Micro-focus CT slice image noise reduction algorithm based on structure extraction and storage medium

Abstract

The invention discloses a micro-focus CT slice image noise reduction algorithm based on structure extraction and a storage medium, and relates to the technical field of industrial CT nondestructive detection.The method comprises the steps that 1, pre-noise reduction processing is conducted on an original micro-focus CT slice image I0, a three-dimensional block matching filtering algorithm is used for removing remarkable noise points, and a pre-noise-reduced image Ipre containing weak noise is generated; 2, performing structure extraction on the pre-denoised image Ipre, adjusting a texture information penalty weight lambda by adopting a relative total variation algorithm to obtain a structure image Istru, and extracting texture information Itexture through difference operation between the Ipre and the Istru; 3, the pixel gray level of the texture information Itexture is enhanced by alpha times, alpha is larger than or equal to 0, then the texture information Itexture is superposed to the pre-noise-reduction image Ipre, and an information enhanced image Ienhanced is generated; and step 4, carrying out final noise reduction processing on the information enhanced image Ienhanced, and outputting a low-noise micro-focus CT slice image Ideniosed of which the detail edge is completely reserved by utilizing a three-dimensional block matched filtering algorithm and combining with the improved similarity measurement parameter sigma fin.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial CT non-destructive testing, and particularly relates to a micro-focus CT slice image denoising algorithm based on structure extraction and a storage medium. Background Technique

[0002] Micro-focus computer tomography (CT) is one of the main non-destructive testing techniques in the current industrial field. This technique uses X-rays to transmit through an object under non-contact and non-destructive conditions. By collecting transmission projection images at different angles and reconstructing them, tomographic imaging of the object's interior is achieved. Its slice images can relatively accurately obtain the internal structure and damage conditions of the object, and it is one of the important means for non-destructive testing of the internal quality of products.

[0003] Micro-focus CT batch detection mainly includes two links: product loading and unloading and micro-focus CT scanning. Using a robot to grasp the product can automatically achieve rapid loading and unloading of the product without interrupting the X-ray beam output, effectively improving the efficiency of micro-focus CT batch inspection of products. However, micro-focus CT scanning takes relatively long time, which has become one of the bottlenecks restricting the improvement of the efficiency of micro-focus CT batch detection of products. On the premise of ensuring the reliability of the internal quality detection of products, improving the micro-focus CT scanning efficiency has become one of the key technologies that need to be solved urgently.

[0004] The micro-focus CT scanning parameters mainly include: tube voltage, tube current, number of projection angles, single-angle acquisition time, etc. Among them: the tube voltage will affect the artifacts, noise, and spatial resolution of CT slice images; the tube current will affect the noise and spatial resolution of CT slice images; the number of projection angles will affect the image noise. For special-shaped parts with large differences in X-ray penetration thickness in all directions, too few projection angles will also cause radial reconstruction artifacts; while the single-angle acquisition time only affects the noise of CT slice images. Generally speaking, the X-ray penetration thickness of most industrial parts has little difference in all directions and does not belong to special-shaped parts. When the number of projection angles is not less than 512, there are almost no radial artifacts. Therefore, in the technical field of industrial CT non-destructive testing, the number of projection angles and the single-angle acquisition time have less impact on the quality of micro-focus CT slice images compared with other CT scanning parameters. In practice, reducing the number of projection angles and the single-angle acquisition time is an important way to reduce the micro-focus CT scanning time. Given that the projection of the micro-focus CT device is obtained after the turntable steps into place, reducing the single-angle acquisition time can only reduce the total acquisition time, while reducing the number of projection angles can reduce both the total turntable stepping time and the total acquisition time. Therefore, reducing the number of projection angles is the most effective method to improve the micro-focus CT scanning efficiency.

[0005] The X-ray source tube power of microfocus CT is relatively low, the distance between the X-ray source and the detector is relatively far, and the total dose of X-rays received by the detector is very small, which results in the slice images of microfocus CT being vulnerable to noise. Reducing the number of projection angles will further reduce the total dose of X-rays received by the detector, deteriorate the quality of microfocus CT slice images, and even affect subsequent tasks such as the determination of the internal quality of products and the quantitative analysis of defects. Therefore, the slice images of microfocus CT often require noise reduction processing.

[0006] The difficulty in the research of noise reduction algorithms lies in how to distinguish between noise and texture information in images. The core of traditional noise reduction algorithms is to estimate the true gray value of the current pixel through the gray distribution information of the neighborhood or similar blocks. The advantage of algorithms such as BM3D compared with previous algorithms is that they use artificially set similarity measurement parameters to preferentially select similar blocks within the entire image range and perform selective weighting on the pixel gray values. Therefore, this algorithm can retain the edge information of image details to a certain extent during noise reduction. However, when the noise is severe, it is often difficult to achieve both good image noise reduction effect and preservation of edge detail information, which is also a common problem faced by most noise reduction algorithms. Summary of the Invention

[0007] Through algorithm research, the present invention provides a noise reduction algorithm for microfocus CT slice images based on structure extraction, which realizes noise suppression of microfocus CT slice images and reduces the influence of noise on the determination of the internal quality of products and the quantitative analysis of product defects.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A noise reduction algorithm for microfocus CT slice images based on structure extraction, comprising the following steps:

[0010] Step 1: Perform pre-noise reduction processing on the original microfocus CT slice image I0, and use a three-dimensional block matching filtering algorithm to remove significant noise points to generate a pre-noise reduction image I pre ;

[0011] Step 2: Perform structure extraction on the pre-noise reduction image I pre , adjust the texture information penalty weight λ using the relative total variation algorithm to obtain a structure image I stru , and extract the texture information I pre through the difference operation between I stru and I texture ;

[0012] Step 3: Enhance the pixel gray value of the texture information I texture by α times, where α ≥ 0, and then superimpose it on the pre-noise reduction image I pre to generate an information-enhanced image I enhanced ;

[0013] Step 4: Perform final noise reduction on the information-enhanced image I enhanced using the 3D block matching filtering algorithm in combination with the enhanced similarity measurement parameter σ fin to output a low-noise micro-focus CT slice image I with complete details and edges preserved denoised .

[0014] In the said Step 1,

[0015] the 3D block matching filtering algorithm is described as:

[0016] Output = BM3D(Input, σ, distribution, wiener) (1)

[0017] where Output is the output image;

[0018] Input is the input image; σ is the similarity measurement parameter for the reference block and candidate blocks selected manually. The larger σ is, the more candidate blocks are similar to the reference block, the larger the 3D array for noise reduction is, and the more obvious the noise reduction effect is. While the smaller σ is, the fewer candidate blocks are similar to the reference block, the smaller the 3D array for noise reduction is, and the less obvious the noise reduction effect is;

[0019] distribution is the noise type of the input image Input, mainly divided into Gaussian noise 'Gauss' and Rice noise 'Rice';

[0020] wiener is 1 or 0, representing whether Wiener filtering is performed.

[0021] In the said Step 2, the value of the texture information penalty weight λ is dynamically adjusted according to the texture richness of the micro-focus CT slice image. When the value of λ is too large, the structure image I stru is over-smoothed and the texture information is lost; when the value of λ is too small, the structure image I stru retains too many texture details.

[0022] In the said Step 2, the relative total variation algorithm is described as:

[0023] Stru = tsmooth(Input, μ, λ, maxTter, ε) (2)

[0024] where Stru is the output image (structure image), Input is the input image, μ is the pixel size of the texture elements in the image, λ is the penalty weight, maxTter is the number of iterations, and ε is the parameter controlling the sharpness of the output image.

[0025] In the said Step 3, the gray-scale enhancement multiple α and I prerelated to the noise intensity inside, and the selection range is usually 0 to 2; by appropriately adjusting the texture enhancement multiple α, on the premise that the noise intensity in I enhanced is equivalent to the noise intensity in I0, make I enhanced have stronger detailed edge information compared to the detailed edge information in I0.

[0026] In step 4, assume that the similarity measurement parameter for pre-denoising is σ pre , and the similarity measurement parameter for final denoising is σ fin , then σ fin should be one order of magnitude higher than σ pre .

[0027] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps of any of the above-mentioned micro-focus CT slice image denoising algorithms based on structure extraction.

[0028] The beneficial effects that may be brought by a micro-focus CT slice image denoising algorithm based on structure extraction disclosed in this application include but are not limited to:

[0029] Using the algorithm provided by the present invention, the contrast-to-noise ratio, signal-to-noise ratio, and edge sharpness of micro-focus CT images can be significantly improved at the same time, and the edge detail information of the image can be better retained, which is beneficial to reducing the number of micro-focus CT scanning angles and significantly and effectively improving the micro-focus CT scanning efficiency on the premise of ensuring the quality of micro-focus CT images and not affecting the internal quality determination and defect quantitative analysis.

[0030] This algorithm initially suppresses the noise of the micro-focus CT original slice image through pre-denoising and structure extraction, and obtains the pre-denoised image and structure image of the micro-focus CT original slice; subtracts the structure image from the pre-denoised image to obtain the texture information of the micro-focus CT slice image; uses the texture information to enhance the information of the pre-denoised image, and performs final denoising on the pre-denoised image with enhanced information to obtain a low-noise micro-focus CT image with good retention of original detailed edges. The present invention can effectively suppress the noise interference brought by rapid micro-focus CT detection, significantly improve the micro-focus CT detection efficiency, reduce the micro-focus CT detection cost, and lay a technical foundation for rapid batch detection of micro-focus CT. Description of the Drawings

[0031] Figure 1 is a flowchart of a micro-focus CT slice image denoising algorithm based on structure extraction;

[0032] Figure 2 is the implementation effect of the algorithm of the present invention. Detailed Embodiment

[0033] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0034] On the contrary, the present application covers any alternatives, modifications, equivalent methods and solutions made on the essence and scope of the present application as defined by the claims. Further, in order to enable the public to have a better understanding of the present application, some specific details are described in detail in the following detailed description of the present application. Those skilled in the art can fully understand the present application without the description of these details.

[0035] A noise reduction algorithm for micro-focus CT slice images based on structure extraction involved in the embodiments of the present application will be described in detail below.

[0036] As Figure 1 shown, a noise reduction algorithm for micro-focus CT slice images based on structure extraction includes the following steps:

[0037] Step 1: Use noise reduction algorithms such as Block-matching and 3D filtering (BM3D) to perform weak pre-noise reduction on the original micro-focus CT slice image I0, remove relatively obvious noise points, and obtain a pre-noise reduction image I pre ;

[0038] Step 2: Use structure extraction algorithms such as Relative total variation (RTV) to perform structure extraction on I pre to obtain a structure image I stru , and then subtract I pre from I stru to obtain texture information I texture ;

[0039] Step 3: After enhancing the pixel gray level of the texture information I texture by α times, superimpose it on the pre-noise reduction image I pre to obtain an information-enhanced image I enhanced ;

[0040] Step 4: Use noise reduction algorithms such as BM3D to perform final noise reduction on the information-enhanced image I enhanced to obtain a low-noise micro-focus CT slice image I denoised with good detail edge preservation.

[0041] In the above Step 1, the BM3D algorithm can be simply described as:

[0042] Output = BM3D(Input, σ, distribution, wiener) (1)

[0043] In the formula, Output is the output image;

[0044] Input is the input image;

[0045] σ is the similarity measurement parameter of the artificially selected reference block and the candidate block. The larger σ is, the more candidate blocks are similar to the reference block, the larger the three-dimensional array used for noise reduction, and the more obvious the noise reduction effect. On the contrary, the smaller σ is, the fewer candidate blocks are similar to the reference block, the smaller the three-dimensional array used for noise reduction, and the less obvious the noise reduction effect;

[0046] distribution is the noise type of the input image Input, mainly divided into Gaussian noise 'Gauss' and Rice noise 'Rice'; wiener is 1 or 0, representing whether Wiener filtering is performed.

[0047] The noise type of the CT slice image is Gaussian noise; Wiener filtering helps to further improve the noise reduction effect, and usually wiener = 1 is selected; while σ is the key parameter of the BM3D algorithm and should be selected according to experience.

[0048] In the above step 2, the RTV algorithm can be simply described as:

[0049] Stru = tsmooth(Input, μ, λ, maxTter, ε) (2)

[0050] In the formula, Stru is the output image (structural image);

[0051] Input is the input image;

[0052] μ is the pixel size of the texture elements in the image (usually taken as 3 - 5);

[0053] λ is the penalty weight (usually taken between 0 and 1);

[0054] maxTter is the number of iterations;

[0055] ε is the parameter to control the sharpness of the output image.

[0056] For microfocus CT slice images, the size of their texture elements is small, and generally μ = 3 is taken; the number of iterations maxTter and the sharpness parameter ε are directly selected as the recommended values of the RTV algorithm (maxTter = 4, ε = 0.02); while the penalty weight λ is the key parameter of the RTV algorithm, representing the penalty intensity for the texture information of the image, and λ needs to be selected according to the texture richness in the actual microfocus CT slice.

[0057] In step 2, a suitable texture information penalty weight λ is selected to obtain a structure image I that contains almost no image texture information. stru , and using the pre-denoised image I pre subtract the structure image I stru to obtain the texture information I of the pre-denoised image texture . In this step, if the penalty weight λ is too large, I stru may be over-smoothed, resulting in I texture containing more structural information and inaccurate texture extraction; if λ is too small, I stru may contain more texture information, and the noise and detail information in I texture are too little, and the texture extraction is also inaccurate; therefore, by selecting a suitable penalty weight λ according to the texture information intensity of the actual micro-focus CT slice image, it can be considered that I texture contains almost no structural information of I pre .

[0058] In step 3 above, the gray-scale enhancement multiple α is related to the noise intensity in I pre , and the selection range is usually 0 to 2. I enhanced Compared with I pre , both the detail edge information and the noise intensity are enhanced. And in step 1, it is known that the noise in I pre is weaker than that in I0, and the detail information in I pre is comparable to that in I0. By appropriately adjusting the texture enhancement multiple α, on the premise that the noise intensity in I enhanced is comparable to the noise intensity in I0, the detail edge information in I enhanced can be made stronger than the detail edge information in I0. So that the detail edge information is less likely to be suppressed in the final denoising process.

[0059] In step 4 above, assume that the similarity measurement parameter for pre-denoising is σ pre , and the similarity measurement parameter for final denoising is σ fin , then σ fin should be one order of magnitude higher than σ pre .

[0060] The following will make a detailed description of a micro-focus CT slice image denoising algorithm based on structure extraction according to the present invention in combination with the embodiments and the accompanying drawings.

[0061] 1) Divide the CT volume data into layers of micro-focus CT slice images in the Z direction (height direction);

[0062] 2) According to the appendix Figure 1Step: Use noise reduction algorithms such as BM3D to pre - reduce the noise of the micro - focus CT slice images (distribution takes 'Gauss', wiener takes 1, σ pre is positively correlated with the image noise and is selected according to experience in practice), and obtain the pre - denoised image I pre ;

[0063] 3) Perform structure extraction on the pre - denoised image I pre (μ takes 3, maxTter takes 4, ε takes 0.02, and λ needs to be selected according to the texture richness in the micro - focus CT slice), and obtain the structure image I stru ;

[0064] 4) Let I enhanced = I pre +α×(I pre - I stru ), where α ranges from 0 to 2;

[0065] 5) Use the BM3D algorithm to perform final noise reduction on the micro - focus CT slice images (distribution takes 'Gauss', wiener takes 1, σ fin is positively correlated with the image noise, and in practice, σ fin should be one order of magnitude higher than σ pre ), and obtain the denoised CT image, and the effect is as shown in the appendix Figure 2 ;

[0066] 6) Repeat steps 2) to 5) until all layers of the micro - focus CT slice images are processed.

[0067] 7) Stack the denoised CT slice images directly together in order to obtain the denoised micro - focus CT volume data.

[0068] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A microfocus CT slice image denoising algorithm based on structure extraction, characterized in that It includes the following steps: Step 1: Perform pre-denoising on the original micro-focus CT slice image I0, and use a three-dimensional block matching filtering algorithm to remove significant noise points, generating a pre-denoised image I with weaker noise pre ; Step 2: For the pre-denoised image I pre perform structure extraction, adjust the texture information penalty weight λ using the relative total variation algorithm, and obtain the structure image I stru , and extract the texture information I pre through the difference operation between I stru and I texture ; Step 3: Enhance the pixel grayscale of the texture information I texture by α times, where α ≥ 0, and then superimpose it on the pre-denoised image I pre to generate an information-enhanced image I enhanced ; Step 4: Perform final noise reduction on the information-enhanced image I enhanced using the 3D block matching filtering algorithm combined with the enhanced similarity measurement parameter σ fin to output a low-noise micro-focus CT slice image I with complete preservation of detailed edges denoised .

2. The noise reduction algorithm for micro-focus CT slice images based on structure extraction according to claim 1, characterized in that: In the said step 1, The 3D block matching filtering algorithm is described as: Output=BM3D(Input,σ,distribution,wiener) (1) In the formula, Output is the output image; Input is the input image; σ is the similarity measurement parameter of the artificially selected reference block and the candidate block. The larger σ is, the more candidate blocks are similar to the reference block, the larger the 3D array for noise reduction is, and the more obvious the noise reduction effect is. On the contrary, the smaller σ is, the fewer candidate blocks are similar to the reference block, the smaller the 3D array for noise reduction is, and the less obvious the noise reduction effect is; distribution is the noise type of the input image Input, mainly divided into Gaussian noise 'Gauss' and Rice noise 'Rice'; wiener is 1 or 0, representing whether Wiener filtering is performed.

3. The micro-focus CT slice image denoising algorithm based on structure extraction according to claim 1, wherein: In the step 2, the value of the texture information penalty weight λ is dynamically adjusted according to the texture richness of the micro-focus CT slice image. When the value of λ is too large, the structure image I stru is over-smoothed, resulting in the loss of texture information; when the value of λ is too small, the structure image I stru retains too many texture details.

4. The noise reduction algorithm for micro-focus CT slice images based on structure extraction according to claim 1, characterized in that: In the said step 2, the relative total variation algorithm is described as: Stru=tsmooth(Input,μ,λ,maxTter,ε) (2) In the formula, Stru is the output image (structural image), Input is the input image, μ is the pixel size of the texture elements in the image, λ is the penalty weight, maxTter is the number of iterations, and ε is the parameter for controlling the clarity of the output image.

5. The micro-focus CT slice image denoising algorithm based on structure extraction according to claim 1, wherein: In the said step 3, the gray-scale enhancement multiple α is related to the noise intensity within I pre and the selection range is usually 0 to 2; by appropriately adjusting the texture enhancement multiple α, on the premise that the noise intensity in I enhanced is equivalent to the noise intensity in I0, make the detailed edge information in I enhanced stronger than the detailed edge information in I0.

6. The micro-focus CT slice image noise reduction algorithm based on structure extraction according to claim 1, characterized in that: In step 4, assume that the similarity measurement parameter for pre-noise reduction is σ pre , and the similarity measurement parameter for final noise reduction is σ fin . Then σ fin should be one order of magnitude higher than σ pre .

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the micro-focus CT slice image noise reduction algorithm based on structure extraction as described in any one of claims 1-6.

Images