A gamma-ray radiation scene image purification method based on spatial axial gradient comparison

Abstract

The present invention relates to a method for image purification of gamma-ray irradiated scenes based on spatial axial gradient comparison. This method integrates the characteristics of gamma-ray-induced image noise and detects the coordinates of all noise pixels in a single frame through inter-frame difference. Grayscale gradient values ​​are then calculated along the horizontal, vertical, and two diagonal axes, centered on the detected noise pixel coordinates. These values ​​are then sorted by size. Adaptive median selection is performed on the two axes corresponding to the maximum and second-largest grayscale gradient values ​​in the sorted results. The two selected medians are weighted and summed based on the grayscale gradient values ​​of the axes to replace the original grayscale values ​​of the noise pixels. Finally, a Wallis sharpening filter is applied to the entire image to enhance the image's detailed edge texture information. This method can effectively eliminate gamma-ray irradiated image noise while preserving the image's detailed texture features. It features high processing speed and a simple algorithm, significantly improving image quality in gamma-ray irradiated scenes and providing methodological and technical support for the design of software modules for nuclear radiation-resistant cameras.

Description

Technical Field

[0001] The present invention belongs to the technical field of radiation image processing, and in particular relates to a gamma-ray radiation scene image purification method based on spatial axial gradient comparison. Background Art

[0002] Clear and reliable visual information is the foundation and prerequisite for nuclear engineering operations. However, the widespread presence of high-energy radiation and particles, such as gamma rays, in strong nuclear radiation environments can produce radiation effects on imaging equipment's image sensors, leading to anomalies in localized areas of the pixel array and the appearance of large amounts of bright, speckled and linear noise in captured images. This noise significantly degrades image quality, threatening the reliability of visual information in strong nuclear gamma radiation environments, hindering related engineering operations and, in severe cases, even leading to catastrophic consequences. Therefore, preventing this radiation-induced noise from interfering with imaging results and ensuring the reliability of visual information in nuclear environments is crucial for nuclear safety.

[0003] Based on this, traditional methods have been used to carry out radiation hardening treatment on imaging equipment such as cameras used in nuclear radiation environments from different angles such as process, circuit design, and shielding materials. However, these methods face problems such as high cost, long cycle, low flexibility, and high complexity.

[0004] In recent years, using image signal processing methods such as filtering and transform domains to address the reliability issues of nuclear radiation and scene visual information has become a research hotspot. However, the grayscale and frequency domain distributions of gamma-ray radiation scene noise differ from classical noise distributions such as Gaussian or Poisson distributions. Common filtering methods such as mean filtering, Gaussian filtering, and median filtering, as well as transform domain analysis methods such as Fourier transform and wavelet transform, are not fully applicable to gamma-ray radiation scene noise processing. While these methods partially reduce noise, they can blur non-noise image information and lack applicability and robustness for image noise reduction in gamma-ray radiation scenes with varying dose rates.

[0005] To address the problem that the reliability of visual information in strong nuclear radiation scenes is threatened, the present invention integrates the ideas and shortcomings of existing noise reduction methods, considers the mechanism of the interaction between gamma rays and image sensors, summarizes the characteristics of gamma-ray radiation scene image noise, and proposes a gamma-ray radiation scene image purification method based on spatial axial gradient comparison. Summary of the Invention

[0006] The present invention aims to provide a more targeted and adaptable image noise reduction method for gamma-ray radiation scenes. This method first detects the spatial coordinates of all noise pixels in the image using continuous inter-frame differencing. Then, with the detected noise pixel as the center, the grayscale gradient values ​​within n pixels adjacent to the noise pixel are calculated along the four axes: the horizontal x-axis, the vertical y-axis, and the diagonal axes d1 and d2. The grayscale gradient values ​​along the four axes are then sorted by magnitude. An adaptive median is selected for the axes with the maximum and second-largest grayscale gradient values. The medians along the two selected axes are weighted, and the result of this weighted sum is used as the new grayscale value at the noise pixel. Finally, the image is sharpened and superimposed using a Wallis sharpening filter window to enhance image detail and edge texture information while reducing image blur. This method can meet the requirements for image noise reduction in scenes with strong nuclear gamma radiation, significantly improving image quality and ensuring the reliability of visual information in such scenes. The method can accurately detect and effectively remove noise from gamma-ray radiation scene images while maintaining image detail texture information, thereby improving the image quality of strong nuclear gamma-ray radiation scenes.

[0007] The method for cleaning a gamma-ray radiation scene image based on spatial axial gradient comparison described in the present invention is specifically performed in the following steps:

[0008] a. Use a camera to continuously capture 20 frames of images to be purified with an integration time of 30ms in a gamma-ray radiation scene;

[0009] b. Set the grayscale threshold T and the number of frames f for inter-frame difference, select an image frame from the images acquired in step a for processing, perform inter-frame difference between the image and its f adjacent frames, and detect the position coordinates of all noise pixels on the image;

[0010] c. Taking the noise pixel detected on the image in step b as the center, calculate the grayscale gradient values ​​of the noise pixel and its adjacent n pixels in the horizontal axis x, vertical axis y, diagonal axis d1 and axis d2, and sort the grayscale gradient values ​​in the four axes by size. The maximum and second largest grayscale gradient values ​​are recorded as G max1 and G max2 ;

[0011] d. The maximum and second largest grayscale gradient values ​​G in step c max1 and G max2 The adaptive median selection is performed on the two axes, and the pixel number range m used for median selection on each axis is adaptively adjusted with a step size of 2. The median values ​​selected in the two directions are respectively recorded as I zmax1 and Izmax2 ;

[0012] e. Gray gradient value G max1 and G max2 As weight, the two medians I selected in step d are zmax1 and I zmax2 Perform weighted summation to obtain a new grayscale value I′(x, y) to replace the original grayscale value I(x, y) at the noise pixel;

[0013] f. Repeat steps ce to process the next noise pixel in turn until all noise pixels detected on the image in step b are traversed to obtain the entire denoised image I′(x, y);

[0014] g. Perform Wallis sharpening on the entire denoised image I′(x, y) processed in steps af to obtain image W(x, y);

[0015] h. Superimpose the sharpened image W(x, y) obtained in step g and the denoised image I′(x, y) obtained in step f to obtain a purified image P(x, y).

[0016] The grayscale threshold T and frame number f used for inter-frame difference in step b and the number of pixels n used to calculate the grayscale gradient in step c should be positively correlated with the radiation dose rate in the γ-ray radiation scene. For the γ-ray radiation scene with a dose rate of 20 rad(Si) / s, the threshold T is set to 15, the frame number f is set to 10, and the pixel number range n is set to 7.

[0017] The pixel number range m used for adaptive median selection in step d is flexibly adjustable, and the initial selected pixel number range m is set to 2.

[0018] The Wallis sharpening filter window in step g is 3×3 and the step size is 1.

[0019] The present invention provides a method for purifying gamma-ray radiation scene images based on spatial axial gradient comparison. The method:

[0020] a. Use a camera-type imaging device to continuously collect 20 frames of images to be processed with an integration time of 30ms in a gamma-ray radiation environment;

[0021] b. Set the grayscale threshold T and the number of frames f for inter-frame difference, select an image frame from the images acquired in step a, and perform inter-frame difference between it and its f adjacent frames before and after it, to detect and determine the position coordinates of all gamma-ray radiation noise pixels on the image;

[0022] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step b is specifically as follows:

[0023] For the current frame image I t , and the f frames before and after it are represented as I t-f , I t-f+1 …I t …I t+f-1 , I t+f , the grayscale value at the pixel position (x, y) on the current frame image can be expressed as I t(x，y) , use formula (1) to determine whether the position (x, y) on the image is noise:

[0024]

[0025] In formula (1), R(x, y) represents the result of judging whether the position (x, y) on the current frame image is noise, 1 represents yes, 0 represents no; I t(x，y) Represents the pixel value at the (x, y) position in the current frame image; I t±i(x，y) represents the grayscale value of the pixel at position (x, y) on the i-th frame before or after the current frame; T represents the grayscale difference threshold between noise and normal background pixels; k is an adjustable input scaling factor, indicating how many frames the pixel at position (x, y) in the current frame must be greater than the pixel at the same position in the current frame to be considered a noise pixel; sgn(x) is the step response function;

[0026] c. Taking the noise pixel detected on the image as the center, calculate the grayscale gradient values ​​of the n adjacent pixels of the noise pixel in the horizontal axis x, vertical axis y, diagonal axis d1 and axis d2 respectively, and sort the grayscale gradient values ​​of the four axes by size. The maximum and second largest grayscale gradient values ​​are recorded as G max1 and G max2 ;

[0027] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step c is specifically as follows:

[0028] Since the pixels in the noise area of ​​the γ-ray radiation scene image tend to gather on a certain axis, that is, the movement path of the γ-ray in the CIS, the grayscale gradient values ​​of the adjacent n pixels on the horizontal axis x, vertical axis y, and diagonal axes d1 and d2 of the noise pixel detected on the image, i.e., the pixel position of R(x, y) = 1, are calculated by equations (2) to (5) respectively, and their sizes are compared by equations (6) and (7). The maximum and second largest grayscale gradient values ​​are recorded as G max1 and G max2 .

[0029]

[0030]

[0031]

[0032]

[0033] G max1 =max(G x , G y , G d1 , G d2 ) (6)

[0034] G max2 =max{(G x , G y , G d1 , G d2 )-G max} (7)

[0035] Where n is the number of adjacent pixels on the axis selected when calculating the grayscale gradient value. It is set according to the size of the radiation environment dose rate. The higher the dose rate, the larger the n setting.

[0036] d. At the maximum and second largest gray gradient values ​​G max1 and G max2 The median selection is performed adaptively on the two axes, and the pixel range m used for median selection on each axis is adaptively adjustable. The median values ​​selected in the two directions are recorded as I zmax1 and I zmax2 ;

[0037] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step d is specifically as follows:

[0038] Let Z xy is the set of gray values ​​of m pixels adjacent to the noise pixel point on a certain axis, Z max 、Z min and Z med Z xy The maximum, minimum and middle values ​​in Z med -Z max <0 and Z med -Z min >0, then output Z med is the median value of the axis;

[0039] If Z is not satisfied med -Z max <0 and Z med -Z min > 0, then set m = m + 2, and judge again until the output is Z med , m is the pixel range of the adaptively selected median;

[0040] When the gray gradient value is G max1 and G max2 Perform step d on the two axes to obtain I zmax1 and I zmax2 ;

[0041] e. The grayscale gradient value G corresponding to the two axes max1 and G max2 The median of the two values ​​I will be selected for the weight zmax1 and I zmax2 Perform weighted summation to obtain I′ (x，y) , instead of the gray value I at the noise pixel (x，y) ;

[0042] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step e is specifically as follows:

[0043] The median values ​​on the two axes are weighted and summed according to the corresponding gray gradient values ​​through formula (8), and the weighted summation result I′ is used (x，y) Replace the original grayscale value of the noise pixel;

[0044]

[0045] In formula (8), G max1 and G max2 are the two larger grayscale gradient values ​​obtained in step c, I zmax1 and I zmax2 is the median value on the two axes obtained in step d;

[0046] f. Repeat step ce to process the next noisy pixel until all the noisy pixels on the image are traversed to obtain the denoised image I′(x, y);

[0047] g. Perform Wallis sharpening on the entire denoised image I′(x, y) obtained after steps af to obtain image W(x, y);

[0048] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step g is specifically as follows:

[0049] Wallis sharpening is based on the Laplace operator and takes into account the logarithmic link in human visual characteristics. Therefore, when sharpening, the logarithmic processing method is used for improvement. The Wallis sharpening operator is shown in formula (9). The window size is 3*3. The Wallis sharpening filter formula (10) can be used to sharpen the image to enhance the detail edge texture information of the image.

[0050]

[0051]

[0052] In formula (10), I′(x, y) is the image before sharpening, and W(x, y) is the image after sharpening. To prevent taking the logarithm of 0, log[I(x, y)+1] is improved to log[I(x, y)+1] during calculation. Since the logarithm value is relatively small, log(256)=5.45, logI(x, y) is improved to 46logI(x, y) during calculation, where 46=255 / log256.

[0053] h. Superimpose the sharpened image W(x, y) obtained in step g and the denoised image I′(x, y) obtained in step f to obtain a purified image P(x, y);

[0054] Furthermore, in the gamma-ray radiation scene image purification method based on spatial axial gradient discrimination, step h is specifically as follows:

[0055] The purified image can be obtained by superimposing the denoised image with the result of Wallis sharpening filtering. While reducing noise, it can also retain the image detail texture information to the greatest extent and reduce image blur. As shown in formula (11);

[0056] P(x,y)=I′(x,y)+W(x,y) (11)

[0057] In formula (11), I′(x, y) is the denoised image in the present invention, and W(x, y) is the result of Wallis sharpening filtering of the denoised image I′(x, y). The two are superimposed to obtain the purified image P(x, y).

[0058] The gamma-ray radiation scene image purification method based on spatial axial gradient comparison described in the present invention has the following significant technical effects:

[0059] 1) Accurately and completely detect the noise caused by gamma-ray radiation on the image: Utilizing the difference between gamma-ray radiation image noise and clean image information, the action time of gamma photons in the image sensor is much shorter than the single-frame integration time. The image noise induced by gamma rays has instantaneous pulse characteristics, while the background information is continuous in the time domain. In other words, by selecting the appropriate inter-frame difference threshold and frame number, the image noise caused by gamma-ray radiation can be accurately detected.

[0060] 2) Effectively remove noise caused by gamma-ray radiation on the image: The image noise induced by gamma-ray radiation is mainly concentrated in a linear spatial geometric feature on a single axis. The noise geometric distribution is determined by the grayscale gradient. The clean information pixels near the noise pixels are weighted by the corresponding axial grayscale gradient value, replacing the original grayscale value of the noise location to achieve the purpose of noise reduction.

[0061] 3) Without losing the image's detailed texture information: The denoised image is further sharpened by using the Wallis sharpening filter that is more adapted to the characteristics of the human eye, so that the image's detailed texture information is retained while removing noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a flow chart of the present invention;

[0063] Figure 2 An image captured by the camera of the present invention in a gamma-ray radiation scene;

[0064] Figure 3 An image in a gamma-ray radiation scene after purification by the method proposed in the present invention;

[0065] Figure 4 A single noise region (dark field background) on the gamma-ray radiation scene image of the present invention;

[0066] Figure 5 For the present invention Figure 4 Taking the pixel with medium stripe noise as an example, the method of the present invention is used to achieve the processing result of a single pixel. DETAILED DESCRIPTION

[0067] The following is a combination of specific embodiments and the appendix of the specification Figure 2 and Figure 3 The present invention further describes a gamma-ray radiation scene image purification method based on spatial axial gradient comparison.

[0068] Example 1

[0069] a. In a gamma-ray radiation environment, use an AR0230 1920×1080 pixel CMOS image sensor camera to continuously capture 20 frames of images to be processed using a 30ms integration time on a standard camera test card. Figure 2 The following figure shows a typical image frame, which will be purified in the following steps;

[0070] b. Set the threshold value T for inter-frame difference to 15, the number of frames to 5, and convert the image selected in step ( Figure 2) and its f=10 adjacent frames before and after it, perform inter-frame difference, judge whether the pixel at the (x, y) position on the image is noise according to formula (1), and determine the position of the noise pixel on the image;

[0071]

[0072] c. Taking the noise pixel detected on the image, i.e., the pixel position where R(x, y) = 1, as the center, calculate the grayscale gradient values ​​of n pixels adjacent to the pixel at (x, y) along the horizontal axis x, vertical axis y, and diagonal axes d1 and d2 using equations (2)-(5), and compare their sizes using equations (6) and (7). The maximum and second-largest grayscale gradient values ​​are recorded as G max1 and G max2 ;

[0073]

[0074]

[0075]

[0076]

[0077] G max1 =max(G x , G y , G d1 , G d2 ) (6)

[0078] G max2 =max{(G x , G y , G d1 , G d2 )-G max} (7)

[0079] d. In the obtained G max1 and G max2 The adaptive median selection is performed on the m adjacent pixels except the point (x, y) on the two corresponding axes, and the results of the median selection are respectively I zmax1 and I amax2 ;

[0080] e. The grayscale gradient value G corresponding to the two axes max1 and G max2 The median of the two values ​​I will be selected for the weight zmax1 and I zmax2 Perform weighted summation and get I′ as shown in formula (8) (x，y) , instead of the gray value I at the noise pixel (x，y) ;

[0081]

[0082] f. Repeat step ce to process the next noisy pixel until all the noisy pixels on the image are traversed to obtain the denoised image I′(x, y);

[0083] g. Perform Wallis sharpening on the overall image I′(x, y) after steps af using equation (10) to obtain image W(x, y);

[0084]

[0085] h. Superimpose the sharpened image W(x, y) obtained in step g and the denoised image I′(x, y) obtained in step f using equation (11) to obtain the purified image P(x, y);

[0086] P(x,y)=I′(x,y)+W(x,y) (11)

[0087] The method of the present invention takes into account the radiation effect mechanism of gamma photons in image sensors, combines the temporal transient and spatial geometric characteristics of gamma-ray radiation scene image noise, and proposes a gamma-ray radiation scene image purification method based on spatial axial comparison. By setting different input parameters such as differential threshold T, number of frames f, proportional coefficient k, pixel range n for calculating grayscale gradient, pixel range m for adaptive median selection, etc., it is found that there are differences in the quality of the purified images. Therefore, in the embodiment, by adjusting the different input parameters of the proposed method, the purified image under the real gamma-ray radiation scene with a dose rate of 20rad(Si) / s is obtained, as shown in FIG. Figure 3 shown.

[0088] In specific embodiment 1 of the present invention, the grayscale threshold T for inter-frame difference is set to 15, the number of frames f is 5, the proportional coefficient k is 0.5, the pixel range n for calculating the grayscale gradient is 12, and the pixel range m for adaptive median selection is 6. The peak signal-to-noise ratio (PSNR) of the purified image reaches 30.6dB, and the structural similarity (SSIM) reaches 0.82, indicating that this method has advanced advantages in radiation image processing.

[0089] The following is a combination of specific embodiments and the appendix of the specification Figure 4 and Figure 5 The present invention further describes a gamma-ray radiation scene image purification method based on spatial axial gradient comparison.

[0090] Example 2

[0091] a. Take a single noise area in a dark field image captured by an AR0230 1920*1080 pixel CMOS image sensor camera in a gamma-ray radiation environment as an example. Figure 4 As shown;

[0092] b. Wherein, white pixels are noise pixels within a single noise region detected after differencing between consecutive frames;

[0093] c. Process the detected noise pixels pixel by pixel. In this example, the pixel with a grayscale value of 142 (marked by a stripe) is taken as the center, and the grayscale gradient values ​​of the adjacent n = 2 pixels on the horizontal axis x, vertical axis y, and diagonal axes d1 and d2 are calculated:

[0094]

[0095]

[0096]

[0097]

[0098] Therefore: G max1 =max(G x , G y , G d1 , G d2 )=G x

[0099] G max2 =max{(G x , G y , G d1 , G d2 )-G max}=G d2

[0100] d. Therefore, in G max1 and G max2 That is G x and G d2 The adaptive median selection is performed on m=4 adjacent pixels except the stripe pixel points in the corresponding x-axis and d-axis directions, and the results of the median selection are recorded as I zmax1 =10 and I zmax2 =10;

[0101] e. According to the corresponding gray gradient value G x and G d2 For the median value I on both axes zmax1 and I zmax2 Perform weighted summation and use the result of weighted summation I′ (x，y) Replace the grayscale value at the noise pixel:

[0102]

[0103] f. Repeat steps ce to process the next noisy pixel until all the noisy pixels on the image are traversed to obtain the denoised image;

[0104] g. Perform Wallis sharpening filtering on the entire image after noise reduction, which is adapted to the visual characteristics of the human eye, to enhance the image's detail edge texture information;

[0105] h. Superimpose the sharpened image obtained in step g and the denoised image obtained in step f to obtain a purified gamma-ray radiation scene image.

[0106] The present invention provides a method for purifying images of gamma-ray radiation scenes based on spatial axial gradient comparison. After multiple experiments on images in real gamma-ray radiation scenes, it was found that the method can accurately detect image noise pixels induced by gamma-ray radiation, achieve good noise reduction effects while retaining the detailed texture features of the image, effectively improve the image quality in scenes with strong gamma-ray nuclear radiation, and provide methods and technical support for the design of software modules for nuclear radiation-resistant cameras.

[0107] The above-described embodiment is only an implementation method of a gamma-ray radiation scene image purification method based on spatial axial gradient comparison described in the present invention, but the protection scope of the present invention is not limited thereto. Any replacement or addition or subtraction that can be understood by anyone familiar with the technology within the technical scope disclosed by the present invention should be included in the scope of the present invention.

Claims

1. A method for γ-ray radiation scene image purification based on spatial axial gradient comparison, characterized in that The specific operation is as follows: a. Use a camera to continuously capture 20 frames of images to be purified with an integration time of 30ms in a gamma-ray radiation scene; b. Set the grayscale threshold T and the number of frames f for inter-frame difference, select an image frame from the images acquired in step a for processing, perform inter-frame difference between the image and its f adjacent frames, and detect the position coordinates of all noise pixels on the image; c. Taking the noise pixel detected on the image in step b as the center, calculate the grayscale gradient values ​​of the noise pixel and its adjacent n pixels in the horizontal axis x, vertical axis y, diagonal axis d1 and axis d2, and sort the grayscale gradient values ​​in the four axes by size. The maximum and second largest grayscale gradient values ​​are recorded as G max1 and G max2 ; d. The maximum and second largest grayscale gradient values ​​G in step c max1 and G max2 The adaptive median selection is performed on the two axes, and the pixel number range m used for median selection on each axis is adaptively adjusted with a step size of 2. The median values ​​selected in the two directions are respectively recorded as I zmax1 and I zmax2 ; e. Gray gradient value G max1 and G max2 As weight, the two medians I selected in step d are zmax1 and I zmax2 Perform weighted summation to obtain a new grayscale value I′ (x，y) , instead of the gray value I at the noise pixel (x，y) ; f. Repeat steps ce to process the next noise pixel in turn until all noise pixels detected on the image in step b are traversed to obtain the entire denoised image I′(x, y); g. Perform Wallis sharpening on the entire denoised image I′(x, y) processed in steps af to obtain image W(x, y); h. Superimpose the sharpened image W(x, y) obtained in step g and the denoised image I′(x, y) obtained in step f to obtain a purified image P(x, y).

2. The method for γ-ray radiation scene image purification based on spatial axial gradient comparison according to claim 1 is characterized in that The grayscale threshold T and frame number f used for inter-frame difference in step b and the number of pixels n used to calculate the grayscale gradient in step c should be positively correlated with the radiation dose rate in the γ-ray radiation scene. For the γ-ray radiation scene with a dose rate of 20 rad(Si) / s, the threshold T is set to 15, the frame number f is set to 10, and the pixel number range n is set to 7.

3. The method for γ-ray radiation scene image purification based on spatial axial gradient comparison according to claim 1 is characterized in that The pixel number range m used for adaptive median selection in step d is flexibly adjustable, and the initial selected pixel number range m is set to 2.

4. The method for γ-ray radiation scene image purification based on spatial axial gradient comparison according to claim 1 is characterized in that The Wallis sharpening filter window in step g is 3×3 and the step size is 1.

Images