SEM image noise reduction method

Abstract

The embodiment of the invention provides an SEM image noise reduction method. The method comprises the following steps: acquiring an SEM image to be denoised as an original image; constructing a noise reduction evaluation function to enable the noise reduction evaluation function to comprise a first item and a second item, and performing noise reduction processing on the original image by using the noise reduction evaluation function to obtain a corresponding noise reduction image; wherein the first item is used for reducing the pixel value fluctuation of the noise reduction image, and the second item is used for reducing the difference between the original image and the noise reduction image. Compared with the prior art, the method has the advantages that more image details in the original image are reserved on the basis of realizing image noise reduction, and the image quality of the noise-reduced image is improved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method for denoising SEM images. Background Technology

[0002] Unlike optical imaging, scanning electron microscopy (SEM) works by illuminating the sample surface with an electron beam and collecting signal electrons through a detector to generate an SEM image. Due to quantum fluctuations, the electron beam is randomly distributed on the sample surface, much like raindrops falling on water in a random pattern. This randomness in the signal electrons results in random noise in the acquired SEM image, known as quantum noise. Quantum noise differs from salt-and-pepper noise or Gaussian noise in optical images.

[0003] Traditional image denoising methods are widely used to process noise types such as Gaussian noise or salt-and-pepper noise. However, when processing quantum noise in SEM images, these methods result in the loss of many image details in the denoised SEM image, which in turn leads to poor image quality. Summary of the Invention

[0004] Based on this, this application provides a SEM image denoising method that can retain more image details while achieving SEM image denoising, thereby improving the image quality of the denoised image.

[0005] This application provides a method for SEM image noise reduction, including:

[0006] Obtain the SEM image to be denoised as the original image;

[0007] A denoising evaluation function is constructed such that the denoising evaluation function includes a first term and a second term, and the original image is denoised using the denoising evaluation function to obtain the corresponding denoised image;

[0008] The first item is used to reduce the pixel value fluctuation of the denoised image, and the second item is used to reduce the difference between the original image and the denoised image.

[0009] In the technical solution provided in this application embodiment, a denoising evaluation function is constructed for the original image and the denoised image. The first term of the denoising evaluation function is used to reduce the pixel value fluctuation of the denoised image, and the second term is used to reduce the difference between the original image and the denoised image. By performing denoising processing on the original image through the constructed denoising evaluation function, the pixel value fluctuation of the obtained denoised image is smaller, thus achieving a denoising effect. Moreover, the difference between the denoised image and the original image is smaller. Compared with the prior art, more image details in the original image are preserved while achieving image denoising, which can improve the image quality of the denoised image. Attached Figure Description

[0010] Figure 1 This is a schematic flowchart of a SEM image noise reduction method provided in an embodiment of this application;

[0011] Figure 2 This is a schematic flowchart illustrating the noise reduction process for an original image provided in an embodiment of this application.

[0012] Figure 3 A schematic diagram illustrating a region division method for the original image provided in an embodiment of this application;

[0013] Figure 4 Another schematic diagram illustrating the region division method of the original image provided in the embodiments of this application;

[0014] Figure 5 Another schematic diagram illustrating the region division method of the original image provided in the embodiments of this application;

[0015] Figure 6 Another schematic diagram illustrating the region division method of the original image provided in the embodiments of this application;

[0016] Figure 7 This is another schematic diagram illustrating the region division method of the original image provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit this application. Furthermore, it should be noted that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, not the entire structure.

[0018] The SEM image denoising method provided in this application embodiment can also be called an image denoising method for SEM images. It is applicable to scenarios where SEM images are denoised. By constructing a denoising evaluation function suitable for SEM images, and using this denoising evaluation function to reduce pixel value fluctuations in the denoised image (i.e., the denoised SEM image) and the difference between the denoised image and the original image, denoising of the SEM image is achieved. The execution subject of this application embodiment can be an SEM image denoising device, which can be implemented by hardware and / or software, and is generally integrated into an electronic device. Optionally, the electronic device can be a terminal with data processing capabilities and the ability to interact with other devices, such as a laptop, tablet, or personal digital assistant. It can also be an independent server or a server cluster composed of multiple servers. For example, the electronic device can be the computer included in a scanning electron microscope. This application embodiment does not limit the specific form of the electronic device.

[0019] Figure 1 This is a schematic flowchart of a SEM image denoising method provided in an embodiment of this application. Figure 1 As shown, the method may include:

[0020] S101. Obtain the SEM image to be denoised as the original image.

[0021] In one embodiment, a SEM image of a sample (such as a wafer) to be denoised is acquired by a scanning electron microscope. The SEM image to be denoised can be referred to as the original image, and the original image is transmitted to an electronic device for image denoising processing.

[0022] In one embodiment, to ensure calculation accuracy, the pixel values ​​of the original image can be converted into floating-point numbers, and subsequent steps can be performed based on the floating-point numbers corresponding to the original image.

[0023] S102. Construct a denoising evaluation function that includes a first term and a second term, and use the denoising evaluation function to denoise the original image to obtain the corresponding denoised image; the first term is used to reduce the pixel value fluctuation of the denoised image, and the second term is used to reduce the difference between the original image and the denoised image.

[0024] The inventors discovered that quantum noise in SEM images typically manifests as sharp fluctuations in pixel values. Therefore, denoising of the original image can be achieved by reducing these pixel value fluctuations. Specifically, a denoising evaluation function is constructed between the original image and the denoised image (i.e., the denoised SEM image). The first term of the denoising evaluation function reduces pixel value fluctuations in the denoised image, thereby reducing quantum noise in the SEM image. The second term reduces the difference between the original image and the denoised image, preserving more image details from the original image. Optionally, the denoising evaluation function can be the sum of the first and second terms.

[0025] In this way, the original image is denoised using the constructed denoising evaluation function. Since the first term of the denoising evaluation function is used to reduce the fluctuation of pixel values ​​in the denoised image, the pixel value fluctuation of the denoised image will be smaller after processing, thus achieving image denoising. Furthermore, since the second term of the denoising evaluation function is used to reduce the difference between the original image and the denoised image, the denoised image will be as close as possible to the original image after processing, thus preserving more image details in the original image.

[0026] The SEM image denoising method provided in this application constructs a denoising evaluation function for the original image and the denoised image. The first term of the denoising evaluation function is used to reduce the pixel value fluctuation of the denoised image, and the second term is used to reduce the difference between the original image and the denoised image. By performing denoising processing on the original image through the constructed denoising evaluation function, the pixel value fluctuation of the resulting denoised image is smaller, thus achieving a denoising effect. Moreover, the difference between the denoised image and the original image is smaller. Compared with the prior art, it retains more image details in the original image while achieving image denoising, thereby improving the image quality of the denoised image.

[0027] Optionally, the expression for the first term of the denoising evaluation function can be a double integral of the horizontal and vertical gradients of the sub-region of the denoised image, and the expression for the second term is the product of the integral of the pixel difference between the sub-region of the denoised image and the sub-region of the original image and the regularization parameter.

[0028] In this context, the sub-region of the denoised image can be any one of the multiple regions contained in the denoised image, or it can refer to the denoised image itself. Similarly, the sub-region of the original image can also be any one of the multiple regions contained in the original image, or it can refer to the original image itself. In the embodiments of this application, the term "sub-region of the image" can be referred to as the "true sub-region of the image" when the image itself is excluded. The difference between the terms "sub-region of the image" and "true sub-region of the image" is similar to the difference between "subset" and "true subset" as known to those skilled in the art.

[0029] In this model, a sub-region of the denoised image and a sub-region of the original image have the same position and size, based on which the pixel difference between the two can be obtained. Furthermore, the regularization parameter value corresponds to the sub-region and is specifically a value obtained within a preset region.

[0030] In one embodiment, a denoised image of the original image is obtained based on the denoising results corresponding to each sub-region of the original image. Optionally, the expression for the denoising evaluation function E(u) can be as follows:

[0031]

[0032] Where u is a sub-region of the denoised image, u0 is a sub-region of the original image, and u0 corresponds to u, that is, u is the result of denoising u0, λ is the regularization parameter, and x and y are the horizontal and vertical orthogonal directions in the denoised image / original image, respectively. In this embodiment, "sub-region" refers to "rectangular sub-region".

[0033] In this embodiment, "horizontal" and "vertical" refer to the directions of two adjacent rectangular sides of the denoised image (or the original image). The two adjacent rectangular sides of each sub-region are parallel to the two adjacent rectangular sides of the original image. That is, for any given sub-region, it has two rectangular sides in the horizontal direction and two rectangular sides in the vertical direction. The aforementioned regularization parameter is used to balance the trade-off between image denoising effect and image detail preservation. A larger value for the regularization parameter results in smaller pixel value fluctuations in the denoised image and a better image denoising effect. However, compared to the original image, the loss of image detail in the denoised image will be greater.

[0034] For example, those skilled in the art can set a smaller value for the regularization parameter to complete the noise reduction with fewer iterations, so as to retain more image details of the original image. In this case, the image noise reduction effect is less important than the preservation of image details, and the preservation of image details is better.

[0035] For example, those skilled in the art can also set a larger value for the regularization parameter to complete denoising with a larger number of iterations, thus preserving less image detail from the original image. In this case, the image denoising effect is more important than the preservation of image detail, resulting in a better image denoising effect. It should be noted that even if the regularization parameter value is large, resulting in less image detail remaining in the denoised result, compared with the prior art, because the denoising evaluation function includes the first and second terms, the method provided in this embodiment still has the technical effect of preserving more image detail from the original image while achieving image denoising.

[0036] In one embodiment, alternatively, such as Figure 2As shown, in step S102 above, the original image is denoised using a denoising evaluation function to obtain the corresponding denoised image, which may include:

[0037] S201. Divide the original image into regions to obtain one or more regions.

[0038] Specifically, the original image can be divided into regions based on its content, resulting in one or more regions. Noise reduction processing can then be applied to each region individually. In some cases, the original image itself can be used as a single region, and noise reduction can be performed directly on the original image.

[0039] In this embodiment, the original image is divided horizontally and vertically to obtain rectangular regions. For example, each of the above-mentioned regions is a rectangular region, and the two adjacent rectangular sides of each region are parallel to the two adjacent rectangular sides of the original image. When the original image is divided into multiple regions, since the original image is rectangular, the regions are rectangular. Dividing the original image into multiple regions easily yields multiple rectangular regions. As described above, when the original image is divided into multiple regions, each region is a true sub-region.

[0040] As an optional implementation, S201 may include: obtaining the total number of pixels in the region of interest in the original image; dividing the original image into regions to obtain one or more divided regions, such that the one or more divided regions include a first divided region, and such that the ratio of the number of pixels in the region of interest in the first divided region to the total number of pixels is greater than or equal to a first preset threshold.

[0041] Specifically, the region of interest (ROI) is extracted from the original image, and the total number of pixels in the ROI is obtained. The original image is then divided according to the ratio of the number of pixels in the ROI within the divided region to the total number of pixels in the ROI. This results in a first divided region in the divided original image. The ratio in the first divided region is greater than or equal to a first preset threshold, which means that a first divided region is obtained that can accommodate more pixels in the ROI. In other words, the ROI is divided into the first divided region as much as possible, so that the corresponding regularization parameter value can be set for the first divided region.

[0042] The number of regions of interest can be one or more, and the number of first partitioned regions can be one or more.

[0043] The first preset threshold can be set based on actual needs. Optionally, the value range of the first preset threshold can be [70%, 100%], such as 70%, 80%, 90%, or 100%. When the first preset threshold is 100%, the ratio of the number of pixels in the region of interest within the first segmented region to the total number of pixels is 1. The first segmented region is the original image, and it includes the complete region of interest. Those skilled in the art can reasonably select the first preset threshold within this value range to divide the original image into one or more segmented regions. For example, the original image can be divided into one segmented region, i.e., the first segmented region, or the original image can be divided into multiple segmented regions, with the first segmented region included in the multiple segmented regions.

[0044] In some examples, the number of regions of interest (ROIs) is greater than the number of first segmented regions. For example, there may be multiple ROIs and one first segmented region. The original image is segmented according to the ratio of the number of pixels in each ROI within the segmented region to the total number of pixels in all ROIs (i.e., the sum of the total number of pixels in each ROI), such that there exists a first segmented region in the segmented original image, and the ratio within this first segmented region is greater than or equal to a first preset threshold.

[0045] In some examples, the number of regions of interest (ROIs) is the same as the number of first partitioned regions. This facilitates obtaining a corresponding first partitioned region for each ROI, ensuring a one-to-one correspondence between the two. For example, ... Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, the number of regions of interest is one, and the number of the first partitioned regions is also one. Furthermore, in Figures 3-6 In the diagram, the area of ​​interest is indicated by a black area.

[0046] In some examples, such as Figure 3 and Figure 4 As shown, by dividing the original image into regions, a first segmented region and other segmented regions can be obtained. These other segmented regions include a second segmented region, and they are all regions of the same size obtained by equally dividing the original image. Specifically, the ratio of the number of pixels in the region of interest (ROI) within the first segmented region to the total number of pixels in the ROI is greater than or equal to a first preset threshold, and this ratio is 1. The second segmented region contains no pixels from the ROI. Figure 3 The diagram illustrates a first subdivision region and three second subdivision regions; in Figure 4 The diagram illustrates a first division region and two second division regions.

[0047] In some examples, such as Figure 5As shown, by dividing the original image into regions, a first division region and other division regions can be obtained. These other division regions include a second division region and a third division region, and these division regions are regions of the same size obtained by equally dividing the original image. Specifically, the ratio of the number of pixels in the region of interest (ROI) within the first division region to the total number of pixels in the ROI is greater than or equal to a first preset threshold (i.e., the first division region contains most of the ROI pixels), for example, if the first preset threshold is 70%, this ratio is 75%. The second division region contains no ROI pixels, and the ratio of the number of ROI pixels within the third division region to the total number of ROI pixels is less than the first preset threshold (i.e., the third division region contains a small number of ROI pixels). Figure 5 The diagram illustrates one first subdivision region and six second subdivision regions. Figure 5 (Only one second division region is marked in the text) and two third division regions.

[0048] In some examples, such as Figure 6 As shown, by dividing the original image into regions, a first segmented region and other segmented regions can be obtained. These other segmented regions include a second segmented region and a third segmented region, and these segmented regions have different sizes. Figure 6 The diagram illustrates a first subdivision region and two second subdivision regions. Figure 6 (Only one second division region is marked in the text) and two third division regions.

[0049] In one embodiment, obtaining the total number of pixels in the region of interest (ROI) of the original image includes: obtaining at least one ROI in the original image, and obtaining the total number of pixels in each ROI. In another embodiment, ensuring that the one or more partitioned regions include a first partitioned region, and that the ratio of the number of pixels in the ROI within the first partitioned region to the total number of pixels is greater than or equal to a first preset threshold, includes: ensuring that the one or more partitioned regions include the same number of first partitioned regions as the ROIs, and that the ratio of the number of pixels in each ROI within its corresponding first partitioned region to the total number of pixels (corresponding to the ROI) is greater than or equal to the first preset threshold. Therefore, there is a one-to-one correspondence between ROIs and first partitioned regions. After region partitioning, each ROI can obtain a first partitioned region capable of accommodating a relatively large number of pixels within that ROI.

[0050] In some examples, the number of regions of interest is the same as the number of the first partitioned regions, and both are multiple, such as... Figure 7 As shown, for example, there are two of each. Figure 7The two regions of interest are indicated by black areas; one region of interest corresponds to one of the first subdivision regions, and the other region of interest corresponds to another of the first subdivision regions. Figure 7 The diagram illustrates two first-level regions and three second-level regions. Figure 7 (Only one second division region and one third division region are marked in the text.)

[0051] In one embodiment, taking a first preset threshold of 70% as an example, the following further details are provided. Figure 7 The first partitioned regions corresponding to the two regions of interest. Specifically, for Figure 7 The region of interest (ROI) in the upper left corner is defined as follows: the number of pixels within the corresponding first partitioned region is equal to the total number of pixels in that ROI. This ratio is 1, or 100%. Since this ratio is greater than a first preset threshold, the first partitioned region can accommodate a significant number of pixels within the ROI; specifically, the first partitioned region contains all the pixels within the ROI. Figure 7 Another region of interest in the first division area has a larger ratio of the number of pixels in the corresponding first division area to the total number of pixels in the region of interest, for example, 80%. This ratio is greater than a first preset threshold, and the first division area can accommodate more pixels in the region of interest.

[0052] The aforementioned region of interest (ROI) can be the area of ​​interest that the user is interested in within the original image. The electronic device extracts the ROI based on the area selected by the user in the original image and obtains the total number of pixels within the ROI. For example, the ROI can correspond to a defect on a sample (such as a wafer), in which case the ROI is the defect area in the original image; the ROI can also correspond to an object to be measured or observed on a sample (such as a wafer), in which case the ROI is the area to be measured or observed in the original image.

[0053] S202. Obtain the parameter values ​​of the regularization parameters corresponding to each division region within the preset interval, assign the sub-regions in the denoising evaluation function as division regions respectively, and perform denoising processing on the division regions based on the denoising evaluation function to obtain the denoising results corresponding to the division regions.

[0054] The above-mentioned preset interval is the range of values ​​for the regularization parameter. Optionally, the above-mentioned preset interval can be [0.01, 1].

[0055] In this embodiment, for each segmented region, the parameter value of the regularization parameter corresponding to that segmented region is obtained within a preset interval. Optionally, the parameter value of the regularization parameter corresponding to the segmented region can be obtained from the preset interval based on the ratio of the number of pixels in the region of interest contained within the segmented region to the total number of pixels in the region of interest. For example, the parameter value of the regularization parameter corresponding to the segmented region is negatively correlated with the ratio of the number of pixels in the region of interest contained within the segmented region to the total number of pixels in the region of interest. That is, if the ratio of the number of pixels in the region of interest contained within the segmented region to the total number of pixels in the region of interest is large, the parameter value of the regularization parameter can be set smaller to retain more image details in the original image. Conversely, if the ratio of the number of pixels in the region of interest contained within the segmented region to the total number of pixels in the region of interest is small, the parameter value of the regularization parameter can be set larger to achieve a better noise reduction effect.

[0056] As an optional implementation, when the original image is divided into multiple regions, the process of obtaining the parameter value of the regularization parameter corresponding to each region within a preset interval may include:

[0057] Within a preset interval, obtain the parameter values ​​of the regularization parameters corresponding to the first partition region and the regularization parameters corresponding to the other partition regions, wherein the parameter value of the regularization parameter corresponding to the first partition region is less than the parameter value of the regularization parameter corresponding to the other partition regions.

[0058] Specifically, for the first segmented region, since the number of pixels in the region of interest is relatively large, a smaller regularization parameter value can be set for the first segmented region to retain more image details in the original image while achieving noise reduction. The other segmented regions contain fewer pixels in the region of interest, so a larger regularization parameter value can be set to achieve better noise reduction.

[0059] In this embodiment, there is one first partition region and one or more remaining partition regions. In one implementation, when there are multiple remaining partition regions, the regularization parameter values ​​corresponding to each remaining partition region can be the same or different, but are not limited to this. In one embodiment, the remaining partition regions include second and third partition regions. The number of second partition regions is at least one, i.e., one or more, and the number of third partition regions is also at least one, i.e., one or more. The regularization parameter value λ2 corresponding to each second partition region is the same, and the regularization parameter value λ3 corresponding to each third partition region is the same.

[0060] For example, see [link to previous article] Figure 3 and Figure 4The original image includes a first segmented region and a second segmented region. The regularization parameter values ​​corresponding to each segmented region are related as follows: the regularization parameter value λ1 corresponding to the first segmented region < the regularization parameter value λ2 corresponding to the second segmented region. (See also...) Figure 5 The original image includes a first segmented region, a second segmented region, and a third segmented region. The regularization parameters corresponding to each segmented region are related as follows: the regularization parameter λ1 corresponding to the first segmented region < the regularization parameter λ3 corresponding to the third segmented region < the regularization parameter λ2 corresponding to the second segmented region.

[0061] Furthermore, based on the obtained parameter values, the regularization parameter in the denoising evaluation function is assigned a value, and the sub-regions of the original image and the sub-regions of the denoised image in the denoising evaluation function are both assigned to the corresponding division regions in the original image (that is, the sub-regions of the original image and the sub-regions of the denoised image are initialized to the corresponding division regions in the original image). The denoising evaluation function corresponding to the division region is then used to perform denoising processing on the division region to obtain the denoising result corresponding to the division region.

[0062] Optionally, the process of performing denoising on the divided regions based on the denoising evaluation function to obtain the denoising results corresponding to the divided regions may include:

[0063] The sub-regions of the denoised image in the denoising evaluation function are iteratively optimized using an optimization algorithm. The sub-regions of the denoised image when the value of the denoising evaluation function meets the preset conditions are taken as the denoising results corresponding to the divided regions.

[0064] The optimization algorithm may include gradient descent, least squares, or other iterative optimization algorithms, and this embodiment does not limit it.

[0065] In one embodiment, an initial value of the denoising evaluation function is calculated. An optimization algorithm is used to iterate over sub-regions of the denoised image. After each iteration, the value of the denoising evaluation function is calculated, and it is determined whether the value of the denoising evaluation function meets a preset condition. If yes, the iteration stops, and the sub-region of the denoised image obtained in this iteration is taken as the denoising result corresponding to the divided region. If not, the next iteration continues until the value of the denoising evaluation function meets the preset condition. The sub-region of the denoised image when the value of the denoising evaluation function meets the preset condition is taken as the denoising result corresponding to the divided region. This process is repeated. If there are multiple divided regions, denoising processing is performed on each divided region based on the denoising evaluation function to obtain a denoising result corresponding to each divided region, resulting in multiple denoising results.

[0066] Optionally, the preset condition can be that the absolute value of the change between the value of the noise reduction evaluation function before and after the iteration is less than or equal to a second preset threshold. The second preset threshold is set based on actual needs, and optionally, the range of the second preset threshold can be

[10] . -6 10 -4 ].

[0067] Specifically, the initial value of the denoising evaluation function is calculated, and the sub-region of the denoised image is iterated using an optimization algorithm. After each iteration, the current value of the denoising evaluation function is calculated, and the absolute value of the difference between the current value and the value of the denoising evaluation function before the iteration (which is the initial value) is calculated. If the absolute value is less than or equal to a second preset threshold, the iteration stops, and the sub-region of the denoised image obtained in this iteration is taken as the denoising result corresponding to the divided region. If the absolute value is greater than the second preset threshold, the next iteration is continued until the absolute value of the change in the value of the denoising evaluation function before and after the iteration is less than or equal to the second preset threshold. The sub-region of the denoised image after the iteration is taken as the denoising result corresponding to the divided region.

[0068] Optionally, the preset condition can be that the value of the noise reduction evaluation function is less than or equal to a third preset threshold. The third preset threshold is set based on actual needs, and optionally, the value range of the third preset threshold can be

[10] . -4 10 -2 For example, 10 -3 .

[0069] S203. Based on the denoising results corresponding to each divided region, obtain the denoised image of the original image.

[0070] Specifically, when only one region is obtained after dividing the original image, the original image is considered the divided region, also known as the first divided region, and the denoising result corresponding to this divided region is used as the denoised image of the original image. When multiple regions are obtained after dividing the original image, the denoising results corresponding to each divided region are combined, and the combined result is used as the denoised image of the original image. The denoised image of the original image can also be referred to as the denoised image corresponding to the original image.

[0071] As an optional implementation, when the original image is divided into regions to obtain a segmented region, the process of obtaining the parameter values ​​of the regularization parameters corresponding to each segmented region within a preset interval may include:

[0072] Obtain the regularization parameter value of the first segmented region, i.e., the original image, within the preset interval.

[0073] As an alternative implementation, for cases where the original image itself is considered as a segmented region, the original image can be denoised using the following process:

[0074] Within a preset range, the parameter values ​​of the regularization parameters corresponding to the original image are obtained. The sub-regions of the original image and the sub-regions of the denoised image in the denoising evaluation function are both assigned to the original image. The original image is then denoised based on the denoising evaluation function to obtain the denoised image of the original image.

[0075] If only one region is obtained by dividing the original image, then as described above, this region is the first region and also the original image. In this case, a regularization parameter value is obtained within a preset interval to serve as the regularization parameter value for the first region, i.e., the original image.

[0076] Specifically, the parameter value of the regularization parameter corresponding to the original image can be obtained within a preset interval based on the distribution information of the region of interest in the original image. For example, if the distribution range of the region of interest in the original image is large, a smaller parameter value of the regularization parameter can be obtained from the preset interval. That is, a smaller parameter value of the regularization parameter is set for the original image in order to retain as many image details in the original image as possible while achieving the noise reduction effect.

[0077] As an optional implementation, obtaining the regularization parameter value corresponding to the original image within the preset interval may include: obtaining the first total number of pixels in the region of interest of the original image and the second total number of pixels in the original image, calculating the ratio between the first total number of pixels and the second total number of pixels, and obtaining the regularization parameter value corresponding to the original image within the preset interval based on the ratio. For example, if the ratio is large, a smaller regularization parameter value can be obtained from the preset interval, that is, a smaller regularization parameter value is set for the original image to retain as many image details as possible while achieving noise reduction. If the ratio is small, a larger regularization parameter value can be obtained from the preset interval, that is, a larger regularization parameter value is set for the original image to improve the noise reduction effect.

[0078] Furthermore, the denoised image in the denoising evaluation function is initialized as the original image, and the denoised image in the denoising evaluation function is iteratively optimized using an optimization algorithm. The denoised image when the value of the denoising evaluation function meets the preset conditions is taken as the denoising result corresponding to the original image. The preset conditions, iterative optimization, and other related content have been introduced above and will not be repeated here.

[0079] In some embodiments of this application, when denoising the original image, the image content in the original image is fully considered. The original image is divided into multiple regions (each region can be called a sub-region), and denoising is performed on each region separately. During denoising, a suitable regularization parameter value is set for each region. Ultimately, while achieving the denoising effect for each region, more image details in the original image are preserved. For example, a smaller regularization parameter value is set for the region containing most pixels of the region of interest (ROI), i.e., the first region. This allows denoising of the region to be completed with fewer iterations, thus preserving more details of the ROI within that region.

[0080] Optionally, after obtaining the denoised image of the original image, there will still be a loss of detail compared to the original image. Therefore, detail enhancement can be performed on the denoised image. The SEM image denoising method also includes detail enhancement operations, which may include: performing a Fourier transform on the denoised image to obtain the corresponding image spectrum; performing a Fourier transform on the Gaussian blur kernel to obtain the spectrum of the Gaussian blur kernel; performing a conjugate complex number operation on the spectrum of the Gaussian blur kernel to obtain a filter; using the filter to filter the image spectrum to obtain the target spectrum; and performing an inverse Fourier transform on the target spectrum to obtain the denoised image with enhanced details.

[0081] In one implementation, to ensure computational accuracy, the above steps are performed based on floating-point numbers corresponding to the original image. Therefore, after obtaining the denoised image with enhanced details, the denoised image with enhanced details can also be converted from floating-point numbers to integer format.

[0082] In this embodiment, performing detail enhancement operations on the denoised image can further improve the image quality of the denoised image.

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0084] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0085] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for denoising SEM images, characterized in that, include: Obtain the SEM image to be denoised as the original image; A denoising evaluation function is constructed such that the denoising evaluation function includes a first term and a second term, and the original image is denoised using the denoising evaluation function to obtain the corresponding denoised image; The first item is used to reduce the pixel value fluctuation of the denoised image, and the second item is used to reduce the difference between the original image and the denoised image.

2. The method according to claim 1, characterized in that, The first term is expressed as the double integral of the horizontal and vertical gradients of the sub-region of the denoised image, and the second term is expressed as the product of the integral of the pixel difference between the sub-region of the denoised image and the sub-region of the original image and the regularization parameter.

3. The method according to claim 2, characterized in that, The original image is denoised using the denoising evaluation function to obtain a corresponding denoised image, including: The original image is divided into regions to obtain one or more regions; Within a preset interval, obtain the parameter values ​​of the regularization parameters corresponding to each of the partitioned regions, assign the sub-regions in the denoising evaluation function to the partitioned regions respectively, and perform denoising processing on the partitioned regions based on the denoising evaluation function to obtain the denoising results corresponding to the partitioned regions. Based on the noise reduction results corresponding to each of the defined regions, the noise-reduced image of the original image is obtained.

4. The method according to claim 3, characterized in that, The original image is divided into regions to obtain one or more regions, including: Obtain the total number of pixels in the region of interest in the original image; The original image is divided into regions to obtain one or more regions, such that the one or more regions include a first region, and the ratio of the number of pixels in the region of interest within the first region to the total number of pixels is greater than or equal to a first preset threshold.

5. The method according to claim 4, characterized in that, Obtaining the total number of pixels in the region of interest in the original image includes: obtaining at least one region of interest in the original image, and obtaining the total number of pixels in each region of interest; Ensuring that the one or more segmented regions include a first segmented region, and that the ratio of the number of pixels in the region of interest within the first segmented region to the total number of pixels is greater than or equal to a first preset threshold, includes: The number of first division regions is the same as the number of regions of interest in the one or more division regions, and the ratio of the number of pixels in each region of interest in the corresponding first division region to the total number of pixels is greater than or equal to a first preset threshold.

6. The method according to claim 3, characterized in that, When the original image is divided into multiple regions, the parameter values ​​of the regularization parameters corresponding to each region are obtained within a preset interval, including: Within the preset interval, obtain the parameter value of the regularization parameter corresponding to the first partition region and the parameter values ​​of the regularization parameters corresponding to the other partition regions, wherein the parameter value of the regularization parameter corresponding to the first partition region is less than the parameter values ​​of the regularization parameters corresponding to the other partition regions.

7. The method according to claim 3, characterized in that, The noise reduction process is performed on the segmented region based on the noise reduction evaluation function to obtain the noise reduction result corresponding to the segmented region, including: The sub-regions of the denoised image in the denoising evaluation function are iteratively optimized using an optimization algorithm. The sub-regions of the denoised image when the value of the denoising evaluation function meets a preset condition are taken as the denoising results corresponding to the divided regions.

8. The method according to claim 7, characterized in that, The preset condition is that the absolute value of the change between the value of the noise reduction evaluation function before iteration and the value after iteration is less than or equal to a second preset threshold; or the preset condition is that the value of the noise reduction evaluation function is less than or equal to a third preset threshold.

9. The method according to claim 4, characterized in that, Each of the defined regions is a rectangular region, and the two adjacent rectangular sides of each defined region are parallel to the two adjacent rectangular sides of the original image; the preset interval is [0.01, 1]; the value range of the first preset threshold is [70%, 100%].

10. The method according to claim 1, characterized in that, Also includes: Perform a Fourier transform on the denoised image to obtain the corresponding image spectrum; Perform a Fourier transform on the Gaussian blur kernel to obtain the spectrum of the Gaussian blur kernel; The filter is obtained by performing conjugate complex number operations on the spectrum of the Gaussian blur kernel; The image spectrum is filtered using the filter to obtain the target spectrum; Perform an inverse Fourier transform on the target spectrum to obtain the denoised image with enhanced details.

Images