X-ray back scattering image processing method

Abstract

The invention relates to an X-ray back scattering image processing method, which belongs to the technical field of X-ray optical and digital image processing, and aims at the characteristics of an image obtained by X-ray staring type back scattering imaging based on a lobster eye lens with a high aspect ratio, the main noise type on the image is fully considered, and the X-ray back scattering image is obtained. Firstly, a cross-shaped background generated by a unique point spread function of a lobster eye lens with a high aspect ratio in an image is removed, then noise reduction is performed on the image according to the characteristics of imaging noise under an extremely weak light condition, bright blocks generated by stray light in the process are subjected to multi-filtering, pixel stretching is performed on the image through histogram equalization, and the image is subjected to high-aspect-ratio image processing. And finally, bright blocks are filtered out through connected domain analysis to obtain a processed image, noise reduction of the X-ray back scattering image is achieved, the function of separating a target from a cross background and enhancing the target is achieved, and the method can greatly improve the signal-to-noise ratio of the back scattering image under the extremely weak light condition.

Description

Technical Field

[0001] The present invention relates to the technical field of X-ray optics and digital image processing, and in particular to an X-ray backscatter image processing method. Background Art

[0002] X-ray imaging plays a vital role in nondestructive testing and has a wide range of applications in various fields such as medical diagnosis, security, and industrial inspection. Conventional transmission X-ray imaging systems perform well in detecting metals or inorganic materials, but face challenges in accurately imaging low atomic number materials. In addition, these systems require the detector and source to be placed on both sides of the target, which limits their practical applications (R. Lanier, “Recent Developments in X-Ray Imaging Technology”). Compared with X-ray transmission systems, X-ray backscatter imaging has better imaging effects on low atomic number materials due to its strong Compton scattering, thereby improving the imaging effect of such targets (D.-C.Dinca, JR Schubert, J. Callerame, et al., “X-ray backscatter imaging”). Currently, the conventional scanning X-ray backscatter system developed by the American Science and Engineering Company (AS&E) adopts the patented “flying spot” technology (J. Callerame, “X-ray backscatter imaging: Photography through barriers”). However, unlike staring imaging systems, this system requires relative motion between the test object and the scanning beam, which limits its imaging distance and resolution.

[0003] Introducing lobster-eye lenses into backscatter imaging systems is an effective way to address the limitations of traditional imaging methods. Lobster-eye optics was first proposed in 1979 (JRP Angel, “Lobster eyes as X-raytelescopes”), inspired by the reflective compound eyes of large crustaceans, and first applied to astronomy as an X-ray telescope. The device consists of thousands of mirror boxes arranged on a common center of curvature (MF Land, “Eyeswith mirror optics”), and its reflection-based focusing mode and spherical structure give the lobster-eye lens a unique focusing capability and a wide field of view (FOV) (V. Dániel, A. Inneman, L. Pína, et al., “X-ray Lobster Eye all-sky monitor for rocket experiment”). In 2005, Gertsenshteyn and his colleagues first proposed a method combining lobster-eye optics with X-ray backscattering detection (M. Gertsenshteyn, T. Jannson, G.Savant, et al., Staring / focusing lobster-eye hard x-ray imaging for non-astronomical objects). In 2007, V. Grubsky et al. first used two sets of parallel gold-coated silicon wafers with a spacing of 1 mm to intersect to form a 14×12 micron channel plate (mcp) to perform X-ray backscattering imaging experiments on clean containers placed behind obstacles (V. Grubsky, M. Gertsenshteyn, T. Jannson, et al., “Non-scanning x-ray backscattering inspection systems based on x-ray focusing”). In 2016, Xu Jie, Wang Xin, et al. proposed a three-lens Schmidt objective by dividing one of the two groups of reflectors into two independent groups of reflectors based on the original double-lens Schmidt geometry (J. Xu, X. Wang, Q. Zhan, et al., A novel lobster-eye imaging system based on Schmidt-type objective for X-ray-backscattering inspection). They stacked three groups of surface data of D263 optical glass into a Schmidt objective and conducted imaging experiments.

[0004] Since the lobster eye lens is reflective focusing based on the total reflection theorem, in order to obtain the best focusing (M. Ouyang, X. Zhao, W. He, et al., Structural design method of the meridional lobster-eye lens with optimal efffciency), the channel must have an ultra-high aspect ratio and the channel sidewall roughness must be less than 1nm (L. Li, C. Zhang, G. Jin, et al., Study on the optical properties of Angel Lobster eye X-ray ffat micro pore optical device), which has always been a challenge for the manufacturing process and coating of metal materials on the inner wall surface of such a narrow square channel.

[0005] In 2022, Li et al. from Northern Night Vision Technology Co., Ltd. (NNVT) reported that they overcame these challenges and tested high aspect ratio (500) lobster eye micro-hole optics for the first time using micro-channel plate technology and atomic layer deposition technology (L. Li, Y. Zhang, M. Ouyang, et al., Fabrication and Performance of Lobster Eye X-Ray Micro Pore Optics with the Ultra-high Aspect Ratio). This method makes it possible to use lobster eye micro-hole optics for hard X-rays.

[0006] However, compared with traditional focused imaging, X-ray backscatter imaging is not only significantly affected by noise due to the weak Compton scattering signal, but also the unique structure of the lobster eye lens introduces inherent cross-shaped artifacts in imaging. In addition, the limitations of the manufacturing process make it impossible for the lens to completely and accurately focus the scattered light, resulting in bright spots in the detector received image caused by the transmission light that directly penetrates the lens and the noise superimposed on the backscattered image, which makes it difficult to obtain useful information directly from the original backscattered image. Therefore, it is urgent to reduce the noise of X-ray backscatter imaging and eliminate the cross-shaped artifacts. Summary of the invention

[0007] In view of the above problems, the present invention proposes an X-ray backscatter image processing method, which fully considers the imaging characteristics of the lobster eye lens and the noise type in the image to perform noise reduction and enhancement processing on the X-ray backscatter image, and eliminates the cross-like background.

[0008] To solve the above problems, the present invention adopts the following technical solutions: The X-ray backscatter image processing method comprises the following steps: Step 1: Acquire the original image obtained by X-ray staring backscatter imaging based on a high aspect ratio lobster eye lens; Step 2: removing the cross-shaped background of the original image to obtain an image after the cross-shaped background is removed. The process of removing the cross-shaped background specifically includes the following steps: Step 2.1: extracting each column pixel and each row pixel that does not contain the target from the original image; Step 2.2: Expand each column and row of pixels extracted in step 2.1 into matrix; Step 2.3: For each Matrix, subtracted from the original image Matrix, get the corresponding result, then sum up all the results and take the average value to get the image after removing the cross-like background; Step 3: According to formula (2), a variance stability transformation is performed on the image after the cross-shaped background is removed, and the multiplicative noise of the image is converted into additive Gaussian white noise to obtain the converted image: (2) in, and Respectively represent the converted image and the image after removing the cross-like background; Step 4: Using a Gaussian low-pass filter to reduce noise on the converted image, filtering out additive Gaussian white noise in the image, and obtaining a reduced-noise image; Step 5: Perform a variance stability inverse transformation on the denoised image according to formula (5) to obtain the original distribution image; (5) in, is the original distribution image, is the image after noise reduction; Step 6: Filter the original distribution image using a Gaussian high-pass filter to obtain a filtered image; Step 7: Using a histogram equalization method to stretch the pixels of the filtered image, filter the bright spots in the image into a plurality of small connected areas whose pixel values ​​are equivalent to the target, and obtain a histogram equalization result; Step 8: Performing connected domain analysis on the histogram equalization result to obtain a processed image. The connected domain analysis process specifically includes the following steps: Step 8.1: Convert the histogram equalization result into a binary image; Step 8.2: Use the classic two-pass sequential region marking to identify and mark each connected region, and calculate the number of pixels each connected region has; Step 8.3: Take the average number of pixels in each connected region as the threshold, treat the connected regions with a value less than the threshold as noise and filter them out, and retain the connected regions with a value greater than the threshold.

[0009] The beneficial effects of the present invention are as follows: the present invention aims at the characteristics of the image obtained by X-ray staring backscatter imaging based on a high aspect ratio lobster eye lens, fully considers the main noise types on the image, first removes the cross-like background generated by the unique point spread function of the high aspect ratio lobster eye lens in the image, and then reduces the noise of the image according to the characteristics of imaging noise under extremely weak light conditions, and after multi-filtering the bright blocks generated by stray light (including transmitted light and background scattered light) in the process, the image is pixel stretched by histogram equalization, and finally the bright blocks are filtered out by connected domain analysis to obtain the processed image, thereby achieving the denoising of the X-ray backscatter image and the function of separating and enhancing the target from the cross-like background. Experiments show that after being processed by the method of the present invention, the structural similarity index (SSIM) of the X-ray backscatter image and the true value image is improved from an average of 0.0526 to 0.5758, indicating that the method of the present invention can greatly improve the signal-to-noise ratio of the backscatter image under extremely weak light conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of the X-ray backscatter image processing method according to an embodiment of the present invention; Figure 2 Schematic diagram of the optical path of the staring X-ray backscattering experiment based on the high aspect ratio lobster eye lens; Figure 3 Schematic diagram of the focusing principle of a high aspect ratio lobster eye lens for focusing X-rays; Figure 4 is the cross point spread function of the lobster eye lens at a photon energy of 33.336keV; Figure 5 The original image of X-ray backscattering when the light source emits photons with an energy of 40kev and the corresponding Compton scattered photons with an energy of 33.336kev. Figure 6The result diagrams of each step in the X-ray backscatter image processing method described in the embodiment of the present invention are shown. For the convenience of display, sub-images (a), (b) and (c) are logarithmically transformed, wherein: sub-image (a) is the original image of X-ray backscatter; sub-image (b) is the image after the cross-shaped background is removed; sub-image (c) is the image after variance stability transformation of sub-image (a); sub-image (d) is the image after noise reduction of sub-image (c); sub-image (e) is the image after inverse variance stability transformation; sub-image (f) is the image after high-pass filtering of sub-image (e); sub-image (g) is the image after histogram equalization; sub-image (h) is the image after connected domain analysis; sub-image (i) is the true value image that is produced to accurately match the target position in the original image of X-ray backscatter; Figure 7 This is the result of the image without stray light processing such as high-pass filtering and connected domain analysis. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] This embodiment provides an X-ray staring backscatter image processing method based on a high aspect ratio (aspect ratio of up to 500) lobster eye lens, such as Figure 1 As shown, the method specifically includes the following steps 1 to 8.

[0013] Step 1 (S1): First, X-ray staring backscatter imaging is performed based on a high aspect ratio lobster eye lens to obtain an original image after imaging. There is often noise in the original image, and there are also cross-shaped artifacts. The focusing principle of the high aspect ratio lobster eye lens used to focus X-rays is as follows: Figure 3 As shown. Figure 2 Taking the test target shown in the figure as an example, the original image obtained is as follows Figure 6 As shown in sub-figure (a) in .

[0014] Step 2 (S2): Based on the unique cross-shaped characteristics of the high aspect ratio lobster eye lens, the cross-shaped background is removed from the original image with low signal-to-noise ratio and the target is submerged behind the cross-shaped background. The removal method can be expressed as: (1) in, and are the number of rows and columns in the original image that do not contain the feature image; the index value and Respectively represent the total number of rows and columns contained in the entire original image; express The original image of the order; and Respectively represent the Line or Column expansion matrix.

[0015] The X-ray backscatter image processing method of this embodiment uses the lobster eye lens point spread function as a cross, such as Figure 4 As shown in the figure, the cross arm intensity is smaller than the central focal spot intensity, and the cross arm is removed. According to formula (1), the process of removing the cross-like background specifically includes the following steps: Step 2.1: Extract each column pixel and each row pixel that does not contain the target from the original image; Step 2.2: Expand the row pixels and column pixels extracted in step 2.1 into matrix; Step 2.3: For each Matrix, subtracted from the original image Matrix, get the corresponding result, that is, subtract these matrices from the original image respectively, then sum up all the results and take the average value to get the result of removing the cross arm, that is, get the image after removing the cross-like background, such as Figure 6 As shown in sub-figure (b) in .

[0016] Step 3 (S3): Perform a variance stability transform (Anscombe's) on the image obtained in step 2 after removing the cross-like background. In X-ray backscatter imaging, since the Compton scattering signal used for imaging is extremely weak, the imaging process is performed under extremely weak light conditions. At this time, the main noise in the image is multiplicative Poisson noise. This noise is specifically manifested as the extremely low average effective number of photons, and the intensity value of each pixel fluctuates around the expected intensity of the pixel. This fluctuation is particularly obvious under weak light conditions, that is, when the intensity expectation is extremely low. It is difficult to separate the multiplicative noise directly from the image, so it is converted into more easily processable additive Gaussian noise to facilitate subsequent noise reduction. In this step, the multiplicative noise of the image is converted into additive Gaussian white noise by performing a variance stability transform on the image after removing the cross-like background, and the converted image is obtained, as shown in FIG. Figure 6 The conversion process can be expressed by the following formula: (2) in, and They represent the converted image and the image before conversion, i.e. the image after removing the cross-shaped background. Figure 6 As shown in sub-figure (c) in .

[0017] Step 4 (S4): Using a Gaussian low-pass filter can effectively reduce the additive Gaussian white noise in the image after step 3 and effectively avoid ringing artifacts in the image. After the Gaussian low-pass filter reduces the noise, the reduced-noise image is obtained, such as Figure 6 As shown in sub-figure (d) in . The Gaussian low-pass filter can be expressed as: (3) in, represents the threshold of the Gaussian low-pass filter, Represents the distance between two points in the frequency domain, expressed as: (4) in, Respectively represent two points of the image in the frequency domain.

[0018] Step 5 (S5): Perform a variance stability inverse transform on the denoised image obtained in step 4, transform the image back to the original distribution, and obtain the original distribution image, such as Figure 6 As shown in subfigure (e) in , the variance-stabilized inverse transform of the debiased model can be expressed as: (5) in, is the original distribution image, is the image after denoising. Since the variance stability transformation is a nonlinear transformation, if the image is directly transformed algebraically during the inverse transformation, the resulting image will be biased. Therefore, formula (5) adopts the inverse transformation method of reducing bias.

[0019] Step 6 (S6): The transmitted X-rays and background stray light will form inconspicuous bright spots in the original backscattered image, which are slightly lower than the target but higher than the cross-like background. However, after the cross-like background is removed in step 2, the contrast of these bright spots will be amplified and become more obvious, such as Figure 7 and Figure 6 As shown in sub-figure (c) in Figure 7 This is the result of the image without stray light processing such as high-pass filtering and connected domain analysis. It can be seen that the target is seriously affected by the light spot formed by the transmitted X-ray. Therefore, this step needs to use a Gaussian high-pass filter to filter these bright blocks in the original distribution image to obtain a filtered image, such as Figure 6 As shown in the sub-image (f) in , the pixels are then stretched and binarized in step 7 to transform them into multiple small connected areas that are not connected to each other, so that they can be easily distinguished from the target and finally filtered out. The Gaussian high-pass filter can be expressed as: (6) in, represents the threshold of the Gaussian high-pass filter, It also represents the distance between two points in the frequency domain. The filtered image is as follows Figure 6 As shown in sub-figure (f) in .

[0020] Step 7 (S7): Use the histogram equalization method to stretch the pixels of the filtered image obtained in step 6 to increase the image contrast and facilitate subsequent operations. The bright spots in the filtered image are filtered into multiple small connected areas with pixel values ​​equivalent to the target by the histogram equalization method. The histogram equalization method can be expressed as: (7) in, is the number of rows of image pixels With the number of columns The product of is the upper limit of the digital interval of the image grayscale, Indicates the gray value is The number of pixels, is the gray value after transformation, is the mapping function, using the mapping function Process the filtered image. The histogram equalized image obtained after step 7 is as follows Figure 6 As shown in sub-figure (g) in .

[0021] Step 8 (S8): After Step 6 and Step 7, the noise is greatly limited and the large bright blocks are filtered into multiple small connected areas whose pixel values ​​are equivalent to the target. The connected domain analysis method is introduced to suppress the small connected areas whose pixel values ​​are consistent with the target and are not connected to each other.

[0022] The process of connected domain analysis specifically includes the following steps: Step 8.1: Convert the histogram equalization result into a binary image; Step 8.2: Then use the classic two-pass (C. Fiorio, J.Gustedt et al "Two linear time Union-Find strategies for image processing") region labeling to identify and label each connected region, and calculate the number of pixels each connected region has; Step 8.3: Take the average number of pixels in each connected area as the threshold, treat the connected areas smaller than the threshold as noise and filter them out, and retain the connected areas larger than the threshold to prevent the target from being misidentified as noise when there are stray blocks larger than the target area. Finally, the processed image is obtained, as shown in Figure 6As shown in sub-figure (h) in .

[0023] The X-ray backscatter image processing method proposed in this embodiment takes into full consideration the main noise types on the image, based on the characteristics of the image obtained by X-ray staring backscatter imaging based on a high aspect ratio lobster eye lens, and first removes the cross-shaped background generated by the unique point spread function of the high aspect ratio lobster eye lens in the image, and then reduces the noise of the image according to the characteristics of imaging noise under extremely weak light conditions. After multiple filtering of bright blocks generated by stray light (including transmitted light and background scattered light) in the process, the image is pixel-stretched through histogram equalization, and finally the bright blocks are filtered out through connected domain analysis to obtain a processed image, thereby achieving noise reduction of the X-ray backscatter image and realizing the function of separating and enhancing the target from the cross-shaped background.

[0024] To further verify the technical effect of the present invention, the present invention also built a staring X-ray backscattering experimental optical path based on a high aspect ratio lobster eye lens, such as Figure 2 As shown in the figure, the letter S of polyamide material is installed as the target to be measured on the target support frame of low scattering material. The target to be measured is placed 1500mm in front of the high aspect ratio lobster eye lens, and the lobster eye lens is fixed by the lens bracket; the Csl scintillator detector is placed 300mm behind the lobster eye lens; the laboratory tabletop X-ray light source is placed 220mm in front of the target to be measured, and the axis of the light source-target to be measured and the axis of the target to be measured-lobster eye lens-detector form an acute angle of 45 degrees ( ).

[0025] The staring X-ray backscattering experimental optical path was used for imaging. The energy of the light source emitted photons was set to 40kev, and the corresponding Compton scattered photon energy was 33.336kev. The original image of the X-ray backscattering was obtained after imaging. Figure 5 For the convenience of display, Figure 6 (a) shows the Figure 5 To evaluate the processing method proposed in the present invention, a true value image is produced that accurately matches the target position in the original image of X-ray backscattering, such as Figure 6 As shown in sub-image (i) in the figure, the structural similarity (SSIM) index between the original image and the true image is calculated. The experimental results show that the SSIM index is improved from an average of 0.0526 to 0.5758, indicating that this method can greatly improve the signal-to-noise ratio of the backscattered image under extremely weak light conditions.

[0026] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0027] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An X-ray backscatter image processing method, characterized in that: The following steps are involved: Step 1: Acquire the original image obtained by X-ray staring backscatter imaging based on a high aspect ratio lobster eye lens; Step 2: removing the cross-shaped background of the original image to obtain an image after the cross-shaped background is removed. The process of removing the cross-shaped background specifically includes the following steps: Step 2.1: extracting each column pixel and each row pixel that does not contain the target from the original image; Step 2.2: Expand each column and row of pixels extracted in step 2.1 into matrix; Step 2.3: For each Matrix, subtracted from the original image Matrix, get the corresponding result, then sum up all the results and take the average value to get the image after removing the cross-like background; Step 3: According to formula (2), a variance stability transformation is performed on the image after the cross-shaped background is removed, and the multiplicative noise of the image is converted into additive Gaussian white noise to obtain the converted image: （2） in, and Respectively represent the converted image and the image after removing the cross-like background; Step 4: Using a Gaussian low-pass filter to reduce noise on the converted image, filtering out additive Gaussian white noise in the image, and obtaining a reduced-noise image; Step 5: Perform a variance stability inverse transformation on the denoised image according to formula (5) to obtain the original distribution image; （5） in, is the original distribution image, is the image after noise reduction; Step 6: Filter the original distribution image using a Gaussian high-pass filter to obtain a filtered image; Step 7: Using a histogram equalization method to stretch the pixels of the filtered image, filter the bright spots in the image into a plurality of small connected areas whose pixel values ​​are equivalent to the target, and obtain a histogram equalization result; Step 8: Performing connected domain analysis on the histogram equalization result to obtain a processed image. The connected domain analysis process specifically includes the following steps: Step 8.1: Convert the histogram equalization result into a binary image; Step 8.2: Use the classic two-pass sequential region marking to identify and mark each connected region, and calculate the number of pixels each connected region has; Step 8.3: Take the average number of pixels in each connected region as the threshold, treat the connected regions with a value less than the threshold as noise and filter them out, and retain the connected regions with a value greater than the threshold.

2. The X-ray backscatter image processing method according to claim 1, characterized in that: The formula of the histogram equalization method is: （7） in, is the number of rows of image pixels, is the number of columns of image pixels, is the upper limit of the digital interval of the image grayscale, Indicates the gray value is The number of pixels, is the gray value after transformation, is the mapping function.

3. The X-ray backscatter image processing method according to claim 1, characterized in that: The Gaussian low-pass filter is expressed as: （3） in, Represent two points of the image in the frequency domain, represents the distance between two points in the frequency domain, Represents the threshold of the Gaussian low-pass filter.

4. The X-ray backscatter image processing method according to claim 3, characterized in that: Gaussian low-pass filter threshold Set to 10.

5. The X-ray backscatter image processing method according to claim 1, characterized in that: The Gaussian high-pass filter can be expressed as: （6） in, Represent two points of the image in the frequency domain, represents the threshold of the Gaussian high-pass filter, Represents the distance between two points in the frequency domain.

6. The X-ray backscatter image processing method according to claim 5, characterized in that: Gaussian high pass filter threshold Set to 4.

7.

7. The X-ray backscatter image processing method according to claim 1, characterized in that: During X-ray staring backscatter imaging, the energy range of the photons emitted by the light source is 33.336kev~40kev.

Images