Image denoising method and device, electronic equipment, readable storage medium and chip

Abstract

The application provides an image denoising method and device, electronic equipment, readable storage medium and chip, and relates to the technical field of image denoising processing, wherein the image denoising method comprises: determining an X-ray original image; performing layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; determining a second image set according to the inter-layer interpolation residual error of the plurality of first image sets; performing nonlinear detail enhancement on the detail image of each resolution layer in the second image set to determine a first enhanced image; constructing a guide image based on the local contrast noise ratio of the detail image of the current level; performing guide filtering on the first enhanced image to determine the filter image corresponding to each resolution layer; and performing layer-by-layer up-sampling processing on the plurality of filter images to determine a denoising enhanced image corresponding to the X-ray original image. Through the scheme of the application, the robustness of X-ray image denoising enhancement is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image denoising processing, in particular to an image denoising method and device, electronic equipment, readable storage medium and chip. BACKGROUND

[0002] The X-ray original image has a very high dynamic range. When directly displaying the X-ray original image, an ordinary display cannot fully present the high dynamic range. The traditional method can only display part of the gray scale at a time, and cannot display the global contrast. At present, the X-ray original image is usually enhanced by using a multi-scale enhancement method. The image is analyzed on different scales. The existing process combines multiple filtering methods to denoise the image on the enhanced image, for example, a non-local mean filtering (Non-Local Means, NL-Means) algorithm and a three-dimensional block matching filtering (Block-matching and 3D filtering, BM3D) algorithm. However, the above filtering methods have high time complexity for denoising the image, and the robustness of the real-time processed image is insufficient. SUMMARY

[0003] The technical scheme of the present application aims to provide an image denoising method, device, electronic equipment, readable storage medium and chip, which can solve the problem of high time complexity of the existing multiple filtering methods for denoising the image and insufficient robustness of the real-time processed image.

[0004] Therefore, the technical scheme of the first aspect of the present application provides an image denoising method.

[0005] The technical scheme of the second aspect of the present application provides an image denoising device.

[0006] The technical scheme of the third aspect of the present application provides an electronic equipment.

[0007] The technical scheme of the fourth aspect of the present application provides a readable storage medium.

[0008] The technical scheme of the fifth aspect of the present application provides a chip.

[0009] In order to achieve the above object, the technical scheme of the first aspect of the present application provides an image denoising method, comprising: determining an X-ray original image; performing layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; determining a second image set according to the inter-layer interpolation residual of the plurality of first image sets; performing nonlinear detail enhancement on the detail image of each resolution layer in the second image set to determine a first enhanced image; constructing a guide image based on the local contrast noise ratio of the detail image of the current level; performing guided filtering on the first enhanced image with the guide image as the constraint to determine the filtered image corresponding to each resolution layer; and performing layer-by-layer up-sampling processing on the plurality of filtered images to determine a denoised enhanced image corresponding to the X-ray original image.

[0010] By the image denoising method provided by the present application, the X-ray original image of high dynamic range is subjected to multi-scale decomposition to determine the first image set and the second image set corresponding to a plurality of resolution layers, and the X-ray original image is taken as the initial level of the first image set. By the multi-scale decomposition manner, different gray scale regions are independently processed, the global contrast imbalance is avoided, and the weak signal is stretched in a targeted manner. The guide image is determined according to the detail image of each level, and the filtered and enhanced synthesized image is generated by using the details in the original image as a guide. By the guided filtering manner, the real local contrast of the original image is preserved, and the noise amplification of the X-ray original image is suppressed.

[0011] As can be understood, by the image denoising method provided by the present application, the X-ray original image is subjected to denoising enhancement, so that the denoised enhanced image can satisfy the high dynamic range, the contrast of the X-ray original image is improved, the image details are enhanced, and the demand of software for real-time processing of images is met.

[0012] In some technical schemes, optionally, the second image set is determined according to the inter-layer interpolation residual of the plurality of first image sets, comprising: determining the number K of levels of the plurality of resolution layers, wherein K is a positive integer; determining the low-pass image of each resolution layer in the first image set; performing bilinear interpolation up-sampling on the low-pass image of the Kth layer to restore the resolution of the low-pass image of the Kth layer to the resolution corresponding to the low-pass image of the K-1th layer; subtracting the up-sampled image from the low-pass image of the K-1th layer pixel by pixel to generate the detail image of the Kth layer; and determining the second image set according to the detail images of the plurality of resolution layers.

[0013] In the present scheme, the image of each resolution layer in the first image set is subtracted from the predicted image after the up-sampling and Gaussian convolution of the previous resolution layer to obtain a plurality of difference images as the detail images. The decomposition process of the second image set comprises four steps, i.e. low-pass filtering, down-sampling, interpolation and band-pass filtering. The image size is reduced by the down-sampling manner; and the image size is enlarged by the interpolation manner.

[0014] It can be understood that, by means of residual interpolation, the image details in the low-pass image in the first image set which are not covered by the low-frequency information of the previous resolution layer are extracted, and the robustness of X-ray original image denoising processing is improved. Moreover, by means of bilinear interpolation upsampling, the image reconstruction speed is improved while ensuring the image accuracy.

[0015] In some technical solutions, the multiple filtered images are optionally processed by layer-by-layer upsampling to determine the denoising enhanced image corresponding to the X-ray original image, comprising: determining the secondary low-pass image after upsampling of the low-pass image of the Kth layer; taking the secondary low-pass image as the compensation input of the first enhanced image of the Kth layer, and determining the filtered image of the Kth layer with the guide image of the Kth layer as the constraint; starting from the K-1th layer, performing bilinear interpolation upsampling on the filtered image as the compensation input of the first enhanced image of each resolution layer, determining the filtered images of the multiple resolution layers, and performing iteration; and iteratively performing to the resolution layer corresponding to the X-ray original image to output the denoising enhanced image.

[0016] In the present solution, starting from the lowest resolution corresponding layer, the low-pass image is upsampled to generate a secondary low-pass image as the compensation input of the first enhanced image of the lowest resolution layer. Starting from the next resolution layer, i.e., the K-1th layer, the upsampling result of the filtered image of the previous resolution layer is taken as the compensation input of each resolution layer, which is combined with the first enhanced image of the same resolution layer to perform guided filtering, and the operations of upsampling, compensation input and guided filtering are performed layer by layer. Iteration is performed to the resolution layer of the X-ray original image, and the compensation input and the filtered result are integrated to output the denoising enhanced image.

[0017] In some technical solutions, the multiple filtered images are optionally processed by layer-by-layer upsampling to determine the denoising enhanced image corresponding to the X-ray original image, further comprising: performing filtering processing on the first enhanced image according to the guide image to determine a secondary filtered image; determining the secondary filtered image corresponding to each resolution layer; determining the secondary low-pass image after upsampling of the low-pass image of the Kth layer; determining the filtered image of the Kth layer according to the secondary low-pass image and the secondary filtered image; starting from the K-1th layer, performing bilinear interpolation upsampling on the filtered image, and determining the filtered image corresponding to each resolution layer according to the filtered image after bilinear interpolation upsampling and the secondary filtered image corresponding to each resolution layer, and performing iteration; and iteratively performing to the resolution layer corresponding to the X-ray original image to output the denoising enhanced image.

[0018] In the scheme, the object of the guided filtering is adjusted on the basis that the up-sampling result of the filtered image of the previous resolution layer is inputted as compensation into the first enhanced image of the same resolution layer and then combined to perform guided filtering. The guided image and the first enhanced image are directly filtered to obtain a secondary filtered image. The up-sampling result of the previous resolution layer and the secondary filtered image are fused to determine the filtered image of the resolution layer.

[0019] In some technical solutions, the guided image is constructed based on the local contrast noise ratio of the detail image of the current level, including: calculating the square value of the original detail image of the current level pixel by pixel; determining at least one size parameter; determining a sliding window according to the size parameter; determining the average value of the square value in the sliding window; and performing square root operation on the average value to determine the guided image reflecting the local contrast noise ratio.

[0020] In the scheme, the original detail image up-sampled in the second image set is squared pixel by pixel, and then moved pixel by pixel with a specific window size to determine the average value of the moving window. The average result is squared to determine the guided image. The local contrast signal-to-noise ratio of the original detail image is calculated to quantify the relative relationship between the contrast and background noise in the original detail image. The original detail image includes high-frequency details of the image.

[0021] It can be understood that, by calculating the root mean square of the images in the second image set to determine the guided image, the robustness of the noise reduction enhancement is improved while further preserving the high-frequency details in the X-ray original image. Moreover, by moving the calculation of the sliding window pixel by pixel, the computational complexity is small, the system calculation efficiency is improved, and the real-time performance of the X-ray original image display is improved.

[0022] In some technical solutions, constructing the guided image further includes: determining a down-sampled image in the first image set; determining a plurality of pixel points in the down-sampled image; determining a background average value of a local window centered on the pixel point; determining an average edge value according to the background average value and the pixel point; determining a local contrast corresponding to the pixel point according to the average edge value; and determining the guided image according to the local contrast.

[0023] The formula of the average edge value is:

[0024] ;

[0025] wherein, the average edge value is, the Laplace operator is, the pixel point is i, and j is the row number and column number of the pixel point, respectively.

[0026] The formula of the local contrast is:

[0027] ;

[0028] wherein, is the local contrast, is the average edge value, is a pixel point, i is a pixel point row number, and j is a pixel point column number.

[0029] In the scheme, the construction of the guide image further comprises: the low-pass image in the first image set is constructed by using the local contrast definition based on edge detection. Wherein, the down-sampling image of each resolution layer in the first image set is determined, the Laplace operator of any pixel point in the down-sampling image is calculated, and the average edge value of the down-sampling image corresponding to the resolution layer is determined. And the local contrast of the down-sampling image is determined according to the average edge value, and the guide image is determined according to the local contrast of each pixel point in the down-sampling image.

[0030] The technical scheme of the second aspect of the application provides an image denoising device, comprising: an image determination module for determining an X-ray original image; a Gaussian processing module for performing layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; an interpolation residual module for determining a second image set according to the interlayer interpolation residual of the plurality of first image sets; a detail enhancement module for performing nonlinear detail enhancement on the detail image of each resolution layer in the second image set to determine a first enhanced image; a guide construction module for constructing a guide image based on the local contrast noise ratio of the current level of the detail image; a guide filtering module for performing guide filtering on the first enhanced image with the guide image as a constraint to determine a filter image corresponding to each resolution layer; and a denoising enhancement module for performing layer-by-layer up-sampling processing on the plurality of filter images to determine a denoising enhanced image corresponding to the X-ray original image.

[0031] The technical scheme of the third aspect of the application provides an electronic device, comprising a processor, a memory, and a program or instruction stored on the memory and executable on the processor, the program or instruction being executed by the processor to implement the steps of the image denoising method in the first aspect.

[0032] The technical scheme of the fourth aspect of the application provides a readable storage medium, the readable storage medium stores a program or instruction, and the program or instruction is executed by a processor to implement the steps of the image denoising method in the first aspect.

[0033] The technical scheme of the fifth aspect of the application provides a chip, the chip comprising a processor and a communication interface, the communication interface and the processor being coupled, the processor being used to run a program or instruction to implement the steps of the image denoising method in the first aspect.

[0034] Additional aspects and advantages of the technical solutions of the present application will become apparent in the description part below, or be learned by practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0036] Figure 2 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0037] Figure 3 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0038] Figure 4 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0039] Figure 5 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0040] Figure 6 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0041] Figure 7 A structural schematic block diagram of an image denoising device according to one embodiment of the present application is shown;

[0042] Figure 8 A structural schematic block diagram of an electronic device according to one embodiment of the present application is shown;

[0043] Figure 9 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0044] Figure 10 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0045] Figure 11 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0046] Figure 12 A construction schematic diagram of a guide image according to one embodiment of the present application is shown;

[0047] Figure 13 A construction schematic diagram of a guide image according to one embodiment of the present application is shown;

[0048] Figure 14 A flowchart of an image denoising method according to one embodiment of the present application is shown;

[0049] wherein, Figure 7 and Figure 8 Correspondence between reference signs and names of parts in the drawings is as follows:

[0050] 900: image denoising apparatus; 902: image determination module; 904: Gaussian processing module; 906: interpolation residual module; 908: detail enhancement module; 910: guide construction module; 912: guide filtering module; 914: denoising enhancement module; 1000: electronic device; 1109: memory; 1110: processor. DETAILED DESCRIPTION

[0051] In order to enable a more complete understanding of the above-mentioned objects, features and advantages of the embodiments of the present application, the embodiments of the present application will be described in further detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0052] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the application, but embodiments of the present application can also be implemented in other ways different from those described herein, and therefore the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0053] X-ray raw images usually have a very high dynamic range, and pixel values can cover a wide range from very low (air region) to very high (metal or bone region). When directly displaying the raw image, it can look very dark or have low contrast, because ordinary displays cannot fully present high dynamic range. Traditional methods such as window width and window level adjustment are used to enhance contrast, but this method can usually display only a part of the gray scale at a time, and cannot display the whole. Therefore, it is usually necessary to enhance the X-ray raw image, and the current enhancement algorithm is multi-scale enhancement. Multi-scale enhancement analyzes the image at different scales, and the scale decomposition methods include pyramid, wavelet transform, etc. After decomposition, the detail image is stretched nonlinearly at different sizes to highlight the details. From the perspective of image quality, the most significant disadvantage of multi-scale enhancement is that it simultaneously enhances the noise in the image, especially at high frequency scales. The existing process practices modern filtering methods such as non-local mean (NLM) algorithm and three-dimensional block matching filter (BM3D) algorithm on the enhanced image. These algorithms have significant denoising effect and can obtain the highest peak signal-to-noise ratio, but have high time complexity and cannot meet the requirements of real-time image processing by software.

[0054] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which: Figures 1 to 14The image denoising method, device, electronic device, readable storage medium and chip provided by the embodiments of the present application are described in detail through specific embodiments and application scenarios. The embodiments provide an image denoising method, as shown in Figure 1

[0055] Step S100: determining an X-ray original image;

[0056] Step S102: performing layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set including multiple resolution layers;

[0057] Step S104: determining a second image set according to the inter-layer interpolation residual error of the multiple first image sets;

[0058] Step S106: performing nonlinear detail enhancement on the detail image of each resolution layer in the second image set to determine a first enhanced image;

[0059] Step S108: constructing a guide image based on the local contrast noise ratio of the detail image of the current level;

[0060] Step S110: performing guided filtering on the first enhanced image with the guide image as a constraint to determine a filter image corresponding to each resolution layer;

[0061] Step S112: performing layer-by-layer up-sampling processing on the multiple filter images to determine a denoising enhanced image corresponding to the X-ray original image.

[0062] By the image denoising method provided by the present application, the X-ray original image of high dynamic range is subjected to multi-scale decomposition to determine a first image set and a second image set corresponding to multiple resolution layers, and the X-ray original image is taken as an initial level of the first image set. By the multi-scale decomposition manner, different gray scale regions are independently processed, the global contrast imbalance is avoided, and the weak signal is stretched in a targeted manner. The guide image is determined according to the detail image of each level, and the filter enhanced composite image is generated by using the details in the original image as a guide. By the guided filtering manner, the real local contrast of the original image is retained, and the noise amplification of the X-ray original image is suppressed.

[0063] It can be understood that, by the image denoising method provided by the present application, the X-ray original image is subjected to denoising enhancement, so that the denoising enhanced image can meet the high dynamic range, the contrast of the X-ray original image is improved, the image details are enhanced, and the demand of software for real-time processing of images is met.

[0064] ​Specifically, the input X-ray original image is iteratively processed, low-pass filtered, and down-sampled to obtain a plurality of low-pass images with decreasing resolution, and a first image set is constructed by a plurality of low-pass image layers with decreasing resolution. The inter-layer residual error in the adjacent first image set is calculated, the low-frequency approximation corresponding to the hierarchical level is up-sampled to the resolution of the previous hierarchical level, and the pixel-by-pixel subtraction is performed to construct a second image set, and the high-frequency details of each resolution layer at the current scale, such as edges, textures, and noises, are determined. The contrast of each layer of detail image is stretched to enhance the weak signal corresponding to each resolution layer to obtain a first enhanced image. The guide image is constructed based on the local contrast-to-noise ratio of the current resolution level, the guide image is generated by squaring the pixel-by-pixel square and averaging in a fixed window, and the local signal-to-noise ratio is quantified. The high value of the local signal-to-noise ratio represents the effective edge in the X-ray original image, and the low value of the local signal-to-noise ratio represents the flat noise region in the X-ray original image. Guided by the local signal-to-noise ratio, the first enhanced image is filtered to make the noise region of the X-ray original image more strongly suppressed and the image details in the edge region of the X-ray original image are retained. In the process of reconstructing the denoising image, the iteration starts from the lowest resolution level, and the filtered image of each resolution level is superimposed and filtered to the next resolution level, and after iteration to the initial level, a denoising enhanced image is obtained, which is the denoising enhanced output result of the X-ray original image.

[0065] Optionally, the first image set includes images corresponding to each resolution layer in the Gaussian pyramid, and the second image set includes images corresponding to each resolution layer in the Laplacian pyramid.

[0066] Illustratively, the original image is taken as the 0th layer of the Gaussian pyramid, a Gaussian kernel (5x5) is used to convolve it, and then the next layer image is obtained by down-sampling (removing even rows and columns) the convolved image. Take this image as input, repeat the convolution and down-sampling operation to get the image of the next layer, and repeat the iteration several times to form a pyramid-shaped image data structure, i.e. the Gaussian pyramid.

[0067] Optionally, the Gaussian pyramid includes a plurality of low-pass filters, and the cut-off frequency gradually increases by a factor of 2 from the previous layer to the next layer, and the plurality of images in the Gaussian pyramid span a large frequency range to meet the denoising needs of different X-ray images.

[0068] Optionally, the Laplacian pyramid is used to reconstruct the image, and each layer of image of the Gaussian pyramid is subtracted from the predicted image after up-sampling and Gaussian convolution of the previous layer image to obtain a series of difference images for Laplacian operator decomposition image.

[0069] The forming process of the Laplace image includes low-pass filtering and down-sampling the original image to obtain a coarse-scale approximate image, i.e., a low-pass approximate image obtained by decomposition, interpolating the approximate image, filtering, and calculating the interpolation of the original image to obtain a band-pass component of the decomposition. The next level of decomposition is performed on the obtained low-pass approximate image, and the multi-scale decomposition is completed iteratively.

[0070] Optionally, the image is smoothed by using a low-pass filter, which is usually a Gaussian filter, and the filter size is usually 5x5.

[0071] Optionally, the image after smoothing is sampled by interlaced rows and columns, and the process of continuously reducing the image resolution is called down-sampling, and the process of continuously increasing the image resolution is called up-sampling, and 0 is inserted in the middle of the rows and columns.

[0072] Optionally, adaptive noise suppression is performed by using a guide image, the guide image is used as a local signal-to-noise ratio proxy, weak filtering is performed on the image edge area corresponding to a high local signal-to-noise ratio to retain the structural details of the X-ray original image, and strong filtering is performed on the flat area of the image corresponding to a low local signal-to-noise ratio to suppress image noise. Quantify the noise level by the local signal-to-noise ratio, so that the filtering strength strictly matches the noise intensity. Compared with fixed parameter filtering, the guided filtering by using the guide image reduces image edge blur and noise residue.

[0073] Optionally, the guide image directly uses the original detail data before enhancement, and the original detail data comes from the second image set, avoiding the case where the noise in the X-ray original image is confirmed as an effective signal due to the enhancement operation.

[0074] Optionally, the time complexity of the guided filtering is O(N), the parallelism of the window operation is high, a single frame is processed through a central processing unit, and the guided filtering improves the efficiency of image denoising calculation, thereby realizing the real-time processing of the X-ray original image by software.

[0075] Optionally, a limited contrast adaptive histogram equalization algorithm is used for detail enhancement, and the contrast limit threshold corresponding to the detail enhancement is inversely proportional to the image resolution of multiple levels. The higher the resolution of the image, the smaller the contrast limit threshold, and the smaller the image stretching intensity; the lower the resolution of the image, the larger the contrast limit threshold, and the greater the image stretching intensity.

[0076] Optionally, the images of multiple resolution levels are fused in a way of filtering and up-sampling layer by layer from the lowest resolution layer to the highest resolution layer to improve the real-time performance of X-ray original image denoising.

[0077] Optionally, the images of the plurality of resolution layers are processed synchronously, and after the filtered images of each resolution layer are determined, interpolation fusion is performed to improve the accuracy and robustness of X-ray original image denoising.

[0078] Exemplarily, the X-ray original image is used in scenes such as industrial detection and medical imaging.

[0079] In some embodiments, optionally, as shown in Figure 2 The second image set is determined according to the inter-layer interpolation residual error of the plurality of first image sets, including:

[0080] Step S1040: Determine the number K of the plurality of resolution layers, wherein K is a positive integer;

[0081] Step S1042: Determine the low-pass image of each resolution layer in the first image set;

[0082] Step S1044: Perform bilinear interpolation up-sampling on the low-pass image of the Kth layer to restore the resolution of the low-pass image of the Kth layer to the resolution corresponding to the low-pass image of the K-1th layer;

[0083] Step S1046: Subtract the up-sampled image from the low-pass image of the K-1th layer pixel by pixel to generate a detail image of the Kth layer;

[0084] Step S1048: Determine the second image set according to the detail images of the plurality of resolution layers.

[0085] In this embodiment, the image of each resolution layer in the first image set is subtracted from the predicted image after up-sampling and Gaussian convolution of the previous resolution layer to obtain a plurality of difference images as detail images. The decomposition process of the second image set, i.e., the Laplacian pyramid, includes four steps, i.e., low-pass filtering, down-sampling, interpolation and band-pass filtering. The image size is reduced by down-sampling; the image size is enlarged by interpolation.

[0086] It can be understood that by means of residual interpolation, the image details in the low-pass image in the first image set that are not covered by the low-frequency information of the previous resolution layer are extracted, and the robustness of X-ray original image denoising processing is improved. And by means of bilinear interpolation up-sampling, the image reconstruction speed is improved while ensuring the image accuracy.

[0087] Specifically, the total number of layers K is set according to the resolution of the X-ray original image, and the value of K determines the decomposition granularity. The K-th layer image includes the key structure in the X-ray original image. For example, the key structure includes the minimum visible size of the metal crack. A Gaussian image is generated by fixing the Gaussian kernel. The image of each resolution layer is enlarged to the previous resolution layer by bilinear interpolation, and a detail image of the previous resolution is output. The detail image includes high-frequency information of the resolution layer, such as image edges and image noise.

[0088] Optionally, the generation of the first image set needs to bind a 5*5 Gaussian kernel function for layer-by-layer low-pass filtering, and the standard deviation corresponding to the low-pass filtering is a preset value, so as to ensure the consistency of the frequency band separation.

[0089] Optionally, the downsampling process adopts an interlaced column sampling strategy, and discards a plurality of pixels in even rows and even columns. The image after discarding the pixels is determined as the down-sampled low-pass image. The pyramid structure is maintained by discarding the pixels in the interlaced column manner.

[0090] Optionally, in the bilinear interpolation upsampling operation, the target pixel value is generated by the weighted average value of the nearest 4 neighboring pixels, and the weight is linearly determined by the sub-pixel offset.

[0091] Optionally, a high-bit-width protection mechanism is introduced in the residual subtraction process, and a 16-bit signed integer is used to store the intermediate result to prevent information truncation.

[0092] Exemplarily, the resolution of the X-ray original image is 2048*2048, and the resolution of the K-th layer is 128*128, wherein the magnification of each adjacent resolution layer is the same.

[0093] Optionally, when the X-ray is used in a medical scene, the upsampling interpolation algorithm is switched to a bicubic interpolation to improve the soft tissue contrast, and a hierarchical reuse mechanism is introduced to allow the K-th layer low-pass image to directly reuse the K+1-th layer upsampling result to reduce the operation load.

[0094] In some embodiments, as shown in Figure 3 Optionally, the plurality of filtered images are subjected to layer-by-layer upsampling processing to determine a denoising enhanced image corresponding to the X-ray original image, including:

[0095] Step S1120: determining a secondary low-pass image after upsampling of the K-th layer low-pass image;

[0096] Step S1122: taking the secondary low-pass image as a compensation input of the first enhanced image of the K-th layer, and determining a filtered image of the K-th layer by taking the guide image of the K-th layer as a constraint.

[0097] Step S1124: starting from the K-1 layer, bilinear interpolation upsampling is performed on the filtered image as the compensation input of the first enhanced image corresponding to each resolution layer, the filtered images corresponding to the plurality of resolution layers are determined, and an iteration operation is performed;

[0098] Step S1126: iteration is performed to the resolution layer corresponding to the X-ray original image, and a denoising enhanced image is output.

[0099] In this embodiment, starting from the lowest resolution corresponding layer, the low-pass image is upsampled to generate a secondary low-pass image as the compensation input of the first enhanced image of the lowest resolution layer. Starting from the next resolution layer, i.e., the K-1 layer, the upsampling result of the filtered image of the previous resolution layer is taken as the compensation input for each resolution layer, which is combined with the first enhanced image of the same resolution layer and then guided filtering is performed, and the operations of upsampling, compensation input and guided filtering are performed layer by layer. Iteration is performed to the resolution layer of the X-ray original image, and the compensation input and the filtered result are integrated to output a denoising enhanced image.

[0100] It can be understood that, by layer-by-layer passing the upsampling compensation of the filtered image, noise suppression is realized while preserving multi-scale details, the reconstruction process is strictly matched with the resolution of the original image, and the robustness of X-ray original image denoising enhancement is improved.

[0101] Specifically, when performing layer-by-layer upsampling processing on the plurality of filtered images, first, the Kth layer of the lowest resolution level is processed, the Kth layer low-pass image is processed by bilinear interpolation upsampling to generate a secondary low-pass image with a resolution matching the K-1 layer, the secondary low-pass image is taken as the compensation input of the first enhanced image of the current resolution layer, and guided filtering is performed with the Kth layer guide image as the constraint to output the Kth layer filtered image. The processing mode of the lowest resolution level is different from that of the plurality of resolution levels in the image pyramid. In the process of sequentially processing the plurality of resolution levels in the image pyramid, the filtered image output by the previous resolution level is bilinearly interpolated and upsampled to generate a compensation input matching the current resolution layer. The result of the compensation input is provided to the first enhanced image of the current layer, and guided filtering is performed with the guide image of the same resolution level as the constraint to determine the filtered image of the current resolution layer. The above steps are repeated until all intermediate layers are processed. Finally, the process is performed to the resolution layer corresponding to the original image, the 2nd layer filtered image is upsampled and input to the 1st layer, guided filtering is performed on the first enhanced image of the 1st layer according to the guide image of the 1st layer, and a denoising enhanced image is output.

[0102] For example, the low-pass image of the lowest resolution layer (128×128 resolution) is taken and enlarged by a factor of 2 (resolution increased to 256×256) using bilinear interpolation to generate a secondary low-pass image. The secondary low-pass image serves as the substrate for carrying high-frequency details in the current layer. The enlarged image is superimposed on the first layer image of the current resolution layer. Using the guiding image of this resolution layer as a constraint, noise in the superimposed image is dynamically suppressed, with strong filtering in flat areas and weak filtering in edge areas, generating the preliminary reconstruction result of the Kth layer. The reconstruction result of the Kth layer is input to the K-1 layer, and bilinear interpolation is performed on the reconstruction result of the Kth layer (increasing the resolution to 512×512). The interpolated image is used as compensation input and provided to the first enhanced image of the K-1 layer for guiding filtering to determine the reconstruction result of the K-1 layer. After multiple iterations, the original resolution denoised and enhanced image (resolution 2048×2048) is output at the first layer, i.e., the global denoised and enhanced image. The denoised and enhanced image includes full-scale information, preserving not only the microscopic details of each resolution layer but also adjusting the image contrast, thus improving the robustness of the denoised and enhanced image.

[0103] Optionally, the multiple resolution layers in the image pyramid include layers K-1 to 1, where layer 1 is the resolution layer corresponding to the original X-ray image.

[0104] Optionally, guided filtering is performed layer by layer on the first enhanced image after compensation input. The guided image comes from the detail layer before image enhancement. The detail layer includes the low contrast characteristics of the original X-ray image. The detail layer can be used directly to reverse the noise reduction of the first enhanced image after enhancement.

[0105] Optionally, in industrial applications, a pixel-level parallel scheme is enabled, with multiple resolution layers independently allocated computing units. Upsampling employs a simplified nearest-neighbor interpolation method, and only 16-bit vertex data is retained when data is transferred between multiple resolution layers. This reduces data transmission volume and lowers the generation latency of the original X-ray image while enhancing and reducing image noise.

[0106] Optionally, in high-precision medical applications, the guided filtering includes dual-channel joint filtering. The main channel of the dual-channel joint filtering uses conventional edge-preserving filtering, while the secondary channel is based on morphological noise modeling from organ anatomical atlases. Upsampling in each resolution layer is performed using bicubic interpolation. The data bit width passed between multiple resolution layers is widened to 32-bit floating-point. This enhances image texture and sharpness while simultaneously reducing noise.

[0107] In some embodiments, optionally, such as Figure 4 As shown, the process involves performing layer-by-layer upsampling on multiple filtered images to determine the denoised and enhanced image corresponding to the original X-ray image. This also includes:

[0108] Step S1130: filtering the first enhanced image according to the guide image to determine a secondary filtered image;

[0109] Step S1132: determining the secondary filtered image corresponding to each resolution layer;

[0110] Step S1134: determining a secondary low-pass image after up-sampling the low-pass image of the Kth layer;

[0111] Step S1136: determining the filtered image of the Kth layer according to the secondary low-pass image and the secondary filtered image;

[0112] Step S1138: starting from the K-1th layer, performing bilinear interpolation up-sampling on the filtered image, determining the filtered image corresponding to each resolution layer according to the bilinear interpolation up-sampled filtered image and the secondary filtered image corresponding to each resolution layer, and performing iteration operation;

[0113] Step S1140: iteratively performing to the resolution layer corresponding to the X-ray original image, and outputting the denoised enhanced image.

[0114] In the embodiment, the object of the guided filtering is adjusted on the basis that each resolution layer originally uses the up-sampled result of the filtered image of the previous resolution layer as a compensation input, combines the first enhanced image of the same resolution layer, and then performs guided filtering. The guided image and the first enhanced image are directly filtered to obtain a secondary filtered image. In the resolution layer, the up-sampled result of the previous resolution layer and the secondary filtered image are fused to determine the filtered image of the resolution layer.

[0115] It can be understood that, by adjusting the object of the guided filtering, the high-frequency detailed information after enhancement is directly filtered to adapt to the scene where the X-ray penetration is insufficient, and the adaptability of the image denoising method to different use scenarios is improved.

[0116] Specifically, the guiding image of the same resolution layer directly performs guided filtering on the first enhanced image to determine the secondary filtered image. The lowest resolution layer, layer K, is processed by upsampling the low-pass image of layer K through bilinear interpolation to generate a secondary low-pass image with a resolution matching layer K-1. The secondary filtered image and the secondary low-pass image are then synthesized to output the filtered image of layer K. The processing method for the lowest resolution layer differs from that for multiple resolution layers in the image pyramid. During the sequential processing of multiple resolution layers in the image pyramid, the guiding image of the same resolution layer directly performs guided filtering on the first enhanced image to generate a secondary filtered image, which serves as the detail layer after the initial denoising. The low-pass image of the previous resolution layer is bilinearly upsampled to the current resolution layer. The upsampled result is synthesized with the secondary filtered image of the current resolution layer to determine the filtered image for that resolution layer. After upsampling to the original resolution, the denoised enhanced image is output.

[0117] By directly applying guided filtering to the guide image and the first enhanced image, noise is pre-filtered before injection into the compensation substrate, thus avoiding noise contamination in image reconstruction. Secondary filtering is used to further remove image noise, and multi-resolution layer stacking provides a clean low-frequency background for the compensation substrate, improving the robustness of noise reduction processing of the original X-ray image.

[0118] In some embodiments, optionally, such as Figure 5 As shown, a guide image is constructed based on the local contrast-to-noise ratio of the detail image at the current level, including:

[0119] Step S1080: Calculate the square value pixel by pixel for the original detail image of the current level;

[0120] Step S1082: Determine at least one size parameter;

[0121] Step S1084: Determine the sliding window based on the size parameters;

[0122] Step S1086: Determine the average value of the squared values ​​within the sliding window;

[0123] Step S1088: Perform a square root operation on the average value to determine the guide image that reflects the local contrast-to-noise ratio.

[0124] In this embodiment, the original detail image obtained by upsampling from the second image set is squared pixel by pixel, and then moved pixel by pixel with a specific window size to determine the average value of the moving window. The square root of the average result is then taken to determine the guide image. By calculating the local contrast-to-noise ratio of the original detail image, the relative relationship between contrast and background noise in the original detail image is quantified. The original detail image includes high-frequency details of the image.

[0125] It can be understood that, by calculating the root mean square of the images in the second image set to determine the guide image, the high frequency details in the X-ray original image are further retained while improving the robustness of the noise reduction enhancement. Moreover, by means of sliding window pixel-by-pixel moving calculation, the operation amount is small, the system calculation efficiency is improved, and the real-time performance of the X-ray original image display is improved.

[0126] Specifically, the high frequency details extracted in the second image set of the current resolution level, i.e. the original detail image, are determined. The original detail image includes effective edges and noise. The size parameter of the sliding window is determined according to the industrial scene to meet the needs of large structure scanning or micro-defect detection. The sliding window is a rectangular window. The larger the size parameter is, the larger the side length of the sliding window is, and the larger the path range scanned by the sliding window is. By means of pixel-by-pixel moving, each part of the original detail image is scanned to ensure full coverage. The arithmetic mean of the square values of all pixels in the sliding window is calculated to quantify the local contrast noise ratio to obtain the guide image.

[0127] Optionally, the guide image is a local standard deviation distribution map.

[0128] Optionally, the size parameter includes but is not limited to 3x3 and 5x5.

[0129] Optionally, the generation of the secondary filtered image needs to bind the adaptive window guided filtering technology. The window size is dynamically adjusted according to the current layer image entropy (when the entropy value is greater than 7.0, a 3x3 window is used, and when the entropy value is less than or equal to 7.0, a 5x5 window is used); the filtering output is stored in high-bit-width hybrid precision. The synthesis process of the secondary low-pass image superimposes a nonlinear compensation factor: when a metal area (gray value>8000) is detected, a 0.5-1.2 times dynamic gain is applied to the secondary low-pass image. By means of dynamic compensation, the thick workpiece penetration attenuation of X-ray is effectively compensated. In the iteration operation, the interlayer noise coupling detection is added. If the structural similarity (Structural Similarity, SSIM) value of the adjacent layer filtered image is greater than 0.98, the residual error discarding mechanism is started, only 10% of the high frequency components of the current layer are reserved to suppress the artifact propagation, reduce the residual image content in the filtered image in multiple resolution layers, and improve the robustness of the noise reduction enhanced image.

[0130] In some embodiments, as shown in Figure 6 , the guide image is constructed, further comprising:

[0131] Step S1090: determining a down-sampled image in the first image set;

[0132] Step S1092: determining a plurality of pixel points in the down-sampled image;

[0133] Step S1094: determining a background average value of the local window centered on the pixel point;

[0134] Step S1096: determining an average edge value according to the background average value and the pixel point;

[0135] Step S1098: determining a local contrast corresponding to the pixel point according to the average edge value;

[0136] Step S1100: determining a guide image according to the local contrast;

[0137] The formula of the average edge value is:

[0138]

[0139] The average edge value is The Laplacian is The pixel point is i is a row number of the pixel point, and j is a column number of the pixel point;

[0140] The formula of the local contrast is:

[0141]

[0142] The local contrast is The average edge value is The pixel point is i is a row number of the pixel point, and j is a column number of the pixel point.

[0143] In the embodiment, the construction of the guide image further includes: constructing the low-pass image in the first image set by using the local contrast definition based on edge detection. Wherein, the down-sampling image of each resolution layer in the first image set is determined, the Laplacian of any pixel point in the down-sampling image is calculated, and the average edge value of the down-sampling image corresponding to the resolution layer is determined. And the local contrast of the down-sampling image is determined according to the average edge value, and the guide image is determined according to the local contrast of each pixel point in the down-sampling image.

[0144] It can be understood that the standard Laplacian filter is susceptible to uniform region noise interference, and the way of determining the local contrast by using the normalized contrast formula separates the edge and the background of the low-pass image in each resolution layer, improves the effective edge recognition rate, enhances the nonlinear characteristics of the X-ray image, and improves the robustness of the noise reduction and enhancement of the X-ray original image.

[0145] Optionally, by using the way of calculating the Laplacian, the workload of the central processing unit for calculating multiple resolution layers is reduced, and the calculation efficiency is improved.

[0146] ​​​​Optionally, when the average edge value of each resolution layer is detected to be suddenly changed, the anti-aliasing operator is switched to reduce the probability of occurrence of the X-ray scanning metal edge artifact.

[0147] Exemplarily, by using the mathematical framework of "Laplacian weighted average and normalized contrast calculation", the two difficult problems of weak edge annihilation in high dynamic range X-ray images and metal artifact interference are solved, the defect recognition accuracy is improved in industrial flaw detection, and the anatomical structure fidelity is improved in medical imaging. The scene realizes double breakthrough, and at the same time guarantees the real-time processing performance (4K image full process ≤20ms).

[0148] In one specific embodiment, the existing multi-scale enhancement algorithm process is shown in Figure 9 , the left is image pyramid decomposition, the right is image synthesis, and the middle is detail enhancement link. The image pyramid is a series of resolution gradually reduced images arranged in the form of a pyramid, which are derived from the same original image, as shown in Figure 9 , which includes n+1 layers of images. This layer-by-layer image is metaphorically referred to as a pyramid. The image pyramid can be obtained by hierarchical downsampling until a certain termination condition is reached. In the downsampling process, the higher the level, the lower the resolution.

[0149] Obtaining the image pyramid generally includes:

[0150] 1. Smoothing the image using a low-pass filter (LP), which is usually a Gaussian filter, and the filter size is usually 5x5.

[0151] 2. Interlaced column sampling is performed on the smoothed image, and the process of continuously reducing the image resolution is called downsampling ( Figure 9 ), and the process of continuously increasing the image resolution is called upsampling ( Figure 9 ). The usual practice is to insert 0 in the middle of the row and column.

[0152] Among them, the Gaussian pyramid is the most basic image tower.

[0153] Principle: First, the original image is taken as the 0th layer of the Gaussian pyramid, and then the convolution is performed using the Gaussian kernel (5x5). Then, the next layer of image is obtained by downsampling (removing even rows and columns) the convolved image. This image is taken as the input, and the convolution and downsampling operations are repeated to obtain the next layer of image. The iteration is repeated several times to form a pyramid-shaped image data structure, i.e. the Gaussian pyramid.

[0154] Gaussian pyramid is a series of down-sampled images obtained by Gaussian smoothing and sub-sampling, that is, the K-th layer of Gaussian pyramid can obtain the K+1-th layer of Gaussian image by smoothing and sub-sampling. The Gaussian pyramid contains a series of low-pass filters, and the cut-off frequency from the upper layer to the lower layer is gradually increased by a factor of 2, so the Gaussian pyramid can span a large frequency range.

[0155] Laplacian pyramid is used to reconstruct graphics, that is, prediction residual, to restore the image to the greatest extent. For example, a small image is reconstructed into a large image. Principle: subtract the prediction image obtained by up-sampling and Gaussian convolution of each layer image of the Gaussian pyramid from the previous layer image to obtain a series of difference images, that is, the Laplacian decomposition image.

[0156] The formation process of the Laplacian image is roughly as follows: low-pass filtering and down-sampling are performed on the original image to obtain a coarse-scale approximate image, that is, the low-pass approximate image obtained by decomposition, and the interpolation, filtering and calculation of the interpolation of the original image are performed to obtain the band-pass component of the decomposition. The next level of decomposition is performed on the obtained low-pass approximate image, and the multi-scale decomposition is completed iteratively. It can be seen that the decomposition process of the Laplacian pyramid includes four steps: low-pass filtering, down-sampling (reducing the size), interpolation (enlarging the size) and band-pass filtering (image subtraction).

[0157] The most common processing of detail enhancement is gray value stretching, which expands the original gray value range to a larger range or adjusts it to a more appropriate range by linear or nonlinear remapping of the original gray value; high-value processing such as adaptive contrast stretching, histogram specification, limited contrast adaptive histogram equalization (CLAHE) and the like.

[0158] Optionally, Figure 9 In g(n) of the Gaussian pyramid, d(n) is the up-sampled detail image in the Laplacian pyramid, d'(n) is the first enhanced image after detail enhancement, and e(n) is the filtered image.

[0159] The overall flowchart of the image denoising method is shown in Figure 10 Guided filtering is used at each layer, and the amplified noise is controlled while the details are enhanced layer by layer. Figure 10 It is a basic implementation, and guided filtering is performed on the synthesized e(n) layer by layer, and the guided image directly comes from the detail layer d(n) before enhancement, because d(n) reflects the low-contrast characteristics of the original layer, and direct use of d(n) can reversely control the enhanced e(n) to reduce image noise.

[0160] Wherein, the guided filter is an edge protection filter algorithm based on a local linear model, and is widely used in image denoising, detail enhancement and other tasks. The core idea of the guided filter is to use a guide image to guide the filtering process. The design of the guide image determines the filtering effect. The output image is a linear transformation of the guide image in a local window. The guided filter has the following advantages: edge preservation: the edge of the output image is consistent with the guide image; no gradient reversal: compared with the bilateral filter, no gradient reversal artifact is generated; high computational efficiency: the algorithm complexity is O(N), which is suitable for real-time processing; multi-purpose: can be used for joint filtering, image matting, image enhancement and the like.

[0161] Optionally, Figure 10 g(n) in the Gaussian pyramid is a down-sampled low-pass image, d(n) is an up-sampled detail image in the Laplacian pyramid, d'(n) is a first enhanced image after detail enhancement, e(n) is a first enhanced image after input synthesis by up-sampling compensation, and e'(n) is a filtered image after guided filtering of each resolution layer.

[0162] Figure 11 It is an image denoising method. In the implementation Figure 10 of the image denoising method, a guided image construction method is proposed, which is based on the local variation degree of the original detail image at the current scale. The guided image is a kind of local standard deviation distribution graph (from a statistical point of view, this method is not a true standard deviation, and the deviation relative to a global average value is not calculated, but the standard deviation of the blurred image of the previous layer is calculated). The guided image g(n) is obtained by calculating the root mean square (RMS) of the detail image d(n) (also the Laplacian pyramid image described above), and the calculation method is as shown in Figure 12

[0163] The square of d(n) is calculated pixel by pixel, and then the average value of the window is calculated pixel by pixel with a specific window size (such as 3x3), and the square root of the average result is obtained to obtain the guided image g(n). The 3x3 convolution kernel in the convolution process is as shown in Figure 13 , which includes a plurality of 1 / 9 average kernels.

[0164] The calculation method is similar to the calculation of the local contrast noise ratio (CNR) of the detail image d(n), which quantifies the relative relationship between the useful signal (contrast) and the background noise in d(n), and d(n) reflects the high-frequency detail information. Using the method to generate g(n) as a guide image can not only significantly reduce noise, but also further preserve high-frequency details. At the same time, it has the advantage of small amount of calculation.

[0165] Another construction method of the guided image is to use an edge detection based local contrast definition to construct g(n) in the Gaussian pyramid. The construction process of this method is as follows: ​

[0166] For any pixel I(i,j) in g(n), calculate the Laplacian operator Δ(i,j), Δ(i,j)=|I(I,j)-IB|, where IB is the background average value of a 3×3 pixel local window centered at I(i,j). When (i,j)≠0, the average edge value E(i,j) is defined as:

[0167] ;

[0168] in, The average marginal value, For the Laplace operator, Let i be the number of rows of pixels and j be the number of columns of pixels.

[0169] The formula for the local contrast at pixel I(i,j) is:

[0170] ;

[0171] in, For local contrast, The average marginal value, Let i be the number of pixels, i be the number of rows of pixels, and j be the number of columns of pixels.

[0172] Optionally, adjustments can be made only to the filtered object, such as... Figure 14 As shown, the enhanced detail d'(x) is directly filtered using the guide image g(x) to obtain d''(x), and then the composite image e(x) at that scale is generated. This method directly filters the enhanced high-frequency detail information and uses stretching algorithms that are more drastic than some detail enhancements, such as exponential stretching. These algorithms are typically used in scenarios where X-ray penetration is insufficient, such as X-ray flaw detection of thick metal parts.

[0173] Understandably, filtering at different scales can both enhance details through nonlinear stretching at each scale and suppress increased noise in a timely manner. Alternatively, different filtering parameters can be adjusted for different scales to adapt to the noise at the current scale. The guide image reflects both the noise level of the original detail image and the local contrast level. Filtering the synthesized enhanced image at this scale under the guidance of this guide image can achieve a good balance between enhancing details, maintaining contrast, and reducing noise.

[0174] Alternatively, different guide maps, different filtering methods, different filtering positions at different scales, or even no filtering at certain scales can all improve image noise.

[0175] like Figure 7As shown in the embodiments of the present application, the image denoising device 900 further comprises: an image determining module 902 configured to determine an X-ray original image; a Gaussian processing module 904 configured to perform layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; an interpolation residual module 906 configured to determine a second image set according to the inter-layer interpolation residual of the plurality of first image sets; a detail enhancement module 908 configured to perform nonlinear detail enhancement on a detail image of each resolution layer in the second image set to determine a first enhanced image; a guide constructing module 910 configured to construct a guide image based on the local contrast-to-noise ratio of the detail image of the current level; a guide filtering module 912 configured to perform guide filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer; and a denoising enhancement module 914 configured to perform layer-by-layer up-sampling processing on the plurality of filtered images to determine a denoising enhanced image corresponding to the X-ray original image.

[0176] As shown in the embodiments of the present application, the image denoising device 900 further comprises: an image determining module 902 configured to determine an X-ray original image; a Gaussian processing module 904 configured to perform layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; an interpolation residual module 906 configured to determine a second image set according to the inter-layer interpolation residual of the plurality of first image sets; a detail enhancement module 908 configured to perform nonlinear detail enhancement on a detail image of each resolution layer in the second image set to determine a first enhanced image; a guide constructing module 910 configured to construct a guide image based on the local contrast-to-noise ratio of the detail image of the current level; a guide filtering module 912 configured to perform guide filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer; and a denoising enhancement module 914 configured to perform layer-by-layer up-sampling processing on the plurality of filtered images to determine a denoising enhanced image corresponding to the X-ray original image. Figure 8 As shown in the embodiments of the present application, the image denoising device 900 further comprises: an image determining module 902 configured to determine an X-ray original image; a Gaussian processing module 904 configured to perform layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; an interpolation residual module 906 configured to determine a second image set according to the inter-layer interpolation residual of the plurality of first image sets; a detail enhancement module 908 configured to perform nonlinear detail enhancement on a detail image of each resolution layer in the second image set to determine a first enhanced image; a guide constructing module 910 configured to construct a guide image based on the local contrast-to-noise ratio of the detail image of the current level; a guide filtering module 912 configured to perform guide filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer; and a denoising enhancement module 914 configured to perform layer-by-layer up-sampling processing on the plurality of filtered images to determine a denoising enhanced image corresponding to the X-ray original image.

[0177] Optionally, the processor 1110 is configured to determine an X-ray original image.

[0178] Optionally, the processor 1110 is further configured to perform layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers.

[0179] Optionally, the processor 1110 is further configured to determine a second image set according to the inter-layer interpolation residual of the plurality of first image sets.

[0180] Optionally, the processor 1110 is further configured to perform nonlinear detail enhancement on a detail image of each resolution layer in the second image set to determine a first enhanced image.

[0181] Optionally, the processor 1110 is further configured to construct a guide image based on the local contrast-to-noise ratio of the detail image of the current level.

[0182] Optionally, the processor 1110 is further configured to perform guide filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer.

[0183] Optionally, the processor 1110 is further configured to perform layer-by-layer up-sampling processing on the plurality of filtered images to determine a denoised enhanced image corresponding to the X-ray original image.

[0184] The memory 1109 can be configured to store software programs and various data. The memory 1109 can mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area can store an operating system, application programs or instructions required by at least one function (such as a sound playing function, an image playing function, etc.), and the like. In addition, the memory 1109 can include a volatile memory or a non-volatile memory, or the memory 1109 can include both a volatile memory and a non-volatile memory. The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory. The volatile memory can be a Random Access Memory (RAM), a Static RAM (SRAM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a Double Data Rate SDRAM (DDR SDRAM), an Enhanced SDRAM (ESDRAM), a Synch link DRAM (SLDRAM), and a Direct Rambus RAM (DRRAM). The memory 1109 in the embodiments of the present application includes but is not limited to these and any other suitable types of memory.

[0185] The embodiments of the present application further provide a readable storage medium, and the readable storage medium stores programs or instructions, the programs or instructions are executed by a processor to realize each process of the above-mentioned image denoising method embodiments and achieve the same technical effects. To avoid repetition, details are not described herein. In addition, the readable storage medium improves the data storage capacity and data processing speed of the image denoising method in the present application.

[0186] The readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media, or electrical signals transmitted through a wire.

[0187] The processor is a processor in the electronic device in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and the like.

[0188] The embodiment of the present application further provides a chip, which includes a processor and a communication interface. The communication interface is coupled with the processor. The processor is used to run programs or instructions, to realize each process of the above image denoising method embodiment, and to achieve the same technical effects. To avoid repetition, details are not described herein. In addition, the chip improves the data processing speed of the method in the present application.

[0189] It should be understood that the chip mentioned in the embodiment of the present application can also be referred to as a system-level chip, a system chip, a chip system, or a system-on-chip, etc.

[0190] In the present application, the terms "first", "second", "third" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance; the term "multiple" refers to two or more, unless otherwise explicitly limited. The terms "mounting", "connecting", "connecting", "fixing" and the like should be understood in a broad sense, for example, "connecting" can be fixed connection, or detachable connection, or integrally connected; "connected" can be directly connected, or indirectly connected through an intermediate medium. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0191] In the description of the application, it should be understood that the terms "upper", "lower", "left", "right", "front", "rear", and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the purpose of facilitating the description of the application and simplifying the description, and do not indicate or imply that the device or unit referred to must have a particular direction, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application.

[0192] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "a specific embodiment" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0193] The above is only the preferred embodiment of the present application, and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An image denoising method, characterized in that, The method comprises: determining an X-ray original image; performing layer-by-layer low-pass filtering and down-sampling processing on the X-ray original image to generate a first image set comprising a plurality of resolution layers; determining a second image set according to the inter-layer interpolation residual of the first image set; performing non-linear detail enhancement on the detail image of each resolution layer in the second image set to determine a first enhanced image; constructing a guide image based on the local contrast-to-noise ratio of the detail image of the current level; performing guided filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer; performing layer-by-layer up-sampling processing on the filtered images to determine a denoising enhanced image corresponding to the X-ray original image; The method comprises: determining the number of levels K of the plurality of resolution layers, wherein K is a positive integer; determining a low-pass image of each resolution layer in the first image set; performing bilinear interpolation up-sampling on the low-pass image of the Kth layer to restore the resolution of the low-pass image of the Kth layer to the resolution corresponding to the low-pass image of the K-1th layer; subtracting the up-sampled image from the low-pass image of the K-1th layer pixel by pixel to generate the detail image of the Kth layer; determining a second image set according to the detail images of the plurality of resolution layers.

2. The image denoising method of claim 1, wherein, The method comprises: determining a secondary low-pass image after up-sampling the low-pass image of the Kth layer; using the secondary low-pass image as a compensation input of the first enhanced image of the Kth layer, and determining a filtered image of the Kth layer with the guide image of the Kth layer as a constraint; starting from the K-1th layer, performing bilinear interpolation up-sampling on the filtered image as a compensation input of the first enhanced image corresponding to each resolution layer to determine the filtered images corresponding to the plurality of resolution layers, and performing an iteration operation; iterating until the resolution layer corresponding to the X-ray original image, and outputting a denoising enhanced image.

3. The image denoising method of claim 1, wherein, The method further comprises: performing filtering processing on the first enhanced image according to the guide image to determine a secondary filtered image; determining the secondary filtered image corresponding to each resolution layer; determining a secondary low-pass image after up-sampling the low-pass image of the Kth layer; determining a filtered image of the Kth layer according to the secondary low-pass image and the secondary filtered image; starting from the K-1th layer, performing bilinear interpolation up-sampling on the filtered image, determining a filtered image corresponding to each resolution layer according to the bilinear interpolation up-sampled filtered image and the secondary filtered image corresponding to each resolution layer, and performing an iteration operation; iterating until the resolution layer corresponding to the X-ray original image, and outputting a denoising enhanced image.

4. The image denoising method of claim 1, wherein, The method comprises: calculating a square value pixel by pixel for a raw detail image of a current level; determining at least one size parameter; determining a sliding window according to the size parameter; determining an average value of the square value in the sliding window; determining a guide image reflecting a local contrast-to-noise ratio by square root operation on the average value.

5. The image denoising method of any of claims 1 to 4, characterized in that, The method of constructing a guide image further comprises: determining a down-sampled image in the first image set; determining a plurality of pixel points in the down-sampled image; determining a background average value of a local window centered at the pixel point; determining an average edge value according to the background average value and the pixel point; determining a local contrast corresponding to the pixel point according to the average edge value; determining a guide image according to the local contrast; wherein the formula of the average edge value is: ； wherein, is the average edge value, is the Laplacian operator, is the pixel point, i is the pixel point row number, and j is the pixel point column number. and the formula of the local contrast is: ； wherein, is a local contrast, is an average edge value, is a pixel point, i is a pixel point row number, and j is a pixel point column number.

6. An image noise reduction apparatus, characterized by comprising: comprises: an image determining module configured to determine an X-ray raw image; a Gaussian processing module configured to perform layer-by-layer low-pass filtering and down-sampling processing on the X-ray raw image to generate a first image set comprising a plurality of resolution layers; an interpolation residual module configured to determine a second image set according to inter-layer interpolation residuals of the first image set; a detail enhancement module configured to perform non-linear detail enhancement on a detail image of each resolution layer in the second image set to determine a first enhanced image; a guide constructing module configured to construct a guide image based on a local contrast-to-noise ratio of the detail image of a current level; a guide filtering module configured to perform guide filtering on the first enhanced image with the guide image as a constraint to determine a filtered image corresponding to each resolution layer; a noise reduction enhancement module configured to perform layer-by-layer up-sampling processing on a plurality of filtered images to determine a noise reduction enhanced image corresponding to the X-ray raw image. The method of determining a second image set according to inter-layer interpolation residuals of a plurality of first image sets comprises: determining a level number K of a plurality of resolution layers, wherein K is a positive integer; determining a low-pass image of each resolution layer in the first image set; performing bilinear interpolation up-sampling on the low-pass image of the Kth layer to restore the resolution of the low-pass image of the Kth layer to a resolution corresponding to the low-pass image of the K-1th layer; subtracting the up-sampled image from the low-pass image of the K-1th layer pixel by pixel to generate the detail image of the Kth layer; determining a second image set according to the detail images of a plurality of resolution layers.

7. An electronic device, comprising: A computer readable storage medium stores a program or instructions, which, when executed by a processor, implement the steps of the image denoising method according to any one of claims 1 to 5.

8. A readable storage medium, characterized by, A chip comprises a processor and a communication interface, the communication interface and the processor are coupled, the processor is configured to run a program or instructions, and implement the steps of the image denoising method according to any one of claims 1 to 5.

9. A chip, characterized by ​

Images