Clear imaging method of mask under optical objective lens

Abstract

The invention discloses a method for clearly imaging a mask under an optical objective, and relates to the technical field of optical detection and image processing, and the method comprises the following steps: S1, constructing a light intensity distribution model based on the boundary condition of the field of view of the optical objective in combination with the reflection characteristic and incident angle change rule of a metal edge material, and determining the coverage range of metal edge reflection crosstalk, generating a mask boundary interference prediction map; s2, according to the mask boundary interference prediction map, performing region division on the micro-reflectivity image, extracting brightness gradient characteristics of reflectivity lifting in an interference region, and generating a brightness gradient parameter set required by image filtering; the method is based on light intensity modeling, fusion direction filtering, pixel stripping, gradient reconstruction and credibility weighted fusion, realizes closed-loop control from interference prediction to pixel restoration, has high resolution and adaptivity, can accurately strip edge reflection artifacts and restore a real image structure, remarkably reduces misjudgment and rework rate, and is suitable for large-scale popularization and application. And the mask yield and the stability of the detection system are improved.

Description

Technical Field

[0001] This invention relates to the field of optical detection and image processing technology, specifically to a method for clear imaging of a mask under an optical objective lens. Background Technology

[0002] "Clear imaging of glass blank masks under optical objectives" refers to the process in optical inspection systems where, when imaging a blank glass mask (i.e., a mask substrate without etched patterns) using a high-precision optical objective, specific optical control and image processing methods are employed to enhance image clarity, contrast, and edge detail reproduction. This allows for the accurate identification of minute defects, stress disturbances, or coating inhomogeneities on the mask's surface or subsurface. Since blank glass masks are typically high-smoothness, low-contrast, and structureless transparent or translucent materials, their imaging is susceptible to light scattering, refraction interference, and focal length deviation, leading to blurred images or insufficient contrast. Therefore, this method, through the coordinated optimization of optical path parameters (such as numerical aperture, depth of focus, and illumination angle), objective structure, and imaging algorithms, ensures high-resolution, clear imaging of blank glass masks under multi-angle, multi-polarization, or specific lighting conditions, providing a stable and reliable image foundation for subsequent inspection and quality assessment.

[0003] Micro-reflectometry mapping effectively enables clear imaging of blank glass masks under optical objectives. Its core function lies in using high-resolution objectives to precisely acquire and spatially map minute differences in reflectivity across various micro-regions of the mask surface. As a patternless, transparent substrate, a blank glass mask is difficult to identify micro-defects in traditional bright-field imaging due to insufficient contrast. Micro-reflectometry mapping, however, captures the highly sensitive response of surface reflectivity to local refractive index changes, film thickness fluctuations, submicron particle adhesion, and internal stress disturbances, enabling quantitative visualization of surface optical inhomogeneities. The imaging results highlight subtle anomalous areas in the form of grayscale or brightness contrast, clearly revealing previously imperceptible defects. This provides crucial detection evidence and imaging support for mask quality assessment and early defect identification.

[0004] The existing technology has the following shortcomings: In existing microreflectivity mapping imaging processes, the edge regions of blank glass masks are often coated with metal or frame materials due to process requirements or structural support. These metal boundaries have significantly higher reflectivity than the glass itself. When a high numerical aperture optical objective is used to scan the mask surface for microreflectivity, the strong reflection signal from the metal edge may be nonlinearly superimposed on the imaging signal of adjacent glass regions through lateral scattering, specular bounce, or multipath interference effects of the optical system, thus causing the so-called optical crosstalk problem. This crosstalk phenomenon can lead to false increases in reflectivity measurements or abnormal image brightness in adjacent defect-free areas, causing the system to mistakenly identify normal areas as defects such as local contamination, film abnormalities, or particle adhesion during automatic identification, seriously affecting the accuracy and reliability of the detection results. Especially in photolithography applications where high mask edge quality is required, if such false defects caused by structural boundary reflection interference cannot be identified and suppressed in time, the entire mask may be misjudged as a defective product, leading to unnecessary rework or resource waste.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a clear imaging method for photomasks under optical objectives. Based on light intensity modeling, it integrates directional filtering, pixel stripping, gradient reconstruction, and confidence-weighted fusion to achieve closed-loop control from interference prediction to pixel restoration. It has high resolution and adaptability, can accurately strip edge reflection artifacts, restore the true image structure, significantly reduce misjudgment and rework rates, improve photomask yield and detection system stability, thereby solving the problems in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for clear imaging of a mask under an optical objective lens, comprising the following steps: S1. Based on the field-of-view boundary conditions of the optical objective lens, combined with the reflection characteristics of the metal edge material and the change law of the incident angle, a light intensity distribution model is constructed to determine the coverage range of metal edge reflection crosstalk and generate a mask boundary interference prediction map. S2, based on the mask boundary interference prediction map, divide the micro-reflectivity image into regions, extract the brightness gradient features of the reflectivity increase in the interference region, and generate the set of brightness gradient parameters required for image filtering. S3, combined with the brightness gradient parameter set, introduces the direction-selective response filtering algorithm to construct an interference stripping operator, performs direction consistency judgment and abnormal response stripping on the interference area, and records the stripped pixel coordinates and the corresponding image response curve; S4. Based on the pixel coordinates and response curves of the stripped area, extract the brightness gradient features of the adjacent undisturbed area, construct a reliable grayscale expansion model, perform pixel-level interpolation reconstruction according to the grayscale trend, and generate grayscale reliability labels. S5, based on grayscale confidence labels, establishes a pixel-level weighted fusion mechanism to weightedly fuse the stable region of the original image with the reconstructed region, generating a composite reflectance image with continuous structure and consistent grayscale. S6 takes the composite reflectance image as input, performs defect identification and reflectance anomaly calibration operations, and feeds back the interference prediction spectrum, reconstructed data and confidence label to the light intensity distribution model to realize the closed-loop update of image restoration, defect detection and modeling parameter optimization.

[0008] Preferably, step S1 includes: Obtain the field diameter, numerical aperture, focal length, and depth of focus of the optical objective, establish a spatial reference coordinate system, and calibrate the position of the mask edge; The reflectivity parameters of the metallic material in the detection band are extracted, and the metal edge reflection response function is constructed by combining the angle between the incident angle and the surface normal. Based on the ray path tracing method, a two-dimensional light intensity superimposed distribution map is generated; Based on the light intensity threshold segmentation method, interference level regions are divided, and a mask boundary interference prediction map is generated and pixel-level labeled.

[0009] Preferably, step S2 includes: Perform spatial alignment processing between the atlas and the image to construct a pixel-level mapping relationship; Divide the area into high-risk interference areas, potential interference areas, and non-interference areas; For interference areas, the local mean contrast method and the brightness change gradient method are combined to identify areas of increased brightness; Directional gradient analysis is performed on the brightness enhancement region to extract brightness gradient parameters and establish a set of brightness gradient parameters.

[0010] Preferably, step S3 includes: The directional response is initialized based on the brightness gradient parameter set, and a directional response weight map is generated. A direction-selective response filtering operator is designed based on the reflection direction of the metal edge, and filtering is performed in the interference prediction region. The filter results are evaluated for response consistency. Abnormal spot pixels are identified and removed, and their pixel coordinates and response curves are recorded. The information of the stripped area is stored in a structured manner to provide a reference for subsequent image reconstruction and fusion processing.

[0011] Preferably, step S4 includes: Obtain the grayscale value and grayscale confidence label of the interpolated reconstructed region, and mark the stable region pixels in the original image that have not been disturbed; Based on the confidence level of each pixel, a fusion weight is assigned, and a weight transition band is established using a bilateral progressive fusion method; Weighted fusion is performed on all pixels to generate a composite reflectance image with consistent structure. The fused image is evaluated for brightness consistency and structure-texture matching score, and the evaluation results are used to adjust the fusion strategy for the next round.

[0012] Preferably, step S5 includes: Obtain the grayscale value and grayscale confidence label of the interpolated reconstructed region, and mark the stable region pixels in the original image that have not been disturbed; Based on the confidence level of each pixel, a fusion weight is assigned, and a weight transition band is established using a bilateral progressive fusion method; Weighted fusion is performed on all pixels to generate a composite reflectance image with consistent structure. The fused image is evaluated for brightness consistency and structure-texture matching score, and the evaluation results are used to adjust the fusion strategy for the next round.

[0013] Preferably, step S6 includes: Using the composite reflectance image as input, perform defect identification processing and reflectance anomaly calibration operations; The interference prediction map, pixel reconstruction data, grayscale confidence labels and recognition results are spatially mapped and integrated to establish a pixel-level data correspondence. Based on the integration results, dynamically adjust the crosstalk judgment threshold and the sensitivity parameters of the filtering operator in the light intensity distribution model; The updated model parameters are reconfigured into the image processing workflow to complete the closed-loop update of image restoration, defect detection, and modeling parameter optimization.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention, based on light intensity distribution modeling, combines directional filtering, pixel stripping, gradient reconstruction, and confidence-weighted fusion to achieve closed-loop control throughout the entire process from interference prediction and response recognition to pixel-level repair, exhibiting high robustness, high resolution, and adaptability. Especially in scenarios where false defects are prone to occur in edge regions, this method can accurately strip reflection artifacts, restore the true grayscale structure, avoid misjudgments by the detection system, effectively reduce false alarm rates and rework rates, and improve mask yield and the stability of the detection system, demonstrating significant engineering application value and technological innovation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method for clear imaging of a mask under an optical objective lens according to the present invention. Detailed Implementation

[0017] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0018] This invention provides, for example Figure 1 The method for achieving clear imaging of a mask under an optical objective lens, as shown, includes the following steps: S1. Based on the field-of-view boundary conditions of the optical objective, a light intensity distribution model for the reflective region of the mask metal edge is constructed. Combining the reflectivity characteristics of the metal edge material with the variation law of the incident angle, the coverage range of the crosstalk light path is determined, and the corresponding mask boundary interference prediction map is generated. To effectively identify and suppress optical crosstalk caused by reflections from metallic materials at the mask edges, a method is proposed to construct a light intensity distribution model of the region affected by reflections at the mask's metallic edges based on the boundary conditions of the optical objective's field of view. This method mainly includes the following steps: Based on the fundamental imaging parameters of the optical objective, including key indicators such as field of view diameter, numerical aperture, focal length, and depth of focus, the spatial mapping relationship between the objective's imaging range and the actual mask boundary is determined. On this basis, a spatial reference coordinate system covering the entire objective's field of view is established, centered on the edge region of the mask, to accurately calibrate the relative positions of the objective boundary and the mask edge. Since high numerical aperture objectives typically have a small depth of field and a large angular convergence characteristic, the metallic region at the mask edge may be located at the boundary of the imaging range, and its reflection path may enter the objective in a non-principal axis direction. Therefore, the non-uniform distribution of the incident angle at the edge of the field of view must be fully considered.

[0019] Based on the type of metal material used at the mask edge (e.g., aluminum, chromium, molybdenum) and its typical reflectance spectrum data, the specular reflectance, scattered reflectance, and angle-dependent reflection distribution characteristics of this material under the used detection light band are extracted. Combining the angle between the incident beam's exit angle and the mask surface normal, the reflection energy intensity attenuation factor at different incident angles is calculated, constructing the metal edge reflection response function. To improve modeling accuracy, correction parameters for temperature, surface roughness, or micro-contamination films on the metal reflectance can be introduced, making the reflection modeling more closely resemble the boundary optical behavior in the actual detection environment.

[0020] The "metal edge reflection response function" is a functional model used to describe the variation of the reflectivity of metal edge materials under different lighting conditions. It expresses how the reflection intensity of incident light changes with different angles, directions, and surface conditions. Its core function is to transform the physical reflection characteristics of mask edge materials into a calculable and predictable model expression, thereby clearly identifying which regions and under what conditions strong reflection will occur during the modeling stage, and locating and quantifying reflection paths that may cause interference.

[0021] The purpose of introducing this function is to move away from relying on simple empirical judgments and instead use a systematic and regular approach to accurately characterize the reflective behavior of metal edges. Edge reflection is not random but closely related to factors such as material type, surface treatment, incident angle, and illumination direction. By establishing this function, we can predict in advance which areas will experience strong reflection interference and which will not during the modeling process, effectively supporting subsequent critical operations such as optical crosstalk prediction, image stripping, and interference repair. This function plays a bridging role between physical properties and image analysis in the entire sharp imaging method, possessing significant engineering value and innovative significance.

[0022] Combining the aforementioned spatial mapping relationship with the metal reflectivity response function, a multipath geometric optics propagation model is used to simulate the propagation paths of reflected light from the metal edge at various angles, identifying non-principal axis light paths that may enter the objective lens via reflection, lateral scattering, or diffuse reflection. For each non-axial light path, it is tracked based on its incident point, reflection angle, and position and angle of entry into the objective lens, recording its possible superimposed position distribution within the objective lens's field of view. A two-dimensional light intensity superposition distribution map is constructed using light energy density as an indicator. This map is used to evaluate the intensity of reflection perturbation exerted by the metal edge region on the adjacent glass region in the imaging area, revealing the potential crosstalk influence range.

[0023] Based on the light intensity superposition distribution map, an interference prediction method based on the light intensity threshold segmentation method is introduced. The entire mask surface is divided into multiple response regions. According to the superposition value of light intensity per unit area caused by the reflected light from the metal edges in each region, it is divided into three levels: high-risk interference zone, potential interference zone, and non-interference zone. Based on this partitioning, a mask boundary interference prediction map is generated that can be dynamically updated with imaging parameters (such as objective lens magnification, incident angle changes, and operating wavelength). Each interference region in the prediction map is labeled at the pixel level so that subsequent image processing steps can use the map as a spatial reference for determining the location and degree of interference sources. This prediction map can not only be used for subsequent image stripping and reconstruction processing of this method, but also serve as the input basis for interference compensation strategies in future detection systems, realizing the combination of model feedforward optimization and interference early warning analysis.

[0024] "Interference prediction method based on light intensity threshold segmentation" refers to: after establishing a light intensity distribution model of the reflective area of ​​the mask metal edge, setting multiple preset threshold intervals according to the magnitude of the reflected light intensity, dividing the entire mask image area into strong, medium, weak or no interference levels, and generating an interference prediction map with spatial positioning capability accordingly.

[0025] Its purpose is to discretize and structure complex, continuously changing optical crosstalk signals, enabling the system to clearly identify which regions are significantly affected by reflections from metal edges and which regions are negligibly affected. This approach provides clear reference boundaries for subsequent image processing steps, guiding operations such as anomaly removal, image inpainting, and credibility assessment.

[0026] The reason for introducing this method is that optical crosstalk caused by reflections from metal edges is spatially non-uniform and numerically gradual. Directly processing the original continuous light intensity image would increase computational costs and the probability of algorithm misjudgment. However, using the light intensity threshold segmentation method not only enables rapid screening and level labeling of interference regions, but also allows for flexible adjustment of the threshold strategy as needed, improving the adaptability and accuracy of predictions. Therefore, this method has significant practical value and engineering feasibility in imaging interference modeling.

[0027] This step serves to provide an accurate and quantifiable prior knowledge base of spatial interference for subsequent image processing and interference removal. When an optical objective performs micro-reflectivity imaging on a mask, the high reflectivity of the metal structure at the mask's edge often generates non-target reflected signals that interfere with neighboring areas, forming what is known as optical crosstalk. Because the imaging field of view of an optical objective has certain boundary limitations, its edge regions are more susceptible to stray light from non-axial, lateral, and complex reflection paths. Therefore, relying solely on direct image data or static boundary masking makes it difficult to accurately predict the location and intensity of interference.

[0028] This step, by combining the boundary constraints of the optical objective lens, such as its field of view parameters, depth of focus, and numerical aperture, with the reflection behavior of the metal edge material at different incident angles, allows for the construction of a spatial light intensity interference distribution map—a mask boundary interference prediction map—based on optical principles. This map clearly indicates the potential range and intensity level of metal edge reflection under different scanning or observation angles, providing a quantitative reference for subsequent processing such as image region segmentation, anomaly signal identification, and image restoration boundary judgment. This prediction mechanism, centered on physical modeling, not only enhances the overall imaging system's understanding of boundary interference but also makes the entire image sharpening process more scientific, controllable, and reproducible, serving as a key prerequisite for achieving high-precision, automated defect detection.

[0029] S2, based on the mask boundary interference prediction map, the micro-reflectivity image is divided into regions, the regional features of the interference region where the reflectivity has an upward trend are identified, and the set of brightness gradient parameters of the reflectivity increase is extracted as the parameter reference for image filtering. To accurately identify mask edge reflection interference and support parameter-driven subsequent image filtering, a processing method is proposed that involves image region segmentation and extraction of brightness gradient parameter sets based on interference prediction maps. This process mainly includes the following steps: The mask boundary interference prediction map generated in the previous step is acquired and spatially aligned with the microreflectivity image of the current frame. During spatial alignment, a pixel-level mapping relationship between the interference map and the image is established based on the imaging magnification, field of view size, and image pixel resolution of the optical objective, ensuring that the interference intensity regions marked in the map accurately correspond to the actual pixel positions in the image. Guided by the spatially aligned map, three spatial regions—high-risk interference regions, potential interference regions, and interference-free regions—are divided in the original reflectivity image, constructing a preliminary image region segmentation layer.

[0030] Brightness analysis is performed on the identified high-risk and potential interference areas to extract image features showing an increasing reflectance trend within these regions. Specifically, grayscale variation analysis is conducted on each row or column of the image using image scanning, focusing on identifying pixel segments where the brightness value shows a rapid upward trend near the boundary region. This is then combined with historical data or the reference brightness change rate of non-interference areas to determine whether the increase exceeds the normal reflectance fluctuation range. To enhance robustness, a joint evaluation using the local mean contrast method and the brightness change gradient method can be introduced to ensure that the extracted reflectance increase trend is significant and stable.

[0031] The "local mean contrast method" and the "brightness variation gradient method" are two brightness feature evaluation methods commonly used in image analysis. In this invention, they are used to determine whether there is an abnormal increase in brightness in the mask edge region, so as to determine whether the region may be affected by metal reflection interference.

[0032] The local mean contrast method involves selecting a certain range (such as a fixed-size pixel window) around the interference prediction area in the image, and calculating the difference between the average brightness value of the pixels within this window and the brightness value of the central pixel. This difference is used to measure the brightness difference between the pixel and its surrounding background. If the difference exceeds a preset threshold, it indicates that the brightness of the area is significantly higher than that of the normal background, which may be a false highlight signal caused by reflection crosstalk.

[0033] The brightness gradient rule evaluates the rate of change of pixel brightness in an image along a certain direction. By calculating the brightness differences between adjacent pixels in the horizontal, vertical, or diagonal directions, it identifies edge structures with prominent brightness changes. In this scenario, if the brightness of a certain area shows a continuous and rapid upward trend near the mask edge, it can be considered that there is a "lifting gradient" caused by metallic reflection.

[0034] In this step, these two methods are combined for interference identification. The specific steps are as follows: First, obtain the pixel data of the interference prediction area; second, based on each pixel as the center, apply the local mean contrast method to analyze its consistency with the brightness of its neighbors and mark the abnormal brightness increase points; third, apply the brightness change gradient method to the area around the marked increase points to determine whether the brightness change direction points to the metal edge and calculate the gradient intensity; finally, only areas that simultaneously meet the conditions of "brightness significantly higher than the neighborhood average" and "stable brightness change trend with clear direction" are identified as interference areas with a reflectivity increase trend.

[0035] The aforementioned joint evaluation mechanism significantly reduces false positives and false negatives, improving the stability and robustness of interference identification. This method fully integrates spatial statistical features and brightness variation trends, making the judgment of reflection interference in images more accurate and providing solid data support for subsequent filtering and stripping processes.

[0036] For the identified brightness-enhanced regions, directional gradient analysis is further performed on their local grayscale distribution to extract representative brightness gradient change features. This process determines the dominant direction and intensity magnitude of the brightness gradient by statistically analyzing the brightness change rate of each pixel within the interference region in different directions (e.g., horizontal, vertical, and diagonal), and saves these values ​​as a parameter set. This brightness gradient parameter set accurately reflects the performance pattern of metallic edge reflection interference in the image and serves as an important basis for subsequent direction-selective filtering operations.

[0037] The extracted brightness gradient parameters are structurally organized and, combined with the spatial location index of each interference level region in the interference prediction map, a complete regional feature parameter reference table is established. This parameter table not only records the pixel coordinate range, brightness rise trend curve, and gradient direction distribution information of each interference region, but also provides specific parameter references for key operations such as operator selection, filter direction control, and interference region pixel marking in subsequent interference stripping processes. This processing method, guided by the prediction map and based on gradient trends, effectively improves the accuracy of interference region identification in images and significantly enhances the response control capability to boundary reflection interference, laying a solid data and structural foundation for ultimately achieving high-fidelity image reconstruction and accurate defect identification.

[0038] This step serves to provide accurate and targeted references for subsequent interference removal and pixel reconstruction processes in the image, thereby effectively improving the accuracy and robustness of image processing. Specifically, by using the mask boundary interference prediction map, image regions that may be affected by metal edge reflections can be preliminarily identified. However, this prediction is based solely on light intensity model simulation and lacks response analysis to actual image brightness changes, potentially leading to spatial redundancy or errors. Therefore, it is necessary to further combine real micro-reflectivity image data to finely segment and extract interference features from suspicious areas in the image. By identifying image regions that are spatially adjacent to metal boundaries and exhibit a significant increase in brightness, the location of the interference source can be pinpointed more accurately.

[0039] Furthermore, the brightness gradients within these regions are quantitatively extracted to form a parameter set, which can be used to describe the directionality, intensity distribution, and boundary morphology of brightness changes within the interference region. These parameters not only serve as the basis for subsequent direction-selective filtering operations but also define the applicable direction and boundary conditions of the interference stripping operator. By combining optical prediction information with image response features, a transition from "potentially interfered" to "specifically abnormal response" is achieved, giving the entire clear imaging method a high degree of targeting and intelligent processing capabilities. This step plays a crucial role in the entire imaging process, serving as a fundamental link for achieving accurate interference identification, precise feature extraction, and adaptive adjustment of filtering parameters.

[0040] S3, combined with the set of brightness gradient parameters, introduces a direction-selective response filtering algorithm to construct an interference stripping operator for the reflection direction of metal edges. Filtering is performed in the interference prediction area. Abnormal spot responses are identified and removed by judging the consistency of response direction. At the same time, the pixel coordinates of the stripped area and its corresponding image response curve are recorded. To effectively identify and remove image interference regions caused by the high reflectivity of metal edges, an interference removal method combining a set of brightness gradient parameters and a direction-selective response filtering algorithm is proposed. Based on the brightness gradient information extracted in the previous step, this method constructs a filtering process with directionality recognition capabilities to accurately determine and remove the directional consistency of abnormal light spots in the mask boundary region. The specific implementation mainly includes the following steps: Based on the extracted set of brightness gradient parameters, directional response initialization processing is performed on each interference prediction region in the image. This processing takes the gradient direction and magnitude of each pixel in the image as input to generate a set of directional response weights, which represent the sensitivity of that pixel to brightness changes in a specific direction. The selection of the response direction is consistent with the main brightness gradient direction identified in the previous step to avoid mistakenly filtering out normal structural edges. By structuring the gradient direction response, it is possible to initially identify which regions in the image exhibit brightness anomalies consistent with the direction of metal edges.

[0041] A direction-selective response filtering algorithm is introduced to construct a set of interference stripping operators designed specifically for the directional reflection characteristics of metal edges. These operators restrict the filtering operation to only apply to pixels with significant directional response weights, thus avoiding the incorrect processing of irrelevant regions. Each stripping operator employs a different convolution kernel shape and intensity weight distribution based on the actual reflection direction to adapt to the complex brightness transition structure near the mask boundary. During execution, the directional response of each interference region is traversed, and a score is assigned based on its response consistency. Stripping is only performed in regions where the directional consistency exceeds a set threshold.

[0042] The "Direction-Selective Response Filtering Algorithm" is an image processing algorithm whose core idea is to selectively enhance or suppress image structures in specific directions based on the directional characteristics of brightness changes in an image. In this invention, the algorithm's main function is to extract and identify brightness anomaly regions in the image that exhibit consistent directionality, targeting the directional interference characteristics of metal edge reflections, and then perform targeted interference stripping on these regions. Since metal edge reflections typically produce strong brightness changes along a specific direction, this algorithm can significantly improve the detection sensitivity of such directional interference signals and avoid misprocessing of genuine defect structures or irrelevant areas.

[0043] In existing image processing algorithms, methods such as "Gabor filter," "directional gradient filtering," "guided filtering," and "directionally separable convolution" can all serve as the basis for implementing direction-selective response filtering algorithms. These algorithms possess direction sensitivity, enabling them to enhance the response of structures in the image that align with a set direction while suppressing signals from non-target directions.

[0044] To construct a set of interference stripping operators for the directional features of metal edge reflections using a direction-selective response filtering algorithm, the following steps can be taken: First, based on the set of brightness gradient parameters extracted in the previous stage, the dominant brightness change directions within the interference region are statistically analyzed, such as along horizontal, vertical, or diagonal directions. Second, a filtering operator with a direction-sensitive weight distribution is designed for each dominant direction. The weight kernel of this operator adjusts the response intensity of brightness changes in the image according to the target direction, enhancing the gradient change in that direction and suppressing the signal response in other directions. Third, these operators are applied to the interference prediction region, filtering is performed on the pixels within the region, and the response intensity map is recorded. Finally, a consistency judgment is made on the response intensity maps, identifying those regions that are spatially clustered, directionally consistent, and continuously increasing in brightness as typical edge reflection interference regions for stripping.

[0045] The directional selective stripping operator constructed in this way can effectively distinguish between "structural edges" (such as pattern contours) and "directional interference" (such as metal reflection spots), thereby achieving more accurate and stable image sharpening processing and improving the reliability and accuracy of subsequent defect identification.

[0046] For regions meeting the stripping criteria, response consistency determination and pixel removal operations are performed. Response consistency determination refers to judging whether the gradient changes of multiple pixels within a region in a specific direction have the same trend and whether the change amplitude is within a set tolerance range. When the determination is consistent, the region is considered a typical area of ​​light spot interference caused by boundary reflection. At this time, stripping processing is performed, that is, the original grayscale values ​​of the pixels in the region are temporarily set to empty or normalized to reconstruction reference values, while the pixel coordinates of the region are marked, and the corresponding brightness change curves are extracted for subsequent reconstruction. This operation can achieve spatial isolation and response separation of local brightness distortion regions, preventing such pseudo-signals from entering the defect identification process.

[0047] The pixel coordinates and response curve information of the area to be stripped are stored in a structured form, forming a stripping feature record containing information such as location, orientation, response amplitude, and stripping intensity. This record can not only be used as a boundary reference in subsequent pixel-level image reconstruction, but also provide decision support for reliable grayscale reconstruction and region fusion performed in subsequent steps. In addition, the response curve can be compared with historical images to build a database of long-term reflectivity characteristics of mask edges, thereby improving the efficiency and accuracy of boundary anomaly identification in future detection processes.

[0048] This step aims to accurately identify and remove image interference areas caused by reflections from metal edges, thus laying a clean and structurally clear foundation for subsequent pixel reconstruction and image fusion. In the mask edge region, due to the high reflectivity and directionality of the metal material, reflected light entering the optical objective forms brightness-enhancing bands or bright spots in a specific direction in the image. These abnormal bright spots are often not real defects but are easily misidentified as contaminants, particles, or film anomalies in automatic recognition algorithms. Therefore, it is necessary to introduce a processing mechanism that can sense direction, understand brightness change trends, and possess removal capabilities.

[0049] This step combines the brightness gradient parameter set extracted in the previous stage to obtain the dominant brightness change direction within the interference region. A direction-selective response filtering algorithm is then introduced, processing only image responses consistent with the direction of metal reflection, thus avoiding interference with normal structural edges. The constructed interference stripping operator is highly targeted, focusing on regions with high directional consistency and prominent brightness changes, performing filtering, detection, and removal operations on them. This approach not only significantly improves the accuracy of interference stripping but also effectively reduces the risk of accidental damage to genuine structures in the image. Compared to traditional general filtering algorithms, this method is more discriminative and selective.

[0050] Furthermore, this step records the pixel coordinates of the extracted region and its corresponding image response curve, providing precise location and feature information for subsequent image grayscale compensation, pixel interpolation reconstruction, and structural fidelity restoration. This direction-centric recognition and processing method effectively isolates interference information in the image, achieving a three-in-one interference control closed loop of "recognition, extraction, and recording," thus improving the accuracy, stability, and engineering practicality of the entire image clarity imaging process. S4. Based on the pixel coordinates and response curve of the stripped area, extract the brightness gradient features of the adjacent undisturbed boundary transition area, construct a reliable grayscale expansion model for pixel reconstruction, perform pixel-level interpolation reconstruction of the stripped area according to the grayscale gradient trend, and generate a grayscale reliability label for each pixel in the reconstructed area. To accurately repair image interference areas caused by metal edge reflections and ensure the continuity of the overall image structure and grayscale consistency, a pixel reconstruction method based on the brightness gradient features of the boundary transition region is proposed. This method utilizes the pixel coordinates of the interference area recorded after the previous stage of stripping and the image response curve, combined with the grayscale distribution trend of the surrounding undisturbed areas, to construct a reliable grayscale expansion model. Pixel-level interpolation reconstruction is then performed on the stripped area, and each reconstructed pixel is assigned a reliability label to evaluate its grayscale reliability. This step specifically includes the following steps: The pixel coordinate set and corresponding image response curve information of the stripped region are read, and the positions of all removed or normalized pixels are marked in the image. A contiguous window of a certain radius is defined centered on each stripped pixel. Neighboring pixels that have not been stripped and have continuous brightness distribution are selected within the window as a reference region for grayscale expansion. This reference region is usually located near the boundary of the stripped region and has spatial continuity and structural similarity to the interference region, thus accurately reflecting the brightness gradient characteristics of the region in the interference-free state.

[0051] Brightness gradient analysis is performed on pixels within the adjacent reference region to extract their grayscale change rates in different directions, and a gradient trend vector set is established to describe the brightness change pattern of the current region. This gradient trend vector set includes not only directional information and change magnitude, but also indicators of local brightness stability and brightness change continuity. By weighted fusion of gradients from multiple directions, the dominant grayscale extension trend of the target region is generated to guide the grayscale filling path and numerical estimation in the subsequent interpolation and reconstruction process.

[0052] Based on the constructed set of gradient trend vectors, pixel-level interpolation reconstruction is performed. For each stripped pixel, multiple reference pixels with consistent gradient directions are selected from the adjacent reference region according to its gradient trend in the grayscale extension direction. The reconstructed grayscale value of the pixel is calculated using a linear or higher-order weighted average method. During the interpolation calculation, a direction consistency coefficient and a neighborhood brightness offset factor are introduced as weight adjustment parameters to ensure that the reconstructed grayscale not only transitions smoothly numerically but also maintains consistency with the original image structure in terms of direction. For pixels located in multi-directional boundary regions, a local grayscale averaging method is preferentially used for weighted processing to avoid abrupt grayscale changes or pseudo-structures.

[0053] During the reconstruction process, a grayscale confidence label is generated for each interpolated pixel to evaluate the reconstruction reliability of that pixel. This label is calculated based on multiple factors, including the number of reference pixels, gradient direction consistency, brightness stability of the original neighborhood, and the magnitude of grayscale changes involved in the interpolation path. The confidence labels are stored as percentages or levels for weighted allocation and quality assessment in subsequent image fusion operations. Simultaneously, the grayscale values ​​of all reconstructed pixels are associated with their confidence labels and archived to construct the data structure of the reconstructed region, providing complete input for the next stage of image fusion.

[0054] This step aims to perform pixel-level precision reconstruction of the image interference regions identified and stripped in the previous stage. This restores the integrity of the image structure and the continuity of brightness distribution, providing a reliable image foundation for subsequent image fusion and defect identification. During micro-reflectivity imaging under optical objectives, strong light interference from metal edge reflections creates abnormal brightness regions in the image. Although these regions have been stripped through directional filtering, the resulting image exhibits blank areas or grayscale discontinuities. Without repair, this will disrupt the overall image structure and affect the accuracy of the defect identification algorithm. Therefore, this step uses the pixel coordinates of the stripped region and the original image response curve as a basis to extract the brightness variation characteristics of the surrounding unaffected areas, establishing a reliable grayscale expansion model to guide the pixel interpolation reconstruction of the stripped region.

[0055] By analyzing the adjacent regions of the stripped edges, the true brightness gradient trend can be obtained, including the direction and magnitude of grayscale changes. This information reflects the natural extension characteristics of the image under undisturbed conditions. Using this gradient trend as the basis for the reconstruction path, pixel reconstruction with directional consistency and grayscale continuity can be achieved, avoiding grayscale jumps and preventing the introduction of artificial structures. More importantly, each reconstructed pixel is assigned a grayscale confidence label during calculation, quantifying the reliability of the pixel reconstruction result. The confidence label is comprehensively evaluated based on factors such as gradient consistency, the number of reference pixels, and the stability of brightness changes, serving as the core basis for weighted reference in subsequent image fusion. Through this mechanism, the system not only completes the "visual restoration" of missing image regions but also constructs a "confidential assessment" at the data level, enabling the image to possess high consistency and reliability at the structural, grayscale, and quality control levels. This processing step embodies the innovative characteristics of "structure guidance + pixel reconstruction + reliable control" in its technical path, and is a key central link in achieving artifact-free, high-fidelity image output.

[0056] S5, based on grayscale confidence labels, establishes a pixel-level region confidence weight fusion mechanism, which weights and fuses the stable region and the reconstructed region in the original image according to their respective pixel confidence, and generates a composite reflectance image with consistent structural continuity. To ensure good grayscale consistency and structural continuity in the image after interference stripping and pixel reconstruction, a pixel-level region confidence weight fusion method based on grayscale confidence labels is proposed. This method uses stable regions and reconstructed regions in the original image as fusion objects, and performs weighted fusion processing based on the confidence label corresponding to each pixel, thereby generating a composite reflectance image with continuous structure, natural texture, and optical realism. The method includes the following steps: The process involves acquiring the interpolated and reconstructed image data along with the corresponding grayscale confidence labels, establishing a one-to-one mapping between confidence labels and pixel coordinates. Based on this, undisturbed pixel regions with stable brightness distributions are extracted from the original image as high-confidence base regions, and their pixel coordinates are labeled. Simultaneously, the confidence label of each pixel in the interpolated and reconstructed region is divided into three confidence levels—high, medium, and low—according to a set standard, forming a complete pixel confidence level map to provide initial weighting for fusion calculations.

[0057] Based on the confidence level map, fusion weights are assigned to each pixel in the entire image. Pixels in stable regions of the original image are assigned fixed high weight values; for pixels in the reconstructed region, weights are dynamically assigned according to their confidence level. Higher confidence levels result in weights closer to the original region, while lower confidence levels result in appropriately lower weights to reduce the impact of potential interpolation errors. At region boundaries, a bilateral progressive fusion method is used to introduce a weight transition band, achieving a seamless connection between the original and reconstructed regions and avoiding abrupt edge changes, brightness jumps, or texture fragmentation.

[0058] A weighted fusion operation is performed on all pixels. Specifically, for each target pixel, its final grayscale value is calculated by weighting the grayscale values ​​in the original image and the reconstructed image according to their respective weights. For pixels located within the reconstructed region, they only participate in the fusion of the reconstructed image and the confidence calculation; while for pixels located in the fusion transition zone, the grayscale information and corresponding weights of both adjacent regions are used for the fusion calculation. This processing method not only preserves the true grayscale information of undisturbed areas in the original image but also takes into account the structural compensation effect of the reconstructed region, achieving a natural brightness transition, smooth boundary fusion, and good detail preservation in the overall image.

[0059] Perform full processing on the weighted fusion image Figure 1 Consistency assessment and structural continuity verification. By constructing a brightness consistency evaluation index and a structure-texture matching scoring mechanism, the fused composite image is comprehensively evaluated in dimensions such as grayscale smoothness, texture continuity, and structural coherence to ensure that the fusion process does not introduce structural distortion, duplicate textures, or false edges. The evaluation results are also fed back into the credibility labeling mechanism to dynamically adjust the fusion strategy parameters in the next batch of image processing, achieving self-learning and optimization of processing parameters. The composite reflectivity image output after fusion has highly consistent optical response characteristics and texture structure, meeting the needs of various applications such as defect identification, reflectivity anomaly detection, and image visualization analysis.

[0060] The "brightness consistency evaluation index" and the "structure and texture matching scoring mechanism" are two key methods for evaluating the quality of image fusion. In the image weighted fusion processing of this step, their role is to objectively and quantitatively evaluate the quality of the fused composite reflectance image, ensuring that there are no visual and data anomalies such as brightness abrupt changes, texture breaks, or structural mismatches introduced by the fusion algorithm between the original region and the reconstructed region. This ensures that the image is visually realistic and natural, and stable and reliable in subsequent image processing steps.

[0061] Specifically, the "brightness consistency evaluation index" refers to the statistical analysis of the grayscale values ​​of the reconstructed region and its adjacent original region in the fused image to assess whether the brightness transition is natural and smooth. Its function is to determine whether there are abrupt changes, such as discontinuities, increases, or decreases in brightness in the fused image, and it typically includes multiple dimensions such as grayscale mean deviation, brightness slope change, and brightness standard deviation fluctuation.

[0062] The "structure-texture matching scoring mechanism" is used to detect whether the fused image is continuous and consistent in terms of texture direction, edge morphology, and local structural integrity, avoiding problems such as duplicate textures, blurred boundaries, or structural distortions caused by reconstruction errors or improper weighted fusion. This scoring mechanism evaluates the texture and structural consistency between the reconstructed region and the adjacent original region by comparing the texture similarity, edge gradient distribution, and structural energy response in a specific direction.

[0063] The specific construction method includes the following steps: Based on the known reconstructed regions and their adjacent original regions in the fused image, several image sub-regions are selected for brightness and structure evaluation; For each sub-region, calculate its grayscale mean, variance, and brightness difference with neighboring pixels to construct a brightness consistency index matrix, which is used to describe the continuity of image brightness transition. For the same sub-region, perform texture direction extraction and edge gradient detection, calculate structural similarity score, including texture direction consistency and boundary structure overlap rate, and generate a structural texture matching score map; By fusing the brightness consistency index with the structure and texture score, the entire image is divided into regions for scoring and risk area labeling. This identifies areas that may have fusion anomalies and uses the results as feedback to adjust pixel weighting parameters or guide reconstruction optimization strategies.

[0064] This dual evaluation mechanism of brightness and structure enables fine-grained and comprehensive control over the quality of image fusion, effectively improving the usability, stability, and realism of the fusion results. It is an indispensable technical step for this method to achieve high-fidelity image output.

[0065] This step aims to perform pixel-level fusion processing between the original and reconstructed images. This ensures a smooth overall brightness transition while preserving the true information of the original regions to the greatest extent possible. It also compensates for structural interruptions or grayscale discontinuities caused by interference stripping, resulting in a composite image with complete structure, visual continuity, and realistic reflectivity. In the preceding steps, the system removed abnormally bright areas caused by reflections from metal edges through stripping and reconstructed pixels based on undisturbed adjacent regions. However, although these reconstructed pixels are interpolated based on true gradient trends, their brightness and texture are still speculatively generated, introducing a degree of uncertainty. Therefore, to further improve the overall image quality, a confidence-guided weighted fusion mechanism must be introduced to unify and integrate the differences in grayscale representation between the original and reconstructed regions.

[0066] By introducing grayscale confidence labels, the reliability of each reconstructed pixel can be quantified and differentiated, thereby enabling dynamic adjustment of weights during the fusion process: reconstructed pixels with high confidence are assigned greater fusion weights, forming a smooth transition with the original region; while pixels with lower confidence are assigned smaller weights to reduce their impact on the overall brightness and structure of the image. Simultaneously, undisturbed pixel regions in the original image maintain high weights to ensure that the true structure and original grayscale are fully preserved. This pixel-level weighted fusion method achieves brightness consistency, texture continuity, and structural integrity in image edge regions, effectively avoiding problems such as visual abrupt changes, texture repetition, or edge distortion in the reconstructed region.

[0067] Furthermore, this step not only completes the visual image stitching but also constructs a quality-controllable data fusion pathway for the entire image through a credibility weight control mechanism. This makes the fusion result more stable and adaptable to subsequent tasks such as defect identification and image segmentation. Therefore, this step plays a bridging role between image restoration and target recognition in the entire sharp imaging method and is an indispensable core link to ensure image reliability and structural continuity.

[0068] S6 takes the composite reflectance image as input, performs defect identification processing and reflectance anomaly calibration operation in the image, and feeds back the interference prediction spectrum, pixel reconstruction data and grayscale confidence label to the light intensity distribution model to dynamically adjust the interference area judgment parameters and filter operator parameters, so as to realize the collaborative closed-loop update of image restoration processing, defect detection processing and modeling parameter optimization processing. To achieve image defect identification, reflectivity anomaly calibration, and dynamic optimization of the entire imaging system parameters, a closed-loop update method based on composite reflectivity image input is proposed. Using the final fused composite reflectivity image as a foundation, defect identification and anomaly calibration are guided, and the identification results are integrated with intermediate data from the image processing process. This dynamically corrects the initial light intensity distribution model and interference judgment conditions, forming a linkage mechanism for image restoration, defect detection, and model updating. This process mainly includes the following steps: Using a composite reflectance image that has undergone pixel-level fusion processing as input, image defect identification and reflectance anomaly calibration are performed. This step employs a defect identification algorithm based on indicators such as brightness abrupt changes, gray-level unevenness, and local contrast enhancement to initially screen abnormal regions in the image. It then determines whether there are any abnormal signals that significantly deviate from the normal reflectance range based on the gray-level distribution characteristics and spatial location of each pixel. For regions identified as potential defects, further contextual structure analysis is used to eliminate the influence of gray-level changes caused by natural variations such as structural edges and texture nodes, ensuring the accuracy and stability of the identification results.

[0069] "Defect identification algorithms based on indicators such as brightness abrupt changes, gray-level unevenness, and local contrast enhancement" refers to a class of image analysis methods that utilize abnormal changes in the reflectance distribution of an image to determine whether there are potential defects in the image. In this step, the algorithm's role is to accurately identify and label defect areas such as contamination, particles, stress disturbances, and film unevenness that may exist in the composite reflectance image, thereby providing a basis for subsequent quality judgment and anomaly analysis. Since the image after interference stripping and pixel reconstruction already has a high degree of structural continuity, defects often exist in the form of "local brightness anomalies" or "minor contrast imbalances," requiring more sensitive and refined identification methods.

[0070] Specifically, abrupt brightness changes refer to a rapid change in brightness in a small area of ​​an image compared to its surrounding background, typically manifested as steep edges, sudden increases or decreases in reflectivity; uneven grayscale refers to random fluctuations or localized spot-like anomalies in the brightness distribution of areas in an image that should have a smooth transition, which may indicate uneven coating thickness or submicron contamination; local contrast enhancement emphasizes that the grayscale difference between a small area of ​​an image and its surrounding areas is greater than a set threshold, which may constitute a valid defect even if the overall brightness is not high due to the strong difference.

[0071] These indicators can serve as triggering conditions for defect identification and can be combined with existing image analysis algorithms such as Laplacian edge detection, local variance analysis, Retinex-based local contrast enhancement detection, or Hessian matrix feature extraction. The algorithm generally includes: multi-scale preprocessing of the image to extract brightness and texture information; calculating the three types of indicators in each sub-region; determining whether a valid defect exists by setting a joint threshold; and finally overlaying the identification results with the original image to output a defect-annotated map. This multi-factor identification method based on image physical features can improve the ability to identify weak, unstructured defects, ensuring the accuracy and comprehensiveness of the final defect detection.

[0072] The image information involved in the identification process is compared with historical processing records to extract known interference prediction maps, pixel reconstruction data, and grayscale confidence labels for each pixel. These data are then spatially mapped and integrated with the defect identification results. By establishing pixel-level correspondences between data from each processing stage in the composite image, it can be determined whether the currently identified abnormal region appears in a region previously identified as interference, reconstruction, or with low confidence, thereby achieving re-verification and data association of defect identification results.

[0073] Based on the aforementioned spatial mapping relationship and recognition verification results, the initially constructed light intensity distribution model is adjusted using feedback. Specifically, this includes updating the influence range of metal edge reflection, correcting the threshold of crosstalk coverage area, dynamically optimizing the reference range of the response curve in the interference area, and adjusting the sensitive parameters of the direction-selective response filter operator. In this process, the false positive rate and false negative distribution from the recognition results are introduced as correction criteria to improve the model's adaptive adjustment capability to different mask structures, edge coating changes, and reflection angle distributions. This enables more accurate interference localization and stripping strategy selection in the next round of image processing.

[0074] The revised model parameters are reconfigured into the image-clearing process to guide subsequent interference removal, pixel reconstruction, and region fusion operations, completing a closed-loop update of the entire image processing and modeling process. Simultaneously, the trajectory of model parameter changes and the accuracy improvement of image processing results are recorded in each update round, forming a traceable optimization data chain to provide data support for later intelligent parameter tuning, autonomous learning, and large-scale deployment. By feeding back the processing results of composite reflectivity images to the optical modeling and processing strategy formulation stage, a closed-loop collaborative control path is achieved from imaging input to defect output and then to strategy self-tuning, constructing an image enhancement and recognition system with continuous self-optimization capabilities.

[0075] This step aims to achieve closed-loop optimization and dynamic self-adjustment of the entire image processing workflow. After image sharpening, pixel reconstruction, and image fusion, the system can reverse-correct the model parameters and processing strategies of previous stages based on the final output, thereby continuously improving the accuracy of image restoration, the precision of defect identification, and the overall system's adaptability and robustness. The composite reflectance image, as the final image after fusion processing, possesses continuous structure and stable texture, making it an ideal input for defect identification and abnormal reflectance calibration. By performing defect detection on this image, not only can potential particulate contamination, stress disturbances, and coating unevenness be discovered, but abnormal areas can also be classified and located using brightness, boundary, and contrast features, resulting in clear defect identification results.

[0076] More importantly, during defect identification, the system compares and integrates the identification results with key intermediate data generated during previous processing, including interference prediction maps, pixel reconstruction records, and grayscale confidence labels. By comparing the spatial overlap between the identified defect location and this data, the system can determine whether the current identification result stems from insufficient processing, excessive stripping, or inaccurate reconstruction. This data linkage mechanism effectively eliminates false defects, improves identification accuracy, and provides feedback information on the "source of identification error" to the system.

[0077] Based on the feedback above, the system will re-evaluate the light intensity distribution model constructed in the initial stage, update the interference judgment threshold, adjust the filtering response direction parameters, and optimize the reconstruction strategy. This will make the next round of image processing more accurate and efficient, truly achieving "data-driven parameter optimization" and "detection-guided image enhancement." This closed loop not only improves the effect of a single image processing operation but also constructs an adaptive image processing framework with learning and iterative capabilities. This enables the entire system to maintain a high level of clear imaging and stable detection under different mask structures and optical interference characteristics, demonstrating significant engineering practicality and technological advancement.

[0078] The aforementioned method for clear imaging of photomasks under optical objectives effectively solves the optical crosstalk problem caused by high reflectivity interference at metal edges, significantly improving the imaging clarity and defect identification accuracy of blank glass photomasks at their edges. Based on light intensity distribution modeling, this method combines directional filtering, pixel stripping, gradient reconstruction, and confidence-weighted fusion to achieve closed-loop control throughout the entire process from interference prediction and response recognition to pixel-level repair, exhibiting high robustness, high resolution, and adaptability. Especially in scenarios where false defects are prone to occur at edges, this method can accurately remove reflection artifacts, restore the true grayscale structure, avoid misjudgments by the detection system, effectively reduce false alarm rates and rework rates, improve mask yield and the stability of the detection system, and has significant engineering application value and technological innovation significance.

[0079] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0080] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0081] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0082] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0083] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0086] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0087] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for clear imaging of a mask under an optical objective lens, characterized in that, Includes the following steps: S1. Based on the field-of-view boundary conditions of the optical objective lens, combined with the reflection characteristics of the metal edge material and the change law of the incident angle, a light intensity distribution model is constructed to determine the coverage range of metal edge reflection crosstalk and generate a mask boundary interference prediction map. S2, based on the mask boundary interference prediction map, divide the micro-reflectivity image into regions, extract the brightness gradient features of the reflectivity increase in the interference region, and generate the set of brightness gradient parameters required for image filtering. S3, combined with the brightness gradient parameter set, introduces the direction-selective response filtering algorithm to construct an interference stripping operator, performs direction consistency judgment and abnormal response stripping on the interference area, and records the stripped pixel coordinates and the corresponding image response curve; S4. Based on the pixel coordinates and response curves of the stripped area, extract the brightness gradient features of the adjacent undisturbed area, construct a reliable grayscale expansion model, perform pixel-level interpolation reconstruction according to the grayscale trend, and generate grayscale reliability labels. S5, based on grayscale confidence labels, establishes a pixel-level weighted fusion mechanism to weightedly fuse the stable region of the original image with the reconstructed region, generating a composite reflectance image with continuous structure and consistent grayscale. S6 takes the composite reflectance image as input, performs defect identification and reflectance anomaly calibration operations, and feeds back the interference prediction spectrum, reconstructed data and confidence label to the light intensity distribution model to realize the closed-loop update of image restoration, defect detection and modeling parameter optimization.

2. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S1 includes: Obtain the field diameter, numerical aperture, focal length, and depth of focus of the optical objective, establish a spatial reference coordinate system, and calibrate the position of the mask edge; The reflectivity parameters of the metallic material in the detection band are extracted, and the metal edge reflection response function is constructed by combining the angle between the incident angle and the surface normal. Based on the ray path tracing method, a two-dimensional light intensity superimposed distribution map is generated; Based on the light intensity threshold segmentation method, interference level regions are divided, and a mask boundary interference prediction map is generated and pixel-level labeled.

3. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S2 includes: Perform spatial alignment processing between the atlas and the image to construct a pixel-level mapping relationship; Divide the area into high-risk interference areas, potential interference areas, and non-interference areas; For interference areas, the local mean contrast method and the brightness change gradient method are combined to identify areas of increased brightness; Directional gradient analysis is performed on the brightness enhancement region to extract brightness gradient parameters and establish a set of brightness gradient parameters.

4. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S3 includes: The directional response is initialized based on the brightness gradient parameter set, and a directional response weight map is generated. A direction-selective response filtering operator is designed based on the reflection direction of the metal edge, and filtering is performed in the interference prediction region. The filter results are evaluated for response consistency. Abnormal spot pixels are identified and removed, and their pixel coordinates and response curves are recorded. The information of the stripped area is stored in a structured manner to provide a reference for subsequent image reconstruction and fusion processing.

5. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S4 includes: Obtain the grayscale value and grayscale confidence label of the interpolated reconstructed region, and mark the stable region pixels in the original image that have not been disturbed; Based on the confidence level of each pixel, a fusion weight is assigned, and a weight transition band is established using a bilateral progressive fusion method; Weighted fusion is performed on all pixels to generate a composite reflectance image with consistent structure. The fused image is evaluated for brightness consistency and structure-texture matching score, and the evaluation results are used to adjust the fusion strategy for the next round.

6. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S5 includes: Obtain the grayscale value and grayscale confidence label of the interpolated reconstructed region, and mark the stable region pixels in the original image that have not been disturbed; Based on the confidence level of each pixel, a fusion weight is assigned, and a weight transition band is established using a bilateral progressive fusion method; Weighted fusion is performed on all pixels to generate a composite reflectance image with consistent structure. The fused image is evaluated for brightness consistency and structure-texture matching score, and the evaluation results are used to adjust the fusion strategy for the next round.

7. The method for clear imaging of a mask under an optical objective lens according to claim 1, characterized in that, Step S6 includes: Using the composite reflectance image as input, perform defect identification processing and reflectance anomaly calibration operations; The interference prediction map, pixel reconstruction data, grayscale confidence labels and recognition results are spatially mapped and integrated to establish a pixel-level data correspondence. Based on the integration results, dynamically adjust the crosstalk judgment threshold and the sensitivity parameters of the filtering operator in the light intensity distribution model; The updated model parameters are reconfigured into the image processing workflow to complete the closed-loop update of image restoration, defect detection, and modeling parameter optimization.

Images