An X-ray image denoising method, system, storage medium and application

Abstract

The application relates to an X-ray image noise reduction method, system, storage medium and application, acquires an original X-ray image to be reduced in noise; performs multi-resolution decomposition on the original X-ray image; performs directional filtering processing on each layer of detail image; calculates the difference value of the detail image and the filtering result; performs a first limiting operation on the initial noise estimation; performs multi-resolution reconstruction on the noise after the first limiting operation and the low-frequency basic noise of zero value of each layer; performs a second limiting operation on the reconstructed full-resolution noise image; subtracts the corrected noise image from the original X-ray image to obtain a final noise reduction image. The application has strong detail retention capability: structure information is extracted in each detail layer through directional filtering, noise is separated through difference operation, and edge blurring and texture loss caused by directly performing threshold processing on high-frequency details are avoided.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to an X-ray image denoising method, system, storage medium and application. Specifically, it relates to an X-ray image denoising method and system based on multi-resolution decomposition and directional filtering, which is particularly suitable for medical imaging scenarios with significant noise, such as low-dose X-ray imaging and dynamic fluorescence fluoroscopy. Background Technology

[0002] X-ray imaging is widely used in clinical diagnosis, interventional surgery guidance, and industrial non-destructive testing. To reduce radiation dose, low-dose imaging protocols are often used, but this leads to a significant increase in image noise, affecting image quality and diagnostic accuracy.

[0003] Existing noise reduction methods can be mainly classified into the following categories: Spatial filtering methods, such as mean filtering, median filtering, and Gaussian filtering, are simple to implement but can easily lead to blurred edges and loss of details.

[0004] Transform domain filtering methods, such as wavelet thresholding and Fourier transform filtering, can suppress noise within a certain frequency range. However, the choice of threshold has a significant impact on the results and is prone to artifacts at singular points.

[0005] Denoising methods based on deep learning require a large amount of paired data for training, the model's generalization ability is affected by the training set, and there is a lack of interpretability in the medical field.

[0006] Multi-resolution analysis methods, such as wavelet transform and pyramid decomposition, can decompose images into different frequency bands for processing. However, effectively separating noise from details remains a challenge. In existing technologies, pyramid decomposition-based methods typically perform thresholding or filtering on high-frequency details directly, which can easily misjudge useful high-frequency information such as weak edges and textures as noise and remove them, resulting in loss of detail.

[0007] Therefore, it is necessary to propose a noise reduction method that can effectively preserve the detailed structure of an image against a background of strong noise.

[0008] In view of the above-mentioned shortcomings, the designer has actively researched and innovated in order to create an X-ray image noise reduction method, system, storage medium and application, so as to make it more industrially valuable. Summary of the Invention

[0009] To address the aforementioned technical problems, the present invention aims to provide an X-ray image noise reduction method, system, storage medium, and application.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: One of the objectives of this invention is: An X-ray image noise reduction method includes the following steps: Step 1: Obtain the original X-ray image I to be denoised; Step 2: Perform multi-resolution decomposition on the original X-ray image I to obtain a series of detail layer images {L0, L1, ..., L...} with different spatial frequencies. k-1} and a low-frequency substrate image G k ; Step 3: For each layer of detail image L i (i=0,1,..., k-1) undergo directional filtering to obtain the filtered result F that preserves edge and texture directions. i ; Step 4: Calculate the detail image L i With the filtering result F i The difference is used to obtain the initial noise estimate N for this layer. i =L i -F i ; Step 5: Estimate N from the initial noise i Perform the first limiting operation to restrict its value to the interval [T]. low i ,T high i Within this range, the final estimated noise N for this layer is obtained. i ; Step 6: Reduce the noise {N0, N1, ..., N} after the first clipping operation on each layer. k-1} and perform multi-resolution reconstruction on a low-frequency substrate with zero values ​​to synthesize a full-resolution noisy image N with the same size as the original X-ray image I. total ; Step 7: Process the reconstructed full-resolution noisy image N total Perform a second limiting operation to restrict its value to the interval [C]. low C high Within ], the corrected noisy image N′ is obtained. total ; Step 8: Subtract the corrected noise image N′ from the original X-ray image I. total The final denoised image I is obtained. denoised .

[0011] As a further improvement of the present invention, the multi-resolution decomposition is performed by decomposing and reconstructing the Laplacian pyramid; the detail layer image is the detail image of each layer of the Laplacian pyramid; and the low-frequency base image is the bottom layer image of the Gaussian pyramid.

[0012] As a further improvement of the present invention, the directional filtering process employs at least one or a combination of the following filters: template-based directional filters, steerable filters, Gabor filters, and anisotropic diffusion filters.

[0013] As a further improvement of the present invention, the directional filtering process includes: filtering the detail layer image using filter banks in multiple directions, and selecting the filtering result corresponding to the direction with the largest response, or fusing the filtering results from multiple directions to obtain the filtering result F. i .

[0014] As a further improvement of the present invention, in the first limiting operation: T low i =-c·σ i T high i =c□σ i , where σ i Estimating the initial noise N′ for the i-th layer i The local or global standard deviation estimate, where c is a constant factor.

[0015] As a further improvement of the present invention, in the first clipping operation: the final estimated noise N i Right now: As a further improvement of the present invention, the threshold C in the second limiting operation... low and C high C is adaptively determined based on the global grayscale range or noise level of the original X-ray image I: low =-d·σ global C high =d□σ global , where σ global This is the noise standard deviation estimate for the flat region of the original image, where d is a constant factor.

[0016] The second objective of this invention is: An X-ray image noise reduction system, comprising: Memory, used to store computer programs; A processor for executing computer programs to implement an X-ray image noise reduction method as described above.

[0017] The third objective of this invention: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the X-ray image noise reduction method described above.

[0018] The fourth objective of this invention: The above-described X-ray image denoising method is applied to denoising medical images or industrial inspection images.

[0019] By means of the above-described solution, the present invention has at least the following advantages: Strong ability to preserve details: Structural information is extracted from each detail layer through directional filtering, and noise is separated by difference operation, avoiding edge blurring and texture loss caused by directly thresholding high-frequency details.

[0020] Dual protection mechanism: Clamp operation is introduced in the noise separation layer and after noise reconstruction to effectively prevent useful signals (especially weak edges and textures) from being misjudged as noise, overcoming the shortcomings of traditional multi-resolution methods that are prone to loss of details.

[0021] Highly adaptable: The noise threshold can be adaptively adjusted according to the characteristics of different resolution layers of the image, which improves the robustness of the method to different noise levels and image content.

[0022] The physical meaning is clear and the interpretability is strong: the entire process is based on the intuitive concept of multi-resolution analysis and noise separation, without relying on black-box models. The results are stable and reliable, and are easy to accept and validate in high-risk fields such as medicine.

[0023] Wide applicability: It is not only applicable to low-dose X-ray images, but can also be extended to other noise-type medical images (such as CT and MRI) or industrial inspection image noise reduction tasks that require strong detail preservation.

[0024] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the following are preferred embodiments of the present invention described in detail with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart of the overall method according to the fifth embodiment of the present invention; Figure 2 This is a schematic diagram of the decomposition and reconstruction of the Laplace Pyramid of the present invention; Figure 3 This is a schematic diagram demonstrating the layered noise separation effect of the present invention; Figure 4 This is a schematic diagram of the directional filtering template of the present invention; Figure 5 This is a schematic diagram of the noise reconstruction and final noise reduction results of the present invention; Figure 6 This is a schematic diagram comparing the effects of direct enhancement of clinical images according to the present invention with the enhancement effect after noise reduction in this scheme; Figure 7 This is a schematic diagram comparing the image enhancement effect of the line pair method of the present invention with the enhancement effect after noise reduction in this solution. Detailed Implementation

[0027] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0029] This invention aims to provide an X-ray image noise reduction method and system. By using Laplacian pyramid multi-resolution decomposition, combined with directional filtering and noise separation techniques, it can effectively suppress noise while preserving anatomical structures, edges, and texture details in the image to the greatest extent, thereby improving image quality and diagnostic value.

[0030] First embodiment of the present invention: The core of this embodiment is: through multi-resolution decomposition, directional filtering is used to extract "noise-free" structural information in each detail layer, the noise component of the layer is separated by difference operation, and clamp operation is used to protect weak signals that may be mixed with noise; finally, through noise reconstruction and secondary clamping, the reconstructed noise is subtracted from the original image to obtain the noise reduction result.

[0031] like Figure 1 As shown, an X-ray image noise reduction method according to this embodiment includes the following steps in sequence: Input: Obtain the original X-ray image I to be denoised. Multi-resolution decomposition: Perform Laplacian Pyramid decomposition on the original image I to obtain a series of detail layer images {L0, L1, ..., L...} at different spatial frequencies. k-1} and a low-frequency substrate image G k Where L0 is the finest layer (highest frequency), L... k-1 As the roughest layer, G k This is the lowest-level, low-frequency image.

[0032] Layered noise estimation and separation: for each layer of detail image L of the Laplacian pyramid i (i=0,1,...,k-1), perform the following sub-steps: a. Directional filtering: for detail layer L i Perform directional filtering to obtain a filtered result F that preserves edge and texture directions. i This filtering aims to simulate the structural information of the resolution layer under "noise-free" conditions. Directional filtering can be achieved using template-based directional filters, Steerable filters, Gabor filters, or anisotropic diffusion filters, etc.

[0033] b. Noise extraction: Calculate detail layer L i With the directional filtering result F i The difference is used to obtain the initial noise estimate N for this layer. i =L i -F i .

[0034] c. Noise Clamp: Estimating N based on the initial noise level. i Perform a Clamp operation to restrict its value to the interval [T]. low i ,T high i Within this range, the final estimated noise N for this layer is obtained. i .Right now: Among them, T low i and T high i The noise threshold for the i-th layer can be adaptively set based on the noise statistics of that layer (such as standard deviation) or preset to a fixed value. This step aims to prevent the over-removal of useful detail signals with small amplitudes (such as weak edges) as noise.

[0035] Noise reconstruction: Reconstructing the noise {N0, N1, ..., N} from each layer after Clamp processing.k-1} and zero-valued low-frequency basal noise (assuming the basal layer is noise-free or negligible), and synthesize a full-resolution noisy image N with the same size as the original image I using the Laplacian pyramid reconstruction algorithm. total .

[0036] Global noise clamping: For the reconstructed full-resolution noisy image N... total Perform a global Clamp operation again to restrict its value to the range [C]. low C high Within ], the corrected noisy image N′ is obtained. total This step serves as a second layer of protection, preventing overshooting of noise estimation in certain areas due to error accumulation during reconstruction, thereby protecting strong edges or high-contrast structures in the original image.

[0037] Image restoration: Subtract the corrected noisy image N′ from the original image I. total The final denoised image I is obtained. denoised : I denoised =IN′ total Output: Output the denoised image I denoised .

[0038] Preferred choice of multi-resolution decomposition and reconstruction scheme: This scheme selects Laplacian pyramid decomposition and reconstruction to obtain the detail layer, where better noise separation can be performed. Therefore, other decomposition and reconstruction schemes, such as wavelet transform, are also possible. However, if wavelet transform is used, the high-frequency sub-band (detail layer) data has directionality, requiring corresponding adjustments to the subsequent directional filtering algorithm: for example, directly canceling directional filtering and directly separating noise from the wavelet-transformed sub-band using thresholding.

[0039] Preferably, the directional filtering algorithm is selected as follows: For performance considerations, this scheme uses template filters with 8 directions for directional filtering. The purpose of directional filtering is to restore the "actual signal" of this detail layer, so other directional filtering methods (such as Steerable filters, Gabor filters, or anisotropic diffusion filters mentioned in the text), or even deep learning (neural network) schemes can be used.

[0040] Preferably, in the directional filtering step, multiple filter banks in multiple directions can be used for processing, and the direction with the largest response or directional fusion can be selected to better track the local structure of the image.

[0041] Preferably, the threshold T in the noise cloamping step lowi and T high i It can be set to: T low i =-c·σ i T high i =c□σ i , where σ i Estimating the initial noise N′ for the i-th layer i The local or global standard deviation estimate, where c is a constant factor (e.g., 1.5, 2, 3, etc.).

[0042] Preferably, the threshold C in the global noise Clamp step low and C high The setting can be based on the global grayscale range or noise level of the original image I, for example, C. low =-d·σ global C high =d□σ global , σ global This is the noise standard deviation estimate for the flat region of the original image, where d is a constant factor.

[0043] The core idea of ​​this embodiment is: explicit noise separation and reconstruction: the noise is separated from the image, protected by double clipping, and then subtracted from the original image.

[0044] The key steps in this embodiment are: decomposition → directional filtering → difference operation (noise separation) → first limiting → noise reconstruction → second limiting → noise reduction of the original image.

[0045] The detailed protection mechanism of this embodiment is as follows: dual limiting: layered limiting prevents weak signals from being mixed in during noise estimation; global limiting prevents overshoot caused by reconstruction errors.

[0046] The noise reduction image generation method (difference) in this embodiment is: original image − reconstructed noisy image.

[0047] In summary, this embodiment employs a "separation + limiting" strategy, first separating noise from structure, then applying double limiting protection to the separated noise, and finally subtracting the noise from the original image. This path selection is not a conventional approach in the field, as traditional concepts tend to directly filter the image or detail layer, rather than explicitly constructing a noisy image and then subtracting it.

[0048] The synergistic effect of dual limiting in this embodiment: The improvements in this embodiment are not only reflected in each step itself, but also in the coordination between the steps, especially the organic combination of "first limiting" and "second limiting": First-level clipping (layered clamping): This applies to the initial noise estimate of each detail layer. Its purpose is to prevent useful signals such as weak edges and textures from being misidentified as noise. After extracting structural information through directional filtering, the initial noise estimate obtained from the difference operation may still contain detail signals with small amplitudes (such as weak edges). If these noises are used directly for reconstruction without clipping, these "pseudo-noises" will be subtracted from the original image, resulting in the loss of true details. First-level clipping sets a threshold range to truncate noise estimates that exceed a reasonable range, ensuring that the noise entering the reconstruction is "clean" noise.

[0049] The second clamping (global clamp): applied to the reconstructed full-resolution noisy image, its purpose is to prevent overshooting of noise estimation in certain areas due to the accumulation of errors during multi-resolution reconstruction. Even if noise estimation at each layer has been clamped, after reconstruction operations such as upsampling and stacking, the synthesized noisy image may have abnormally large values ​​in local areas (e.g., reconstruction artifacts near edges). Without the second clamping, these excessively large noise values ​​would be treated as noise and removed during the subtraction step, thus damaging strong edges or high-contrast structures in the image. The second clamping acts as a "safety valve," further ensuring the reliability of the final noise reduction result.

[0050] This dual-limiting mechanism forms a "layered control + global fallback" protection system, achieving multi-layered and all-round protection for useful signals. This synergistic effect cannot be achieved by simply superimposing a single limiting operation: a single limiting operation cannot solve the reconstruction error problem, while only two limiting operations can simultaneously eliminate the risk of detail loss in both the noise separation and reconstruction stages. While existing technologies protect details through classification and directional filtering, their protection mechanism is placed before the filtering process and does not address the error problem in the reconstruction process. The dual-limiting mechanism in this embodiment is placed after noise separation and reconstruction, which is completely different from the protection path of existing technologies, and its synergistic effect is neither disclosed nor implied in existing technologies.

[0051] Analysis of key steps / module modifications in this embodiment: 1. Directional filtering module: directional structure extraction in multi-resolution detail layers. In this technical field, directional filtering (such as Gabor and directional filtering) is indeed a common technique in image processing, often used for edge detection, texture analysis, etc.

[0052] However, the improvement in this embodiment is as follows: Unique application location: This embodiment applies directional filtering to the detail layers after multi-resolution decomposition, rather than the original image. The detail layers themselves contain information from different frequency bands. Applying directional filtering on this basis can more accurately extract the structural orientation within that frequency band, avoiding interference from frequency band aliasing on orientation estimation.

[0053] Purpose of Innovation: The purpose of using directional filtering in this embodiment is to obtain a "structural information image under noise-free conditions" so that noise can be separated through difference operations. This application is not common in existing technologies. Directional filtering is typically used for enhancement or detection, but this embodiment uses it as a pre-processing tool for "noise separation," achieving a fine decomposition of noise.

[0054] Connection with subsequent steps: The directional filtering result is directly used for difference calculation, and its accuracy directly affects the quality of subsequent noise estimation. Therefore, directional filtering is not an isolated technique in this embodiment, but a crucial link in the entire noise separation chain.

[0055] 2. Difference Operation Noise Extraction Module: An Innovative Application of Explicit Noise Separation In this technical field, it is common practice in image processing to obtain high-frequency or noise components by using the difference between the original signal and the filtered signal. For example, the Laplacian pyramid itself is constructed by difference.

[0056] However, the improvement in this embodiment is as follows: The objects of the difference operation differ: traditional Laplacian pyramid differences are performed between Gaussian images of different scales to obtain detail layers; while the difference operation in this embodiment is performed between the same detail layer and its directional filtering result to obtain noise estimation for that layer. Both the objects and purposes of this difference operation are entirely new.

[0057] The separated noise has a clear physical meaning: the difference obtained through the "original detail layer - structural information layer" represents the components in that layer that are independent of the directional structure, namely noise and a small number of weak details not captured by the directional filter. This separation method is more interpretable than directly setting the coefficients to zero in the transform domain and also provides a basis for subsequent clipping protection.

[0058] 3. Amplitude Limiting Module: A Clever Design of Dual Amplitude Limiting In this technical field, clamping is a basic digital signal processing operation used to limit a value within a certain range.

[0059] However, the improvement in this embodiment is as follows: The object and timing of the amplitude limiting are unique: the amplitude limiting operation in this embodiment is not performed directly on the original image or the filtered image, but on the separated noisy image. This approach of amplitude limiting the noise itself reflects the reverse thinking of "preserving details by starting with the noise", which is completely different from the conventional approach of "smoothing or thresholding the image".

[0060] Dual-limiting structural design: As mentioned above, the first and second limiting are at different stages in the process and solve different problems (noise separation error, reconstruction error), and neither can be omitted.

[0061] Adaptive setting of the limiting threshold: This embodiment further explains that the threshold is adaptively calculated based on the statistical characteristics (standard deviation) of noise, enabling the limiting operation to adapt to different image content and noise levels, avoiding the limitations of a fixed threshold. This adaptive mechanism makes limiting no longer a simple truncation, but an intelligent protection measure.

[0062] 4. Noise Reconstruction Module: Unique application for constructing full-resolution noisy images. In this technical field, multi-resolution reconstruction (such as Laplacian pyramid reconstruction) is a common technique in image processing and is often used for image fusion, super-resolution, etc.

[0063] However, the improvement in this embodiment is as follows: The object of reconstruction is noise: multi-resolution reconstruction is usually used to reconstruct the image itself, while this embodiment reconstructs noise estimation images of each layer, resulting in a full-resolution noise map. This concept of treating noise as a "reconstructable image" breaks through traditional thinking.

[0064] The reconstructed noise map is used as a subtraction mask: the reconstructed full-resolution noise map is used as a mask to be subtracted from the original image. This noise reduction method is equivalent to "adaptive noise removal". Compared with direct filtering, it can more accurately control the amount of noise removal and avoid over-smoothing.

[0065] 5. Image Restoration Module: Subtraction noise reduction from the original image to the noise-reducing image. In this technical field, image subtraction is the most basic arithmetic operation.

[0066] However, the improvement in this embodiment is as follows: The subtraction in this subtraction is the reconstructed noisy image: conventional denoising typically employs direct smoothing or filtering, which essentially process the image itself or its representation coefficients directly. In contrast, this embodiment subtracts a carefully constructed and double-limited noisy image, representing the noise components to be removed, rather than high-frequency information. This subtractive denoising is essentially a noise component removal process, rather than a simple frequency separation.

[0067] Dependence on the overall process: This subtraction step is the final step in the entire process, and its effectiveness depends on the synergistic effect of all preceding steps. Without accurate structure extraction through directional filtering, noise separation through difference operations, and protection through double clipping, this subtraction may damage details or leave residual noise. Therefore, the innovation of this module lies in its role within the overall system, rather than in its individual operation.

[0068] The second embodiment of the present invention: An X-ray image noise reduction system according to this embodiment includes: Memory, used to store computer programs; A processor for executing computer programs to implement an X-ray image noise reduction method as described above.

[0069] The third embodiment of the present invention: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the X-ray image noise reduction method described above.

[0070] Fourth embodiment of the present invention: This embodiment illustrates the application of the X-ray image denoising method described above in medical imaging or industrial inspection image denoising.

[0071] The fifth embodiment of the present invention (such as Figures 1-7 (as shown) like Figure 1 The X-ray image noise reduction method of this embodiment includes the following steps in sequence: S1: Input a 1024x1024 pixel low-dose digital X-ray image.

[0072] S2: Construct a 4-layer Laplace pyramid (k=4) to obtain the detail layers {L0, L1, ..., L k-1} and low-frequency substrate G k (G k = 0, meaning all pixel values ​​are set to 0. For example... Figure 2 .

[0073] S3: For each detail layer L i : S3.1: Apply a set of template filters in 8 directions (e.g., Figure 4 The maximum value of the response in each direction at each point (e.g., the direction with the smallest variance) is taken, and then fused (e.g., the mean value in that direction is taken) to obtain the directional filtering result F. i .

[0074] S3.2: Calculate N′ i =L i -Fi .

[0075] S3.3: Estimate N′ i global standard deviation σ i Set c=2.0 and perform Clamp: N i =clamp(N′ i , -2 σ i , 2σi).

[0076] S4: The noise from layers N0, N1, N2, and N3, along with the zero-value G4, are reconstructed into a 1024x1024 full-resolution noisy image N using a Laplacian pyramid upsampling and addition algorithm. total .like Figure 3 .

[0077] S5: Estimate the noise standard deviation σ of the background region in the original image. global Let d=2.5, for N total Perform a global clone: ​​N′ total =clamp(N total -2.5σ global 2.5σ global ).

[0078] S6: Calculate the denoised image: I denoised =IN′ total .

[0079] S7: Output I denoised The results showed that background noise was significantly suppressed, while the detailed structures of tissue textures (such as blood vessels and bones) were clearly visible. This was confirmed in online card membrane testing (see...). Figure 5 This noise reduction scheme will not reduce the precision of image line pairs.

[0080] The parameters k=4, c=2.0, d=2.5 are typical values ​​and do not represent fixed values. They can be adjusted according to the characteristics of the image.

[0081] like Figure 6 For the direct enhancement of clinical images of the present invention ( Figure 6 (Left side) and the enhanced noise reduction of this solution ( Figure 6 (Right side) Comparison diagram of effects.

[0082] Since directly comparing the noise level of an image before enhancement is affected by various factors such as the image's dynamic range, it is difficult to observe directly with the naked eye. Therefore, we enhanced the image using a multi-scale enhancement scheme commonly used for X-ray images, and the same enhancement method and processing parameters were used for both the left and right images.

[0083] As you can see, Figure 6 The enhancement result after noise reduction on the right side and Figure 6 The direct enhancement results on the left show no visible loss of information about large and small blood vessels and ribs, but the granular noise in the background is significantly reduced, and the overall image becomes clearer.

[0084] like Figure 7 For the line-pair image direct enhancement of the present invention ( Figure 7 (Left side) and the enhanced noise reduction of this solution ( Figure 7 (Right side) Comparison diagram of effects.

[0085] As you can see, Figure 7 The enhancement result after noise reduction on the right side and Figure 7 The direct enhancement results on the left show that the low contrast (disc-shaped) index at the top of the image and the spatial resolution (line pairs) index at the bottom of the image did not decrease, but the granular noise in the background was significantly reduced, and the image as a whole became clearer.

[0086] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0087] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for denoising X-ray images, characterized in that: The steps are as follows: Step 1: Obtain the original X-ray image I to be denoised; Step 2: Perform multi-resolution decomposition on the original X-ray image I to obtain a series of detail layer images {L0, L1, ..., L...} with different spatial frequencies. k-1 } and a low-frequency substrate image G k ; Step 3: For each layer of detail image L i (i=0,1,...,k-1) undergo directional filtering to obtain the filtered result F that preserves edge and texture directions. i ; Step 4: Calculate the detailed image L i With the filtering result F i The difference is used to obtain the initial noise estimate N for this layer. i =L i -F i ; Step 5: Estimate N from the initial noise. i Perform the first limiting operation to restrict its value to the interval [T]. low i ,T high i Within this range, the final estimated noise N for this layer is obtained. i ; Step 6: Reduce the noise {N0, N1, ..., N} after the first clipping operation on each layer. k-1 } and perform multi-resolution reconstruction on a low-frequency substrate with zero values ​​to synthesize a full-resolution noisy image N with the same size as the original X-ray image I. total ; Step 7: Process the reconstructed full-resolution noisy image N total Perform a second limiting operation to restrict its value to the interval [C]. low C high Within ], the corrected noisy image N′ is obtained. total ; Step 8: Subtract the corrected noise image N′ from the original X-ray image I. total The final denoised image I is obtained. denoised .

2. The X-ray image noise reduction method as described in claim 1, characterized in that, The multi-resolution decomposition is performed by decomposing and reconstructing the Laplacian pyramid; the detail layer image is the detail image of each layer of the Laplacian pyramid; the low-frequency base image is the bottom layer image of the Gaussian pyramid.

3. The X-ray image noise reduction method as described in claim 1, characterized in that, The directional filtering process employs at least one or a combination of the following filters: template-based directional filters, steerable filters, Gabor filters, and anisotropic diffusion filters.

4. The X-ray image noise reduction method as described in claim 1, characterized in that, The directional filtering process includes: filtering the detail layer image using filter banks in multiple directions, and selecting the filtering result corresponding to the direction with the largest response, or fusing the filtering results from multiple directions to obtain the filtering result F. i .

5. The X-ray image noise reduction method as described in claim 1, characterized in that, In the first limiting operation: T low i =-c·σ i T high i =c□σ i , where σ i Estimating the initial noise N′ for the i-th layer i The local or global standard deviation estimate, where c is a constant factor.

6. The X-ray image noise reduction method as described in claim 1, characterized in that, In the first clipping operation: the final estimated noise N i Right now: .

7. The X-ray image noise reduction method as described in claim 1, characterized in that, Threshold C in the second limiting operation low and C high C is adaptively determined based on the global grayscale range or noise level of the original X-ray image I: low =-d·σ global C high =d□σ global , where σ global This is the noise standard deviation estimate for the flat region of the original image, where d is a constant factor.

8. An X-ray image noise reduction system, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement an X-ray image noise reduction method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of an X-ray image noise reduction method as described in any one of claims 1 to 7.

10. The application of the X-ray image denoising method according to any one of claims 1 to 7 in medical imaging or industrial inspection image denoising.

Images