:

[0017]

[0018] in Represents pixels, where i is the row number and j is the column number. The threshold value set.

[0019] Preferably, in step S5, each pixel value of the new odd-numbered row difference image MnGrayMeanDiff obtained in step S4 is summed with each corresponding pixel value of the odd-numbered rows of the original image, and boundary processing is performed to ensure that the compensated pixel values ​​are within the effective range, thereby correcting the brightness of the odd-numbered rows of the original image so that it is consistent with the adjacent even-numbered rows, and finally obtaining the image result with the horizontal stripes removed.

[0020] The beneficial effects obtained by adopting the above technical solution are as follows:

[0021] (1) Perform separate statistics and processing for each frame of the image to avoid using the compensation matrix of the current scene for the next frame of the image, which would cause deviation in the motion scene. Adapt to the video image of the dynamic scene and ensure that the image quality is always good during the video playback.

[0022] (2) It has low computational complexity: it only corrects odd-numbered row pixels and does not process even-numbered row pixels. The filter window height is 1, and it only requires mean filtering and column mean calculation. The amount of computation is small, which is suitable for real-time processing in embedded systems.

[0023] (3) A combined local and global processing method is adopted. Local differences are calculated by horizontal window filtering to avoid large grayscale changes and preserve image details. The vertical column mean reflects the overall trend of odd and even row response. The global constraint adjustment it provides avoids overcompensation of the target edge by the local difference. That is, it can smooth the jagged misalignment of the diagonal edge of the image and ensure that the horizontal lines are removed without affecting the image quality. Attached Figure Description

[0024] Figure 1 This is a flowchart of an infrared image horizontal stripe removal method according to the present invention.

[0025] Figure 2 This is the original image containing horizontal stripes and a magnified view of a portion thereof.

[0026] Figure 3 This is the initial odd-numbered row difference image.

[0027] Figure 4 The image and magnified view are obtained by directly using the initial odd-numbered row difference compensation to remove horizontal stripes. There are jagged edges and misalignments on the diagonal edges of the buildings in the scene.

[0028] Figure 5 This is a new odd-row difference image adjusted using the column-direction mean.

[0029] Figure 6 The image shows the removal of horizontal lines and a magnified view of the scene after applying the adjusted odd-row difference compensation. The diagonal edges of the buildings in the scene are smooth and natural. Detailed Implementation

[0030] The technical solution of the present invention will now be described more clearly and completely with reference to the accompanying drawings.

[0031] Input parameters and definitions:

[0032] Image resolution: width nWid=1280, height nHei=1024 (i.e., 1280 columns × 1024 rows).

[0033] Line definition:

[0034] Even rows: row numbers 0, 2, 4, ..., 1022 (base row, no correction).

[0035] Odd-numbered rows: row numbers 1, 3, 5, ..., 1023 (rows to be corrected).

[0036] The pixel is denoted as (nCurrLine, j), where nCurrLine is the row number and j is the column number.

[0037] (1) For example Figure 2The original image with horizontal stripes shown is used to calculate the pixel grayscale difference between odd-numbered rows and even-numbered rows.

[0038] For each odd-numbered row nCurrLine, process pixel by pixel: Using the current pixel (nCurrLine, j) as the center, employ 1×W, for example... The horizontal window is subjected to mean filtering, and the resulting low-frequency grayscale value is denoted as nCurGrayMean, with the following formula:

[0039]

[0040] Where I(nCurrLine, y) is the pixel value of the current row;

[0041] Using the same window, perform mean filtering on the same column position (i.e., the vicinity of coordinate j) of the even-numbered rows nCurrLine-1 of the previous row. The resulting low-frequency grayscale value is denoted as nLastGrayMean, and the formula is:

[0042]

[0043] Where I(nCurrLine-1, y) is the pixel value of the current row;

[0044] Calculate the difference of odd-numbered rows :

[0045] .

[0046] (2) The resolution is The differences of all odd-numbered rows of the original image are used to form a matrix to generate the initial odd-numbered row difference image, denoted as iGrayMeanDiff. The resolution of iGrayMeanDiff is 1280 pixels wide and 512 pixels high.

[0047] Since the initial odd-row difference image iGrayMeanDiff also statistically analyzes some target edges in the observed scene, using it for image compensation will cause jagged and misaligned target edges, such as... Figure 4 As shown, in the image results compensated with the initial odd-number row difference iGrayMeanDiff, although the horizontal stripes disappear, jagged and misaligned mosaic effects appear on the diagonal edges of the buildings in the scene.

[0048] (3) Calculate the column mean array for the initial odd-numbered row difference image.

[0049] For each column j of the initial odd-numbered row difference image iGrayMeanDiff, calculate the mean aMeanDiff of all rows i:

[0050]

[0051] The output aMeanDiff is of size An array representing the mean grayscale value for each column.

[0052] (4) Adjust the difference image of odd-numbered rows

[0053] The initial odd-row difference image iGrayMeanDiff is a local difference, and directly using it for image compensation will cause jagged, misaligned mosaic effects on the target edges. The vertical column mean aMeanDiff, however, reflects the overall trend of the odd-even row response, and its global constraint can prevent overcompensation of the target edges by the local difference. Therefore, iGrayMeanDiff is adjusted using aMeanDiff to obtain a new odd-row difference image MnGrayMeanDiff, calculated as follows:

[0054]

[0055] in Represents pixels, where i is the row number and j is the column number. The threshold value set.

[0056] The new odd-row difference image MnGrayMeanDiff, adjusted using the column-direction mean aMeanDiff, is shown below. Figure 5 As shown.

[0057] (5) Compensate for odd-numbered rows in the original image to obtain the image after removing horizontal stripes.

[0058] The pixel values ​​of each element in the new odd-row difference image MnGrayMeanDiff are summed with the corresponding pixel values ​​in each odd-row of the original image. Boundary processing is then performed to ensure that the compensated pixel values ​​are within an effective range. This corrects the brightness of the odd-rows in the original image, making it consistent with the adjacent even-rows. The final image with stripes removed is shown below. Figure 6 As shown, the sloping edges of the buildings in the scene become smooth and natural.

Claims

1. A method for removing horizontal stripes from infrared images, characterized in that, Includes the following steps: Step S1: For the original image containing horizontal stripes, calculate the difference in pixel grayscale between the odd-numbered rows and the adjacent even-numbered rows, and record it as the difference between the odd-numbered rows. Step S2: Generate an initial odd-row difference image from all odd-row differences in the original image; Step S3: For the initial odd-numbered row difference image, calculate and obtain its column mean array; Step S4: Adjust the initial odd row difference image generated in step S2 using the column mean array obtained in step S3 to obtain a new odd row difference image; Step S5: Based on the new odd-numbered row difference image obtained in step S4, correct the brightness of the odd-numbered row pixels in the original image to obtain the image result with horizontal stripes removed.

2. The infrared image horizontal stripe removal method according to claim 1, characterized in that, In step S1, to calculate the grayscale difference between the pixels in the odd-numbered rows and the adjacent even-numbered rows, each odd-numbered row is processed pixel by pixel, centered on the current pixel. The mean filter is applied to the horizontal window, and the resulting low-frequency grayscale value is denoted as nCurGrayMean. The same window is then used to apply the mean filter to the corresponding pixels in the previous even-numbered row of the current odd-numbered row, and the resulting low-frequency grayscale value is denoted as nLastGrayMean. The difference between these two corresponding low-frequency grayscale values ​​is calculated and denoted as the odd-numbered row difference. The calculation formula is: 。 3. The infrared image horizontal stripe removal method according to claim 2, characterized in that, In step S2, the resolution is The differences of all odd-numbered rows of the original image are used to form a matrix to generate the initial odd-numbered row difference image, denoted as iGrayMeanDiff. The resolution of iGrayMeanDiff is 1280 pixels wide and 512 pixels high.

4. The infrared image horizontal stripe removal method according to claim 3, characterized in that, In step S3, the mean value is calculated for each column of the initial odd-row difference image iGrayMeanDiff, resulting in a column mean array, denoted as aMeanDiff, with a size of [missing information]. , representing the average compensated grayscale value for each column.

5. The infrared image horizontal stripe removal method according to claim 4, characterized in that, In step S4, the initial odd-row difference image iGrayMeanDiff is adjusted using the column mean array aMeanDiff to obtain a new odd-row difference image MnGrayMeanDiff. The calculation formula is as follows: in Represents pixels, where i is the row number and j is the column number. The threshold value set.

6. The infrared image horizontal stripe removal method according to claim 5, characterized in that, In step S5, each pixel value of the new odd-row difference image MnGrayMeanDiff obtained in step S4 is summed with each corresponding pixel value of the odd-row of the original image, and boundary processing is performed to ensure that the compensated pixel values ​​are within the effective range. This corrects the brightness of the odd-row of the original image so that it is consistent with the adjacent even-row, and finally the image result with horizontal stripes removed is obtained.

Images