Image screening method, device, equipment and medium

By using the feature vectors and clustering methods of frequency domain images in the lithography process, the representative image sets are screened out, and the problem of low processing efficiency in the collaborative optimization technology of light source mask is solved, and efficient and accurate lithography process is achieved , suitable for a variety of scenarios.

CN118691865BActive Publication Date: 2024-11-15HUAXINCHENG (HANGZHOU) TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411185630.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-11-15
Estimated Expiration
2044-08-27

AI Technical Summary

Technical Problem

The existing light source mask collaborative optimization technology relies on key graphics selection, high accuracy and computing power requirements, and cannot handle the full chip requirements, resulting in low processing efficiency, especially in time-critical application scenarios, which is difficult to achieve fast and efficient lithography processes.

Method used

By obtaining the frequency domain images of the image set to be filtered, generating feature vectors and clustering, filtering out representative image sets, reducing data dimensions and calculation costs, using sparse sampling and clustering methods, suitable for various scenarios, and improving processing efficiency and precision.

Benefits of technology

It realizes that without relying on historical data and experience, the processing efficiency and accuracy of the lithography process is improved, and is suitable for various scenarios, reducing computing power requirements, simplifying the process, and improving the generalization of the method.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118691865B_ABST
    Figure CN118691865B_ABST
Patent Text Reader

Abstract

The present application discloses an image screening method, device, equipment and medium, including: obtaining a set of images to be screened, and determining the frequency domain images of all the images to be screened in the set of images to be screened at the target frequency, and generating a feature vector for each frequency domain image; clustering the frequency domain images based on the feature vectors to obtain a number of cluster image sets; screening out a number of cluster images from each cluster image set respectively, so as to obtain a screened image set that can characterize the set of images to be screened; the number of images in the screened image set is less than the number of images in the set of images to be screened. It can be seen that the present application can only use frequency domain images at the target frequency to achieve sparse sampling, which is conducive to improving processing efficiency, can be applied to various scenarios, improve the generalization of the method, and does not rely on historical data and experience combined with subsequent clustering.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an image screening method, device, equipment and medium. Background Art

[0002] See also Figure 1 As shown in the figure, a typical scene diagram of the existing screening of the image to be screened (the whole set of patterns) is shown. Starting from the initial whole set of patterns, an algorithm is used to screen and generate a subset of representative patterns. After performing SMO (Source Mask Optimization) on the subset, the result of SMO is verified. If the result is not good, the hotspot image is added to the subset and SMO is performed again and verified; if the result is good, the whole set is corrected and verified by OPC (OLE (Object Linking and Embedding) for Process Control) using the subset until the performance of the whole set and the subset meet the requirements (that is, until convergence and no serious hotspots), and can be comparable, then the iteration can be ended, and the subsequent lithography process can be carried out using this source, that is, the iteration result. This process is mainly to find a suitable light source for the whole chip. Since the calculation of SMO is very time-consuming, it is necessary to select a representative pattern for calculation. SMO is calculating the light source, and verification is checking whether there is a problem with the result. The reason for iteration is that it is often impossible to achieve the optimal result in one calculation, and it is necessary to reverse the result and find the hotspots before adding them to the original representative pattern set for light source optimization. The lithography standard in the figure may change, so it is necessary to determine in real time whether the lithography standard has changed, and if so, change the lithography standard.

[0003] At present, the accuracy of light source mask collaborative optimization is heavily dependent on the selection of key graphics, which comprehensively and effectively determines the final accuracy of OPC, and thus directly affects the yield rate. In addition, light source mask collaborative optimization requires extremely high computing power, and usually cannot handle the needs of the entire chip, but instead processes limited key graphics. Using fast key graphics screening can make many high-computing power or time-sensitive application scenarios possible. Therefore, the industry currently generally uses key graphics selection to improve speed and accuracy, most of which are based on experience and node database selection. A small number of selection methods are based on image processing and machine learning.

[0004] The first type is based on experience or existing databases. The advantages are accuracy and targeting. The disadvantages are that it is unexplainable, relies heavily on expert experience, requires multiple trials and errors, and experience is difficult to summarize. In addition, the use of experience and databases often has the problem of instability and difficulty in compressing the number of patterns.

[0005] The second category is based on image feature extraction and machine learning. The advantage is automation, but the disadvantage is that it is heavily dependent on data and algorithms, unstable, poor generalization, high computing power requirements, inability to quickly obtain information in a large number of scenarios, high trial and error costs, and in some scenarios, the algorithm fails due to the inability to obtain confidential data. Summary of the invention

[0006] In view of this, the object of the present invention is to provide an image screening method, device, equipment and medium, which can improve processing efficiency, is applicable to various scenarios, and does not rely on historical data and experience. The specific scheme is as follows:

[0007] In a first aspect, the present application discloses an image screening method, comprising:

[0008] Acquire a set of images to be screened, determine frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generate a feature vector of each frequency domain image;

[0009] Clustering the frequency domain image based on the feature vector to obtain a plurality of cluster image sets;

[0010] A number of cluster images are respectively screened out from each of the cluster image sets to obtain a screened image set capable of representing the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0011] Wherein, generating a feature vector of each frequency domain image includes:

[0012] Dividing a plurality of target intervals from the frequency domain image according to a preset interval division rule;

[0013] Integrating the light source intensity for each target interval to obtain a first interval intensity;

[0014] The feature vector of the frequency domain image is determined based on the first interval intensity to generate a feature vector of each of the frequency domain images.

[0015] The step of dividing a plurality of target intervals from the frequency domain image according to a preset interval division rule includes:

[0016] Dividing a plurality of target intervals from the frequency domain image based on symmetry;

[0017] Alternatively, a specific image in a specific direction is divided from the frequency domain image, and a plurality of target intervals are divided from the specific image; the number of directions in the specific direction is not less than 1.

[0018] The step of determining the feature vector of the frequency domain image based on the first interval intensity to generate a feature vector of each frequency domain image includes:

[0019] Integrate the target interval by a preset integration method to obtain an integrated interval;

[0020] Obtaining a second interval intensity of each of the integrated intervals based on the first interval intensity, and determining a feature vector of the frequency domain image based on the second interval intensity, so as to generate a feature vector of each of the frequency domain images;

[0021] The preset integration method includes interval merging and interval removal.

[0022] Wherein, determining the feature vector of the frequency domain image based on the second interval intensity includes:

[0023] Determine an intensity combination corresponding to the frequency domain image based on the second interval intensity;

[0024] The intensity of the second interval in the intensity combination is subjected to intensity normalization processing to obtain a feature vector of the frequency domain image.

[0025] Before selecting a plurality of cluster images from each of the cluster image sets, the method further comprises:

[0026] Sorting the cluster images in each of the cluster image sets to obtain a priority of each of the cluster images;

[0027] Accordingly, a plurality of cluster images are selected from each of the cluster image sets, including:

[0028] A plurality of cluster images are respectively selected from each of the cluster image sets based on the priority.

[0029] Wherein, after filtering out a plurality of cluster images from each of the cluster image sets to obtain a filtered image set capable of representing the image set to be filtered, the method further comprises:

[0030] Calculate the inter-set similarity and intra-set similarity of the filtered image set;

[0031] and / or, calculating the percentage of the frequency domain occupancy of the screened image in the frequency domain occupancy of the entire set of images to be screened;

[0032] And / or, comparing the key frequency occupancy determined based on the historical image set with the frequency domain occupancy of the screened image to evaluate the degree to which the screened plurality of cluster images represent the image set to be screened.

[0033] In a second aspect, the present application discloses an image screening device, comprising:

[0034] A feature vector generation module is used to obtain a set of images to be screened, determine frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generate a feature vector of each frequency domain image;

[0035] A clustering module, used for clustering the frequency domain image based on the feature vector to obtain a plurality of cluster image sets;

[0036] The screening module is used to screen out a number of cluster images from each of the cluster image sets, so as to obtain a screened image set that can represent the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0037] In a third aspect, the present application discloses an electronic device, including:

[0038] Memory, used to store computer programs;

[0039] A processor is used to execute the computer program to implement the image screening method disclosed above.

[0040] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the image screening method disclosed above is implemented.

[0041] It can be seen that the present application obtains a set of images to be screened, determines the frequency domain images of all the images to be screened in the set of images to be screened at the target frequency, and generates a feature vector for each of the frequency domain images; clusters the frequency domain images based on the feature vectors to obtain a number of cluster image sets; selects a number of cluster images from each of the cluster image sets, so as to obtain a screened image set that can characterize the set of images to be screened; the number of images in the screened image set is less than the number of images in the set of images to be screened. It can be seen that the present application obtains a feature vector at the target frequency in the frequency domain, that is, no overall image is required, only the frequency domain image at the target frequency of the image is required, so sparse sampling greatly reduces the dimension of the original data by extracting feature vectors, which is conducive to improving speed and accuracy; in addition, the clustering method can effectively classify the images to be screened in the set of images to be screened, so that different sets of images to be screened are representative, and then the images are selected from a number of cluster image sets respectively so that the selected images can better characterize the set of images to be screened; the sparse sampling and clustering methods of the present application can be applied to various scenarios, improve the generalization of the method, and do not rely on historical data and experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a typical scene of existing screening of images to be screened;

[0044] Figure 2 A flow chart of an image screening method disclosed in this application;

[0045] Figure 3 A schematic diagram of an image screening process disclosed in this application;

[0046] Figure 4 A flowchart of a specific image screening method disclosed in this application;

[0047] Figure 5 This is a schematic diagram of an interval division disclosed in this application;

[0048] Figure 6 This is another schematic diagram of interval division disclosed in this application;

[0049] Figure 7 A flowchart of a specific image screening method disclosed in this application;

[0050] Figure 8 This is a schematic diagram of the structure of an image screening device disclosed in this application;

[0051] Fig. 9 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] At present, the accuracy of light source mask collaborative optimization is heavily dependent on the selection of key graphics, which comprehensively and effectively determines the final accuracy of OPC, and thus directly affects the yield rate. In addition, light source mask collaborative optimization requires extremely high computing power, and usually cannot handle the needs of the entire chip, but instead processes limited key graphics. Using fast key graphics screening can make many high-computing power or time-sensitive application scenarios possible. Therefore, the industry currently generally uses key graphics selection to improve speed and accuracy, most of which are based on experience and node database selection. A small number of selection methods are based on image processing and machine learning.

[0054] The first type is based on experience or existing databases. The advantages are accuracy and pertinence. The disadvantages are that it is unexplainable, relies heavily on expert experience, requires multiple trials and errors, and is difficult to summarize. In addition, the use of experience and databases often has the problem of instability and difficulty in compressing the number of patterns (images).

[0055] The second category is based on image feature extraction and machine learning. The advantage is automation, but the disadvantage is that it is heavily dependent on data and algorithms, unstable, poor generalization, high computing power requirements, a large number of scenarios cannot quickly obtain information, the cost of trial and error is high, and in some scenarios, the algorithm fails due to the inability to obtain confidential data.

[0056] To this end, an embodiment of the present application proposes an image screening solution that can improve processing efficiency, is applicable to various scenarios, and is independent of historical data and experience.

[0057] The present application embodiment discloses an image screening method, see Figure 2 As shown, the method includes:

[0058] Step S11: obtaining a set of images to be screened, determining frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generating a feature vector of each frequency domain image.

[0059] In this embodiment, the number of original images in the image set to be screened is relatively large, and it is necessary to screen out images that can represent the image set to be screened from the image set to be screened, so as to improve the speed of subsequent image processing. Only the frequency domain image at the target frequency is used to achieve sparse sampling, and very high efficiency can be achieved with extremely low computational cost. This method is highly related to lithography behavior and can achieve very high prediction accuracy. Specifically, sparse sampling mainly reduces the dimension of the original data by greatly extracting feature vectors. For example, a picture has a dimension of 128×128 in both the spatial domain and the frequency domain. After extracting the feature vector, the dimension can be reduced to a one-dimensional vector of 1×10. Since the information is more concentrated after sparse sampling, this brings huge benefits to data processing, algorithms, and computing power, and is also conducive to improving accuracy.

[0060] See also Figure 3 As shown, it is a schematic diagram of an image screening process; the figure includes three key steps, specifically: first, generating a frequency domain image and generating a feature vector; second, clustering; third, screening.

[0061] In this embodiment, a periodic pattern often has a discrete frequency domain spectrum, but a non-periodic pattern has a continuous frequency domain spectrum. This method does not distinguish whether the spectrum source is a periodic or non-periodic structure, and uniformly processes, simplifies the process and has universality.

[0062] In this embodiment, in the application of SMO and some specific scenarios, not all frequency intervals are of equal importance. Therefore, it is possible to directly select to generate a frequency domain image at the target frequency, and there is no need to generate a frequency domain image of the entire image, thereby further reducing the pressure of subsequent processing.

[0063] In this embodiment, the target frequency can be set according to actual conditions and is again not specifically limited.

[0064] Step S12: clustering the frequency domain image based on the feature vector to obtain a plurality of clustered image sets.

[0065] In this embodiment, the clustering method can be fuzzy clustering or strict clustering; strict clustering (strictmatch / group) means that the patterns within the class are strictly identical from the perspective of custom resolution; fuzzy clustering (fuzzy match / group) means that the patterns within the class are not strictly identical, but similar.

[0066] Step S13: Screening out a number of cluster images from each of the cluster image sets to obtain a screened image set that can represent the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0067] In this embodiment, before filtering out a number of cluster images from each of the cluster image sets, it includes: sorting the cluster images in each of the cluster image sets to obtain the priority of each of the cluster images; correspondingly, filtering out a number of cluster images from each of the cluster image sets includes: filtering out a number of cluster images from each of the cluster image sets based on the priority.

[0068] It should be pointed out that the priority can be calculated using various methods, again without specific limitation. In a specific embodiment, for example, the distance between two images can be calculated according to a certain mathematical similarity, and a priority can be obtained by sorting according to this distance, and the front image is taken according to this priority.

[0069] In this embodiment, after filtering out a number of cluster images from each of the cluster image sets to obtain a filtered image set that can represent the image set to be filtered, it also includes: evaluating the degree of representation of the image set to be filtered by the filtered out cluster images; if the representation degree does not meet the preset degree, continuing to filter out a number of cluster images from each of the cluster image sets based on the priority, and then jumping to the step of evaluating the degree of representation of the image set to be filtered by the filtered out cluster images, until the representation degree meets the preset degree.

[0070] It should be pointed out that the use of interval algorithms can reduce the complexity of the algorithm, and is compatible with both strict and fuzzy scenarios, and can be combined with screening algorithms and verification algorithms for iterative optimization.

[0071] It can be seen that the present application obtains a set of images to be screened, determines the frequency domain images of all the images to be screened in the set of images to be screened at the target frequency, and generates a feature vector for each of the frequency domain images; clusters the frequency domain images based on the feature vectors to obtain a number of cluster image sets; screens out a number of cluster images from each of the cluster image sets respectively, so as to obtain a screened image set that can characterize the set of images to be screened; the number of images in the screened image set is less than the number of images in the set of images to be screened. It can be seen that the present application obtains a feature vector at the target frequency in the frequency domain, that is, no overall image is required, only a frequency domain image at the target frequency of the image is required. Therefore, sparse sampling greatly reduces the dimension of the original data by extracting feature vectors, which is beneficial to improving speed and accuracy. In addition, the clustering method can effectively classify the images to be screened in the image set to be screened, so that different image sets to be screened are representative, and then images are screened from several clustered image sets respectively so that the screened images can better represent the image set to be screened. The sparse sampling and clustering methods of the present application can be applied to various scenarios, improve the generalization of the method, and do not rely on historical data and experience. The calculation of the degree of representation and the subsequent re-screening of the clustered images ensure the degree of representation of the clustered image set for the image set to be screened.

[0072] The embodiment of the present application discloses a specific image screening method. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 4 As shown, specifically including:

[0073] Step S21: obtaining a set of images to be screened, and determining frequency domain images of all images to be screened in the set of images to be screened at a target frequency.

[0074] Step S22: dividing a plurality of target intervals from the frequency domain image according to a preset interval division rule.

[0075] In this embodiment, dividing a number of target intervals from the frequency domain image according to a preset interval division rule includes: dividing a number of target intervals from the frequency domain image based on symmetry; or dividing a specific image in a specific direction from the frequency domain image, and dividing a number of target intervals from the specific image; the number of directions in the specific direction is not less than 1.

[0076] It should be pointed out that the method of dividing the frequency domain image into several target intervals based on symmetry is suitable for scenarios that care about images in all directions; the method of dividing a specific image in a specific direction from the frequency domain image and dividing several target intervals from the specific image is suitable for scenarios that care about images in some directions.

[0077] In a specific embodiment, see Figure 5As shown, it is a schematic diagram of interval division, which is to divide a specific image in a specific direction from the frequency domain image, and divide a number of target intervals from the specific image; the number of directions in the specific direction is not less than 1. Specifically, in the figure, a Y interval is taken out from the two-dimensional Fourier spectrum ( ) in the X direction ( ) on the spectrum, the two-dimensional spectrum fragment is partitioned according to the basic interval of X. The figure is divided into five areas, marked as 0, 1, 2, 3, and 4 (due to symmetry, only half of the area is partitioned).

[0078] It should be pointed out that the spectra in multiple directions of a Y interval can also be extracted from the two-dimensional Fourier spectrum, and multiple two-dimensional spectrum segments can be partitioned according to the basic intervals in their respective directions.

[0079] In another specific embodiment, see Figure 6 As shown, it is another schematic diagram of interval division, which is to divide several target intervals from the frequency domain image based on symmetry. Specifically, the figure divides the two-dimensional Fourier spectrum into concentric circles ( ) segmentation method, the figure is divided into 6 areas, marked with 0, 1, 2, 3, 4, and 5 respectively.

[0080] It should be pointed out that not only concentric circle segmentation can be performed, but any segmentation method that satisfies symmetry can be used, such as concentric square segmentation, concentric rectangle segmentation, etc.

[0081] Step S23: integrating the light source intensity for each target interval to obtain a first interval intensity; determining a feature vector of the frequency domain image based on the first interval intensity to generate a feature vector for each frequency domain image.

[0082] In this embodiment, the light source intensity is integrated for each target interval to obtain the first interval intensity, specifically, the intensity is integrated for each partition. The specific integration formula is as follows:

[0083] , is the frequency, For this interval end frequency, The starting frequency for this interval.

[0084] In a specific embodiment, see Figure 5 As shown in the figure, due to the symmetry, we only need to integrate half of it.

[0085] In another specific embodiment, see Figure 6 As shown, each partition is integrated.

[0086] In this embodiment, the feature vector of the frequency domain image is determined based on the first interval intensity to generate a feature vector for each of the frequency domain images, including: integrating the target interval by a preset integration method to obtain an integrated interval; obtaining a second interval intensity of each of the integrated intervals based on the first interval intensity, and determining the feature vector of the frequency domain image based on the second interval intensity to generate a feature vector for each of the frequency domain images; wherein the preset integration method includes interval merging and interval removal. Merging intervals can reduce the dimension of a vector. After merging, the interval is regarded as a frequency placeholder. Interval merging or interval removal can be used alone to integrate intervals, or interval merging and interval removal can be used simultaneously to integrate intervals.

[0087] In a specific embodiment, see Figure 5 As shown, the partitions are 0, 1, 2, 3, and 4. When integrating, you can consider deleting dispensable partitions, for example, delete 4. In this case, 10 of the first interval intensities 150, 80, 30, 25, and 10 becomes 0. When integrating, you can also consider merging the intensities of two important intervals, such as 1 and 2. In this case, the first interval intensities 150, 80, 30, 25, and 10 become 150, 110, 30, and 10. At this time, the intensities of 1 and 2 are improved compared to before.

[0088] In another specific embodiment, see Figure 5 As shown, the partitions are 0, 1, 2, 3, 4, and 5. When integrating, you can consider deleting dispensable partitions, for example, deleting 0. In this case, 150 in the first interval strength 150, 80, 30, 25, 10, 5 becomes 0. When integrating, you can also consider merging the strengths of two important intervals, such as 1 and 2. In this case, the first interval strengths 150, 80, 30, 25, 10 become 150, 110, 30, 10, 5. At this time, the strengths of 1 and 2 are improved compared to before, and 110 is regarded as a frequency placeholder.

[0089] The preset integration method may also include setting weights, which may be used alone or together with interval merging and interval removal. For example, in the first interval intensities of 150, 80, 30, 25, and 10, a weight is set for each intensity. For example, if 1 and 2 are important, the weights of 80 and 30 are set to 40% and 40% respectively, and the other 150 is 10%, and 20 and 10 are both 5%. Thus, the second interval intensities may be calculated to be 15, 32, 12, 1.25, and 0.5.

[0090] It should be pointed out that for periodic images, the preset integration method can also be as follows: when the frequency is occupied, if there are f, 2f, and 3f, multiple frequency spectra with a multiple relationship can be regarded as a placeholder, because the maximum effective frequency can meet the needs of SMO. At this time, the placeholder only counts the maximum frequency placeholder as 1, and the rest are 0.

[0091] In this embodiment, determining the feature vector of the frequency domain image based on the second interval intensity includes: determining an intensity combination corresponding to the frequency domain image based on the second interval intensity; and performing intensity normalization processing on the second interval intensity in the intensity combination to obtain the feature vector of the frequency domain image. It should be noted that the normalized data is simpler and is conducive to subsequent calculations.

[0092] In a specific embodiment, if 150, 80, 30, 25, and 10 are normalized to 1, 0.5, 0.3, 0.1, and 0.05, they are no longer normalized. Figure 5 150, 110, 30, 10 and Figure 6 The normalized results of 150, 110, 30, 10, and 5 are described in detail.

[0093] It should be noted that before normalization, noise reduction processing can be performed on data that is less than a preset value to prevent the data from being entirely provided by noise. The preset value can be determined according to actual conditions and is not specifically limited here.

[0094] Step S24: clustering the frequency domain image based on the feature vector to obtain a plurality of clustered image sets.

[0095] Step S25: Screening out a number of cluster images from each of the cluster image sets to obtain a screened image set that can represent the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0096] In the present embodiment, it should be pointed out that multiple target frequencies can be set in advance, and images of the same or different directions can be extracted from frequency domain images of different target frequencies to generate a feature vector corresponding to each spectrum image. The feature vectors of different spectrum images of the same image to be screened can be combined to obtain a new feature vector, so that the new feature vector can be used to complete clustering and screening operations later.

[0097] It can be seen that the present application obtains a set of images to be screened, and determines the frequency domain images of all the images to be screened in the set of images to be screened at the target frequency; divides a number of target intervals from the frequency domain image according to a preset interval division rule; integrates the light source intensity for each of the target intervals to obtain a first interval intensity; determines the feature vector of the frequency domain image based on the first interval intensity to generate a feature vector for each of the frequency domain images; clusters the frequency domain image based on the feature vector to obtain a number of cluster image sets; screens a number of cluster images from each of the cluster image sets to obtain a screened image set that can characterize the set of images to be screened; the number of images in the screened image set is less than the number of images in the set of images to be screened. It can be seen that the present application proposes a variety of target interval division methods, which can be used in a variety of scenarios to improve the generalization of the method.

[0098] The embodiment of the present application discloses a specific image screening method. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. Figure 7 As shown, specifically including:

[0099] Step S31: obtaining a set of images to be screened, determining frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generating a feature vector of each frequency domain image.

[0100] Step S32: clustering the frequency domain image based on the feature vector to obtain a plurality of clustered image sets.

[0101] Step S33: a plurality of cluster images are respectively screened out from each of the cluster image sets to obtain a screened image set capable of representing the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0102] Step S34: Calculate the inter-set similarity and intra-set similarity of the screened image set; and / or, calculate the percentage of the frequency domain occupancy of the screened image in the full set frequency domain occupancy of the image set to be screened; and / or, compare the key frequency occupancy determined based on the historical image set with the frequency domain occupancy of the screened image to evaluate the degree of representation of the image set to be screened by the screened several cluster images.

[0103] In this embodiment, before performing SMO, the degree of representation can be pre-evaluated, so that when the degree of representation does not meet the preset degree, the image is screened again to the screened image set for iterative optimization until the degree of representation meets the preset degree. There are three ways to evaluate the degree of representation of the selected clustered images to the image set to be screened. The first is to calculate the inter-set similarity and intra-set similarity of the screened image set, wherein the higher the intra-set similarity, the lower the inter-set similarity, and the higher the degree of representation. The degree of representation can be calculated according to the weight and the intra-set similarity and inter-set similarity; the second is to calculate the percentage of the frequency domain occupancy of the screened image in the frequency domain occupancy of the entire set of the image set to be screened, and the higher the percentage, the higher the degree of representation; the third is to compare the key frequency occupancy or hotspot determined based on the historical image set with the frequency domain occupancy or hotspot frequency occupancy of the screened image, and the more the same comparisons, the higher the degree of representation. The historical image set is the historical screened image set.

[0104] It should be noted that for the second method, assuming the base frequency is , within the effective frequency range, 100 intervals (total number of digits) can be divided based on this frequency interval. When all 100 frequency positions are occupied, it is 100%. The placeholders of the selected filtered image set can be calculated. The more placeholders there are and the more evenly distributed they are, the higher the placeholder ratio and the better the coverage. Due to the problem of frequency doubling, the calculation of coverage needs to take frequency doubling into account. Multiple frequency doublings are regarded as one placeholder and need to be processed in advance. There may be some frequencies that are more important than other frequencies. In this case, weights can be set in each basic frequency interval, and the number of digits occupied is expressed as weighted sum. In a specific embodiment, the frequency occupancy of the entire set of images to be screened is first calculated. Specifically, it is assumed that the distribution of the occupancy in the entire set is known. According to the resolution definition, there are n different frequency bits, and the frequency bits of the entire set are <1, 1, 1, 1, 1, 1, 0, 1>. If there are weights, the weighted sum is calculated, and it is assumed that there are 100 occupants in total. Then, the occupancy of the screened image set can be calculated, that is, the occupancy of all the screened images is put together, and as long as there is a place, the frequency bit is changed to 1, and the counting is no longer repeated. If there is a weight, the weighted sum is calculated, and it is assumed that the result is 80. Then, the number of occupied bits is divided by the frequency occupancy (total number of bits) of the entire set of images to be screened to obtain a coverage rate of 80%. Then, the frequency occupancy of a single screened image can be calculated, that is, the frequency occupancy of a single screened image is calculated, and the result is 5. The frequency occupancy of the screened image is divided by the frequency occupancy (total number of bits) of the entire set of images to be screened to obtain a coverage rate of 5%.

[0105] It should be pointed out that SMO can also be performed on the clustered images first, and the obtained results can be used to judge the quality of the screening, specifically using tools such as LMC for evaluation, which is not specifically limited here.

[0106] It should be pointed out that if three different algorithms are used for clustering, three different filtered image sets are finally obtained, and SMO is performed on the filtered images to obtain three light source mask image sets. When historical data is used to judge the degree of representation, SMO can also be performed on the historical filtered images to obtain a historical light source mask image set. The three light source mask image sets are compared with the historical light source mask image set. The one with the highest similarity has the highest coverage and the highest degree of representation.

[0107] It can be seen that the present application obtains a set of images to be screened, determines the frequency domain images of all the images to be screened in the set of images to be screened at the target frequency, and generates a feature vector for each of the frequency domain images; clusters the frequency domain images based on the feature vectors to obtain a number of cluster image sets; screens out a number of cluster images from each of the cluster image sets respectively, so as to obtain a screened image set that can characterize the set of images to be screened; the number of images in the screened image set is less than the number of images in the set of images to be screened; calculates the inter-set similarity and intra-set similarity of the screened image set; and / or, calculates the percentage of the frequency domain occupancy of the screened image in the full set frequency domain occupancy of the set of images to be screened; and / or, compares the key frequency occupancy determined based on the historical image set with the frequency domain occupancy of the screened image, so as to evaluate the degree of characterization of the set of images to be screened by the screened several cluster images. It can be seen that the present application proposes multiple methods for calculating the degree of representation of the several clustered images to the image set to be screened, so as to obtain the degree of representation and ensure that the degree of representation of the image set to be screened finally obtained after the screening meets the conditions.

[0108] Correspondingly, the present application also discloses an image screening device, see Figure 8 As shown, the device comprises:

[0109] The feature vector generating module 11 is used to obtain a set of images to be screened, determine the frequency domain images of all the images to be screened in the set of images to be screened at a target frequency, and generate a feature vector of each of the frequency domain images;

[0110] A clustering module 12, configured to cluster the frequency domain image based on the feature vector to obtain a plurality of cluster image sets;

[0111] The screening module 13 is used to screen out a number of cluster images from each of the cluster image sets, so as to obtain a screened image set that can represent the image set to be screened; the number of images in the screened image set is less than the number of images in the image set to be screened.

[0112] Among them, the more specific working process of each of the above modules can refer to the corresponding content disclosed in the above embodiments, which will not be repeated here.

[0113] Furthermore, an embodiment of the present application also provides an electronic device. Fig. 9 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0114] Fig. 9 The present invention provides a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a display screen 23, an input / output interface 24, a communication interface 25, a power supply 26, and a communication bus 27. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the image screening method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0115] In this embodiment, the power supply 26 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 25 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 24 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0116] In addition, the memory 22, as a carrier for storing resources, may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon may include a computer program 221, and the storage method may be temporary storage or permanent storage. In addition to including a computer program that can be used to complete the image screening method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 221 may further include a computer program that can be used to complete other specific tasks.

[0117] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the image screening method disclosed above is implemented.

[0118] The specific steps of the method can refer to the corresponding contents disclosed in the above embodiments, which will not be repeated here.

[0119] The various embodiments in this application are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0120] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0121] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0122] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0123] The above is a detailed introduction to an image screening method, device, equipment, and storage medium provided by the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An image screening method, characterized in that: include: Acquire a set of images to be screened, determine frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generate a feature vector of each frequency domain image; Clustering the frequency domain image based on the feature vector to obtain a plurality of cluster image sets; Screening out a number of cluster images from each of the cluster image sets, respectively, so as to obtain a screened image set capable of representing the image set to be screened; The number of images in the screened image set is less than the number of images in the to-be-screened image set; Wherein, generating a feature vector of each frequency domain image includes: Dividing a plurality of target intervals from the frequency domain image according to a preset interval division rule; Integrating the light source intensity for each target interval to obtain a first interval intensity; determining a feature vector of the frequency domain image based on the first interval intensity to generate a feature vector of each of the frequency domain images; The step of determining the feature vector of the frequency domain image based on the first interval intensity to generate a feature vector of each frequency domain image includes: Integrate the target interval by a preset integration method to obtain an integrated interval; Obtaining a second interval intensity of each of the integrated intervals based on the first interval intensity, and determining a feature vector of the frequency domain image based on the second interval intensity, so as to generate a feature vector of each of the frequency domain images; The preset integration method includes interval merging and interval removal.

2. The image screening method according to claim 1, characterized in that: The dividing a plurality of target intervals from the frequency domain image according to a preset interval division rule comprises: Dividing a plurality of target intervals from the frequency domain image based on symmetry; Or, a target image in a target direction is divided from the frequency domain image, and a plurality of target intervals are divided from the target image; the number of directions of the target direction is not less than 1.

3. The image screening method according to claim 1, characterized in that: The determining the feature vector of the frequency domain image based on the second interval intensity includes: Determine an intensity combination corresponding to the frequency domain image based on the second interval intensity; The intensity of the second interval in the intensity combination is subjected to intensity normalization processing to obtain a feature vector of the frequency domain image.

4. The image screening method according to claim 1, characterized in that: Before selecting a plurality of cluster images from each of the cluster image sets, the method includes: Sorting the cluster images in each of the cluster image sets to obtain a priority of each of the cluster images; Accordingly, a plurality of cluster images are selected from each of the cluster image sets, including: A plurality of cluster images are respectively selected from each of the cluster image sets based on the priority.

5. The image screening method according to claim 1, characterized in that: After respectively selecting a plurality of cluster images from each of the cluster image sets to obtain a screened image set capable of representing the image set to be screened, the method further includes: Calculate the inter-set similarity and intra-set similarity of the filtered image set; and / or, calculating the percentage of the frequency domain occupancy of the screened image in the frequency domain occupancy of the entire set of images to be screened; And / or, comparing the key frequency occupancy determined based on the historical image set with the frequency domain occupancy of the screened image to evaluate the degree to which the screened plurality of cluster images represent the image set to be screened.

6. An image screening device, characterized in that: include: A feature vector generation module is used to obtain a set of images to be screened, determine frequency domain images of all images to be screened in the set of images to be screened at a target frequency, and generate a feature vector of each frequency domain image; A clustering module, used for clustering the frequency domain image based on the feature vector to obtain a plurality of cluster image sets; A screening module, used for screening out a number of cluster images from each of the cluster image sets, so as to obtain a screened image set capable of representing the image set to be screened; The number of images in the screened image set is less than the number of images in the to-be-screened image set; The feature vector generation module is specifically used to divide a number of target intervals from the frequency domain image according to a preset interval division rule; Integrating the light source intensity for each target interval to obtain a first interval intensity; determining a feature vector of the frequency domain image based on the first interval intensity to generate a feature vector of each of the frequency domain images; The step of determining the feature vector of the frequency domain image based on the first interval intensity to generate a feature vector of each frequency domain image includes: Integrate the target interval by a preset integration method to obtain an integrated interval; Obtaining a second interval intensity of each of the integrated intervals based on the first interval intensity, and determining a feature vector of the frequency domain image based on the second interval intensity, so as to generate a feature vector of each of the frequency domain images; The preset integration method includes interval merging and interval removal.

7. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the image screening method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the image screening method according to any one of claims 1 to 5 is implemented.