A method for efficiently identifying and detecting defects in surface printing of a metal plate

Abstract

The present application relates to the technical field of defect detection, and particularly relates to a high-efficiency identification and detection method for printing defects on the surface of a metal plate. The method comprises: obtaining a difference representation value and a suspected judgment index value of a pixel point to be detected, and obtaining a suspected defect pixel point according to the suspected judgment index value; clustering all suspected defect pixel points according to the suspected judgment index value and the coordinate value of the suspected defect pixel points, obtaining each cluster, and obtaining an abnormality degree representation value of the suspected defect pixel point according to the difference representation value of the suspected defect pixel point, the gradient direction of the suspected defect pixel point in the cluster to which the suspected defect pixel point belongs, and the standard deviation and mean value of the gray values of all suspected defect pixel points in the cluster to which the suspected defect pixel point belongs; and identifying printing defects on the surface of the printing metal plate to be detected according to the abnormality degree representation value. The present application can improve the accuracy of identifying the printing defect area on the surface of the printing metal plate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection, and in particular to a high-efficiency identification and detection method for surface printing defects of a metal plate. BACKGROUND

[0002] Since surface printing defects of a metal plate will affect the appearance quality, service life and functionality of a product, current surface printing defect detection of a metal plate has become an important link in quality control in the manufacturing industry, that is, surface printing defect detection of a metal plate is crucial to the appearance quality and subsequent use performance of the metal plate in the production process of the metal plate.

[0003] In addition, since printed metal plates are generally produced in batches, a printing template usually exists before the printed metal plate is printed. However, when the printed metal plate is subjected to printing defect detection, the gray-scale difference between the surface image of the printed metal plate to be detected and the standard image is generally obtained first, and then the defect pixel points are identified based on a preset gray-scale difference threshold, and the printing defect area on the surface of the printed metal plate to be detected is identified based on the identification result. The standard image refers to the surface image of the printing template. However, the metal plate itself not only has a high light reflection feature, but also has a phenomenon of uneven light collection, that is, the influence of the collection environment will cause uneven light when collecting the surface image of the printed metal plate. The influence of the collection environment mainly refers to the influence of various environmental factors such as mechanical vibration, and the high light reflection feature of the metal plate itself and the phenomenon of uneven light collection will directly affect the accuracy of subsequent printing defect area identification. That is, the high light reflection feature of the metal plate itself and the phenomenon of uneven light caused by environmental factors during the process of collecting the surface image of the printed metal plate will cause some normal area pixel points to be in an uneven light area or a light interference area, resulting in a large gray-scale difference between the normal area pixel points and the standard image, so that some normal area pixel points are determined as defect area pixel points, and thus the accuracy of surface printing defect area detection of the printed metal plate to be detected is low. Therefore, how to improve the accuracy of surface printing defect area detection of the printed metal plate has become a problem to be solved. SUMMARY

[0004] To solve the above problems, the present application provides a high-efficiency identification and detection method for surface printing defects of a metal plate, and the technical solution is as follows:

[0005] One embodiment of the present application provides a high-efficiency identification and detection method for surface printing defects of a metal plate, which comprises the following steps:

[0006] obtaining a surface image of a printed metal plate to be detected and a standard image;

[0007] According to the difference between the to-be-detected pixel point and the standard pixel point at the same position, a difference representation value of each to-be-detected pixel point is obtained, wherein the to-be-detected pixel point belongs to a surface image, and the standard pixel point belongs to a standard image;

[0008] According to the difference representation value of the to-be-detected pixel point, the gray level co-occurrence matrix eigenvalue of the to-be-detected pixel point and the neighborhood to-be-detected pixel point of the to-be-detected pixel point, a suspected judgment index value of the to-be-detected pixel point is obtained, and a suspected defect pixel point is obtained according to the suspected judgment index value;

[0009] According to the suspected judgment index value and the coordinate value of the suspected defect pixel point, all suspected defect pixel points are clustered to obtain each cluster, and according to the difference representation value of the suspected defect pixel point, the gradient direction of the suspected defect pixel point in the cluster to which the suspected defect pixel point belongs and the standard deviation and mean value of the gray values of all suspected defect pixel points in the cluster, an abnormality degree representation value of the suspected defect pixel point is obtained.

[0010] According to the abnormality degree representation value, a surface of a to-be-detected printed metal plate is identified for printing defects.

[0011] Beneficial effects: The surface image and the standard image of the to-be-detected printed metal plate are obtained first, then according to the difference between the to-be-detected pixel point and the standard pixel point at the same position, a difference representation value of each to-be-detected pixel point is obtained, then according to the difference representation value of the to-be-detected pixel point, the gray level co-occurrence matrix eigenvalue of the to-be-detected pixel point and the neighborhood to-be-detected pixel point of the to-be-detected pixel point, a suspected judgment index value of the to-be-detected pixel point is obtained, and a suspected defect pixel point is obtained according to the suspected judgment index value, then according to the suspected judgment index value and the coordinate value of the suspected defect pixel point, all suspected defect pixel points are clustered to obtain each cluster, and according to the difference representation value of the suspected defect pixel point, the gradient direction of the suspected defect pixel point in the cluster to which the suspected defect pixel point belongs and the standard deviation and mean value of the gray values of all suspected defect pixel points in the cluster, an abnormality degree representation value of the suspected defect pixel point is obtained, and finally according to the abnormality degree representation value, a surface of a to-be-detected printed metal plate is identified for printing defects. And on the basis of the known difference representation value of the to-be-detected pixel point, the abnormality degree representation value obtained by analyzing the characteristics of the interference area and the defect area, that is, the abnormality degree representation value obtained by combining the suspected judgment index value of the to-be-detected pixel point, the gradient direction of the suspected defect pixel point in the cluster to which the suspected defect pixel point belongs and the standard deviation and mean value of the gray values of all suspected defect pixel points in the cluster, can improve the accuracy of identifying the printing defect area on the surface of the to-be-detected printed metal plate. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required by the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0013] Figure 1 A flow chart of a metal plate surface printing defect efficient identification and detection method of the present application. DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art belong to the scope of protection of the embodiments of the present application.

[0015] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0016] The present embodiment provides a metal plate surface printing defect efficient identification and detection method, which is described in detail as follows:

[0017] As shown in Figure 1 The metal plate surface printing defect efficient identification and detection method comprises the following steps:

[0018] Step S001, obtaining the surface image and the standard image of the printing metal plate to be detected.

[0019] The present embodiment mainly improves the accuracy of identifying the printing defect area on the basis of analyzing the difference between the surface image of the printing metal plate to be detected and the standard image, and combining the characteristics of the uneven illumination area and the metal reflection interference area. That is, the present embodiment mainly combines multiple dimensions to improve the detection accuracy of the printing defect area. In addition, the printing metal plate in the present embodiment is mainly applied to the fields of electronic products, automobile parts, household appliances, etc.

[0020] The embodiment first selects any printed metal plate as a to-be-detected printed metal plate in any batch of printed metal plate products, and obtains a surface image of the to-be-detected printed metal plate, wherein the surface image of the to-be-detected printed metal plate is a gray-scale image, and the materials, printing conditions, printing requirements and the like used when printing all the metal plates in the same batch are the same, that is, all the printed metal plate products in the same batch are printed according to the same printing template; the surface image of the to-be-detected printed metal plate in the embodiment is collected by a high-definition camera installed on a metal plate conveying track, that is, the printed metal plate is conveyed on the metal plate conveying track, and the parameters of the high-definition camera when collecting the image need to be set by the implementer according to the actual situation such as the size of the metal plate and the speed of the conveying track.

[0021] Since the embodiment needs to analyze the differences between the surface image of the to-be-detected printed metal plate and the standard image in combination with other dimensions in the subsequent process, the embodiment needs to obtain a standard image of the to-be-detected printed metal plate, and the standard image of the to-be-detected printed metal plate in the embodiment is a surface gray-scale image of a printing template of the to-be-detected printed metal plate; and as another real-time method, a surface gray-scale image of a printed metal plate product that is qualified in appearance quality or does not have any defects but is consistent with the to-be-detected printed metal plate in size, material, printing requirement and the like can also be selected as the standard image of the to-be-detected printed metal plate.

[0022] Therefore, the embodiment can obtain the surface image and the standard image of the to-be-detected printed metal plate through the above process, and record the pixel points on the surface image of the to-be-detected printed metal plate as to-be-detected pixel points and record the pixel points on the standard image as standard pixel points; in addition, the conditions and angles when collecting the surface image and the standard image are consistent, so the pixel points with the same coordinates on the surface image and the standard image belong to the same positions of different printed metal plates.

[0023] In step S002, a difference representation value of each to-be-detected pixel point is obtained according to the differences between the to-be-detected pixel points and the standard pixel points at the same positions.

[0024] Due to the high light reflection characteristics of the metal plate itself and the uneven light caused by various environmental factors such as mechanical vibration in the process of collecting the surface image of the printed metal plate, some normal area pixel points may be in the uneven light area or the light interference area, which will cause the gray value of the normal area pixel point on the printed metal plate to be greatly different from the gray value of the pixel point at the same position on the standard image. This phenomenon will cause the normal area pixel point to be identified as a defect pixel point based on the preset gray difference threshold, resulting in low accuracy of the surface printing defect detection of the printed metal plate to be detected. In order to improve the detection accuracy, the difference between the surface image of the printed metal plate to be detected and the standard image is analyzed, and then the feature difference between the defect area and the interference area is analyzed to exclude the influence of the interference area on the detection, thereby improving the defect detection accuracy. The interference area refers to the uneven light area or the metal light interference area. Therefore, it is known that the embodiment needs to analyze the difference between the surface image of the printed metal plate to be detected and the standard image next, that is, the embodiment needs to obtain the difference representation value of each detected pixel point according to the difference between the detected pixel point and the standard pixel point at the same position. The difference representation value is an important parameter for obtaining the suspected defect pixel point and the abnormality degree representation value. The specific process of obtaining the difference representation value of the detected pixel point is as follows:

[0025] First, take each detected pixel point on the surface image as the center to establish a window with a preset first length of long and wide, and mark it as the first local window corresponding to the detected pixel point. Take each standard pixel point on the standard image as the center to establish a window with a preset first length of long and wide, and mark it as the first local window corresponding to the standard pixel point. In specific applications, the implementer needs to set the preset first length according to the actual situation, but it cannot be too large and must be an odd number. For example, the preset first length can be set to 5 or 7 in this embodiment. If the preset first length is 5, the size of the first local window is 5x5, that is, the maximum number of pixel points that can be accommodated in the first local window is 25.

[0026] After the first local window is obtained, the specific acquisition process of the difference representation value of any to-be-detected pixel point A on the surface image is described, that is, the specific acquisition process of the difference representation value of the to-be-detected pixel point A is as follows: first, on the standard image, a standard pixel point with the same coordinate as the to-be-detected pixel point A is obtained, and is recorded as the same-position standard pixel point of the to-be-detected pixel point A; then, in the first local window of the to-be-detected pixel point A, all gray value types appearing in the first local window of the to-be-detected pixel point A are obtained, and a set constructed by all gray value types appearing in the first local window of the to-be-detected pixel point A is recorded as a gray value type set corresponding to the to-be-detected pixel point A; then, the frequencies of each gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the to-be-detected pixel point A are obtained, and the ratio of the frequency of each gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the to-be-detected pixel point A to the total number of to-be-detected pixels in the first local window of the to-be-detected pixel point A is recorded as the probability of the corresponding gray value type appearing in the first local window of the to-be-detected pixel point A; the frequencies of each gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the same-position standard pixel point of the to-be-detected pixel point A are obtained, and the ratio of the frequency of each gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the same-position standard pixel point of the to-be-detected pixel point A to the total number of standard pixels in the first local window of the same-position standard pixel point of the to-be-detected pixel point A is recorded as the probability of the corresponding gray value type appearing in the first local window of the same-position standard pixel point of the to-be-detected pixel point A, that is, the probability of the a-th gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the to-be-detected pixel point A is the ratio of the frequency of the a-th gray value type appearing in the first local window of the to-be-detected pixel point A to the total number of to-be-detected pixels in the first local window of the to-be-detected pixel point A, and the probability of the a-th gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the same-position standard pixel point of the to-be-detected pixel point A is the ratio of the frequency of the a-th gray value type appearing in the first local window of the same-position standard pixel point of the to-be-detected pixel point A to the total number of standard pixels in the first local window of the same-position standard pixel point of the to-be-detected pixel point A; then, according to the gray difference between the to-be-detected pixel point A and the same-position standard pixel point of the to-be-detected pixel point A and the probabilities of each gray value type in the gray value type set corresponding to the to-be-detected pixel point A appearing in the first local window of the to-be-detected pixel point A and the first local window of the same-position standard pixel point of the to-be-detected pixel point A respectively, the difference representation value of the to-be-detected pixel point A is obtained.

[0027] and according to the gray level difference between the to-be-detected pixel point A and the same-position standard pixel point of the to-be-detected pixel point A and the probabilities of each gray level type in the gray level type set corresponding to the to-be-detected pixel point A to appear in the first local window of the to-be-detected pixel point A and the first local window of the same-position standard pixel point of the to-be-detected pixel point A respectively, a specific process of obtaining the difference representation value of the to-be-detected pixel point A is as follows:

[0028] The absolute value of the gray level value difference between the to-be-detected pixel point A and the same-position standard pixel point of the to-be-detected pixel point A is calculated and recorded as the gray feature difference value of the to-be-detected pixel point A, the probability difference value corresponding to each gray level type in the gray level type set corresponding to the to-be-detected pixel point A is obtained, and the mean value of the probability difference values corresponding to all gray level types in the gray level type set corresponding to the to-be-detected pixel point A is recorded as the probability feature difference value of the to-be-detected pixel point A, the product of the gray feature difference value of the to-be-detected pixel point A and the probability feature difference value of the to-be-detected pixel point A is calculated and recorded as the difference representation value of the to-be-detected pixel point A, and the probability difference value corresponding to the bth gray level type in the gray level type set corresponding to the to-be-detected pixel point A is the absolute value of the difference between the probability of the bth gray level type to appear in the first local window of the to-be-detected pixel point A and the probability of the bth gray level type to appear in the first local window of the same-position standard pixel point of the to-be-detected pixel point A.

[0029] In addition, the calculation expression of the difference representation value of the to-be-detected pixel point A is as follows:

[0030]

[0031] wherein, D A is the difference representation value of the to-be-detected pixel point A, H A is the gray level value of the to-be-detected pixel point A, H' A is the gray level value of the same-position standard pixel point of the to-be-detected pixel point A, B is the total number of gray level types in the gray level type set corresponding to the to-be-detected pixel point A, P A is the probability of the bth gray level type in the gray level type set corresponding to the to-be-detected pixel point A to appear in the first local window of the to-be-detected pixel point A, P' A is the probability of the bth gray level type in the gray level type set corresponding to the to-be-detected pixel point A to appear in the first local window of the same-position standard pixel point of the to-be-detected pixel point A. When |H A -H' A | is larger, it indicates that the difference between the to-be-detected pixel point A and the same-position standard pixel point of the to-be-detected pixel point A is larger, and vice versa when |H A -H' A | is smaller. ​The smaller the difference is, the smaller the difference between the to-be-detected pixel point A and the standard pixel point at the same position of the to-be-detected pixel point A is.

[0032] Therefore, the embodiment can obtain the difference representation value of each to-be-detected pixel point through the above process. Due to the existence of interference factors such as uneven illumination and reflection, there may be pixel points in the normal region in the to-be-detected pixel point with a large difference representation value. Therefore, the embodiment still needs to further analyze in combination with the characteristics of the defect region and the interference region. The interference region is mainly caused by uneven illumination and metal reflection. Therefore, the essence of the interference region refers to the region affected by factors such as uneven illumination and metal reflection. Alternatively, the interference region can also be referred to as an uneven illumination region and a metal reflection interference region.

[0033] In step S003, a suspected judgment index value of the to-be-detected pixel point is obtained according to the difference representation value of the to-be-detected pixel point, the gray level co-occurrence matrix eigenvalue of the to-be-detected pixel point, and the neighborhood to-be-detected pixel point of the to-be-detected pixel point. A suspected defect pixel point is obtained according to the suspected judgment index value.

[0034] Since the size of the difference representation value of the to-be-detected pixel point cannot accurately screen out the defect pixel point, the embodiment will analyze the suspected determination index value of the to-be-detected pixel point next. The suspected determination index value in the embodiment only analyzes the features between uneven illumination and the defect region. Therefore, the embodiment can only perform the initial screening of the normal pixel point and the defect pixel point based on the suspected determination index value. In order to further ensure the identification accuracy of the defect pixel point, the subsequent screening will be combined with the metal reflection interference region. It can be known based on the above analysis that the embodiment will first obtain the suspected determination index value next. It is known that the suspected determination index value is related to the features of the difference representation value, the uneven illumination region and the real defect region. Therefore, when the suspected determination index value is obtained, the features of the uneven illumination region and the real defect region need to be analyzed based on the known difference representation value. The features of the uneven illumination region and the real defect region are that the to-be-detected pixel points in the uneven illumination region are affected by uneven illumination during image acquisition, so that there is a difference between part of the to-be-detected pixel points and the standard pixel points, but the overall texture is not damaged. The local texture features of the real defect region change. The uneven brightness region is the change of the brightness of the whole image or most of the image regions. Therefore, the local features of the pixel points in the uneven brightness region under different window scales are not much different. The real defect is generally a local defect, so the local feature difference of the defect pixel points under different window scales is large. The embodiment will obtain the suspected determination index value of the to-be-detected pixel point next based on the above description. Since the surface of the printed metal plate usually has patterns, the local features will also change. In order to avoid the influence of the patterns on the surface of the printed metal plate on the accurate identification of the defect pixel point in the subsequent analysis, the gray value involved in the analysis of the local feature difference under different window scales will be replaced by the gray feature difference value in the subsequent analysis of the embodiment, so as to avoid the influence of the patterns on the surface of the printed metal plate on the accurate identification of the defect pixel point. The gray feature difference value of any to-be-detected pixel point is the absolute value of the gray value difference between the to-be-detected pixel point and the standard pixel point at the same position of the to-be-detected pixel point.

[0035] First, a window with a preset second length of long and wide is established with the pixel point A to be detected as the center, and is recorded as the second local window of the pixel point A to be detected, then a window constructed by the gray feature difference values of all the pixel points to be detected in the second local window of the pixel point A to be detected is obtained, and is recorded as the second feature window of the pixel point A to be detected, a window constructed by the gray feature difference values of all the pixel points to be detected in the first local window of the pixel point A to be detected is obtained, and is recorded as the first feature window of the pixel point A to be detected, the cth gray feature difference value in the first feature window of the pixel point A to be detected is the gray feature difference value of the cth pixel point to be detected in the first local window of the pixel point A to be detected, and the fth gray feature difference value in the second feature window of the pixel point A to be detected is the gray feature difference value of the fth pixel point to be detected in the second local window of the pixel point A to be detected; then the gray co-occurrence matrix of the pixel point A to be detected is obtained according to the gray values of the pixel points to be detected in the first local window of the pixel point A to be detected, that is, in the process of obtaining the gray co-occurrence matrix of the pixel point A to be detected, the pixel pair is composed of the gray values; then the inverse distance feature value of the gray co-occurrence matrix of the pixel point A to be detected is obtained, the inverse distance feature value can reflect the local texture uniformity, and the process of constructing the gray co-occurrence matrix of the pixel point A to be detected and calculating the inverse distance feature value of the gray co-occurrence matrix under the known window of the pixel point A to be detected is known; then the inverse function is used to map the inverse distance feature value of the gray co-occurrence matrix of the pixel point A to be detected, and the mapping result is recorded as the local texture representation value of the pixel point A to be detected, then the absolute value of the difference between the mean value of all the gray feature difference values in the first feature window of the pixel point A to be detected and the mean value of all the gray feature difference values in the second local window of the pixel point A to be detected is obtained, and is recorded as the window mean difference value of the pixel point A to be detected, then the defect representation value of the pixel point A to be detected is obtained according to the local texture representation value of the pixel point A to be detected and the window mean difference value of the pixel point A to be detected, the defect representation value is the key to subsequently obtain the suspected judgment index value, and the defect representation value is mainly determined by analyzing the features of the uneven illumination area and the real defect area, the defect representation value of the pixel point A to be detected is the product of the local texture representation value of the pixel point A to be detected and the window mean difference value of the pixel point A to be detected; and in specific application, the implementer needs to set the preset second length according to the preset first length and the actual situation, but the preset second length is required to be different from the preset first length, for example, the preset second length can be set to 9 or 15, if 9, then the size of the second local window is 9*9.

[0036] In addition, the specific calculation expression of the defect representation value of the pixel point A to be detected is wherein, W A is the inverse distance feature value of the gray co-occurrence matrix of the pixel point A to be detected, is the result of mapping the inverse disparity eigenvalue of the gray level co-occurrence matrix of the pixel point A to be detected by using the inverse function, and is the local texture feature value of the pixel point A to be detected A is the mean value of all gray level feature difference values in the first feature window of the pixel point A to be detected, and A is the mean value of all gray level feature difference values in the second feature window of the pixel point A to be detected, and A -θ2 A is the window mean difference value of the pixel point A to be detected; and W A is smaller, indicating that the local texture of the pixel point A to be detected is more uneven, and A -θ2 A is larger, indicating that the local feature difference of the pixel point A to be detected under different window scales is larger. Since the defect region is relative to the uneven illumination region, the local feature difference of the defect pixel point in the defect region under different window scales is larger, and the local texture feature of the defect pixel point in the defect region will change, that is, the local texture of the defect pixel point in the defect region is more uneven. Therefore, when W A is smaller and A -θ2 A is larger, it indicates that the pixel point A to be detected belongs to the defect region or the probability of belonging to the defect pixel point is larger, and vice versa.

[0037] After the defect characterization value is obtained, the result obtained by multiplying the defect characterization value of the pixel point A to be detected and the difference degree characterization value of the pixel point A to be detected and then performing normalization processing is recorded as a suspected determination index value of the pixel point A to be detected, and the normalization processing uses a normalization function Norm(), and the suspected determination index value is 0 to 1; and the size of the suspected determination index value can reflect the possibility of the pixel point A to be detected being a defect pixel point, and the suspected determination index value is the basis for primary screening, and the smaller the suspected determination index value, the greater the probability of the pixel point A to be detected being a normal pixel point, and the larger the suspected determination index value, the smaller the probability of the pixel point A to be detected being a normal pixel point, so after the suspected determination index value of the pixel point A to be detected is obtained, whether the pixel point A to be detected is a suspected defect pixel point is determined according to the suspected determination index value of the pixel point A to be detected, and the specific process is: judging whether the suspected determination index value of the pixel point A to be detected is greater than a preset suspected determination threshold, if yes, the pixel point A to be detected is recorded as a suspected defect pixel point, otherwise, the pixel point A to be detected is recorded as a normal pixel point. In addition, in specific applications, the implementer needs to set the preset suspected determination threshold according to the value range of the suspected determination index value, experimental statistics and actual conditions, etc., for example, the suspected determination index value can be set to 0.7 in this embodiment.

[0038] Therefore, the suspected defect pixel points on the surface image are obtained through the above process in this embodiment.

[0039] In step S004, all suspected defect pixel points are clustered according to the suspected determination index value and the coordinate value of the suspected defect pixel points, and each clustering cluster is obtained, and the abnormal degree characterization value of the suspected defect pixel point is obtained according to the difference characterization value of the suspected defect pixel point, the gradient direction of the suspected defect pixel point in the clustering cluster to which the suspected defect pixel point belongs, and the standard deviation and mean value of the gray values of all suspected defect pixel points in the clustering cluster to which the suspected defect pixel point belongs.

[0040] Since the above-mentioned suspected defective pixel point screening only analyzes the features between uneven illumination and the defect area, the embodiment based on the suspected determination index value can only perform the initial screening, that is, there may be normal pixels caused by metal reflection in the suspected defective pixel points, so the embodiment needs to combine the features of the real defect area and the metal reflection interference area to obtain the defect confidence of the suspected defective pixel point, and the features of the real defect area and the metal reflection interference area are that the real defect area has lower regional gray uniformity and lower overall brightness uniformity, has no significant highlight feature, and has high local complexity, the metal reflection interference area changes the local texture, but the gray uniformity in the area is high, the overall brightness of the metal reflection interference area is high, and the area has a local highlight feature and low complexity, and in addition, the real defect area and the metal reflection interference area will damage the local texture of the image; therefore, based on the above analysis, the embodiment will analyze the gray distribution uniformity, brightness, and complexity of the area to which the suspected defective pixel point belongs to obtain the defect confidence of each suspected defective pixel point, the defect confidence can not only further reflect the possibility that the suspected defective pixel point is a real defective pixel point, but also is an important parameter for obtaining the abnormal degree representation value of the suspected defective pixel point subsequently, that is, in the embodiment, the defect confidence of the suspected defective pixel point needs to be obtained before the abnormal degree representation value of the suspected defective pixel point is obtained, and the suspected defective pixel point needs to be clustered before the defect confidence of the suspected defective pixel point is obtained, so the specific process of clustering all suspected defective pixel points on the surface image is as follows:

[0041] First, the vector composed of the suspected determination index value, the horizontal coordinate value, and the vertical coordinate value of each suspected defective pixel point is recorded as the feature vector of the corresponding suspected defective pixel point, and the first parameter in the feature vector of any suspected defective pixel point is the suspected determination index value of the suspected defective pixel point, the second parameter is the horizontal coordinate value of the suspected defective pixel point, and the third parameter is the vertical coordinate value of the suspected defective pixel point, then based on the known feature vector of the suspected defective pixel point, the mean shift clustering algorithm is used to cluster all suspected defective pixel points on the surface image to obtain each cluster, and the distance between any two suspected defective pixel points involved in clustering all suspected defective pixel points on the surface image by using the mean shift clustering algorithm is the Euclidean distance between the feature vectors of the two suspected defective pixel points; in addition, the specific process of clustering by using the mean shift clustering algorithm is a known technology, and therefore will not be described in detail.

[0042] After obtaining the cluster, the specific process of obtaining the defect confidence of any suspected defective pixel point S is described based on the cluster to which the suspected defective pixel point belongs, that is, the specific process of obtaining the defect confidence of the suspected defective pixel point S is as follows:

[0043] First, the cluster to which the suspected defective pixel S belongs is obtained and denoted as the cluster to be analyzed. Then, the gradient direction of each suspected defective pixel in the cluster to be analyzed is obtained, and the information entropy of the gradient directions of all suspected defective pixels in the cluster to be analyzed is calculated and denoted as the complex characterization value of the grayscale distribution of the suspected defective pixel S. The processes for obtaining the gradient direction and information entropy of the pixels are known techniques. Next, the mean grayscale value of all suspected defective pixels in the cluster to be analyzed is obtained and denoted as the grayscale mean of the cluster to be analyzed. The standard deviation of the grayscale values ​​of all suspected defective pixels in the cluster to be analyzed is also obtained and denoted as the standard deviation of the grayscale values ​​of the cluster to be analyzed. The standard deviation of the clusters is analyzed, and the mean gray value of all pixels to be detected on the surface image is obtained and recorded as the global gray mean. Next, the ratio of the gray mean of the cluster to be analyzed to the global gray mean is obtained and recorded as the feature ratio. Then, the feature ratio is mapped using an inverse function to obtain the regional brightness characterization value of the suspected defective pixel S. Finally, the product of the standard deviation of the clusters to be analyzed, the regional brightness characterization value of the suspected defective pixel S, and the complex gray distribution characterization value of the suspected defective pixel S is obtained and used as the defect confidence score of the suspected defective pixel S. The specific calculation expression for the defect confidence score of the suspected defective pixel S is as follows:

[0044]

[0045] Where Z represents the defect confidence score of the suspected defective pixel S, σ represents the standard deviation of the cluster to be analyzed, and G represents the feature ratio. H represents the brightness characterization value of the region of the suspected defective pixel S. S H represents the complex grayscale distribution of the suspected defective pixel S, and is also the information entropy of the gradient direction of all suspected defective pixels in the cluster to be analyzed. A larger σ indicates lower grayscale uniformity within the region corresponding to the cluster to be analyzed, and a smaller feature ratio G indicates that the brightness of the region corresponding to the cluster to be analyzed is darker relative to the overall brightness of the surface image, meaning that the region corresponding to the cluster to be analyzed has fewer significant highlight features. S The larger the value of H, the higher the gray-level distribution complexity within the region corresponding to the cluster to be analyzed. Furthermore, since the actual defect region has lower gray-level uniformity and overall brightness uniformity compared to the metal reflection interference region, and lacks significant highlight features, and has higher local complexity, a larger value of H and a smaller value of G indicates a higher degree of complexity. S The larger Z is, the greater the probability that the suspected defective pixel S is actually a defective pixel. Conversely, the smaller Z is, the greater the probability that the suspected defective pixel S is actually a normal pixel.

[0046] Since the defect is a regional feature rather than a single-point feature, the local continuity of the pixel points needs to be combined to obtain the real abnormality degree, that is, on the basis of the defect confidence and the difference representation value of the suspected defect pixel point S, the local continuity of the suspected defect pixel point needs to be combined to determine the abnormality degree representation value of the suspected defect pixel point S, and then the specific acquisition process of the abnormality degree representation value of the suspected defect pixel point S is as follows:

[0047] First, all the to-be-detected pixel points in the eight-neighborhood of the suspected defect pixel point S are obtained, and the set of all the to-be-detected pixel points in the eight-neighborhood of the suspected defect pixel point S is denoted as the neighborhood set of the suspected defect pixel point S. Then, the mark value of each detection pixel point in the neighborhood set is obtained, and the specific acquisition process of the mark value of each detection pixel point in the neighborhood set is as follows: for the jth to-be-detected pixel point in the neighborhood set, if the jth to-be-detected pixel point is a suspected defect pixel point, the difference representation value of the jth to-be-detected pixel point is directly taken as the mark value of the jth to-be-detected pixel point, and if the jth detection pixel point is not a suspected defect pixel point, a preset constant is directly taken as the mark value of the jth detection pixel point. In specific applications, the implementer needs to set the preset constant according to the actual situation, for example, the preset constant is set to 0 in this embodiment.

[0048] Then, the mean value of the label values of all the to-be-detected pixel points in the neighborhood set is obtained and is denoted as a neighborhood label value mean, and the standard deviation of the label values of all the to-be-detected pixel points in the neighborhood set is obtained and is denoted as a neighborhood label value standard deviation. Then, the ratio of the neighborhood label value mean to the neighborhood label value standard deviation is obtained and is denoted as a local continuity representation value of the suspected defect pixel point S. Finally, the product of the local continuity representation value of the suspected defect pixel point S, the difference representation value, and the defect confidence is obtained, and the product result is normalized. The normalized result is denoted as an abnormality degree representation value of the suspected defect pixel point S. Here, the normalization function Norm() is used to normalize the product result. Since the larger the local continuity representation value is, the better the continuity of the difference representation value in the neighborhood of the suspected defect pixel point S is, that is, the better the continuity of the local difference representation value of the suspected defect pixel point S is, and when the continuity of the local difference representation value is better, the probability that the suspected defect pixel point S is located in the defect region is larger, and the probability that the suspected defect pixel point S is a real defect pixel point is larger, the larger the local continuity representation value is, the larger the abnormality degree representation value of the suspected defect pixel point S should be. Therefore, the larger the local continuity representation value of the suspected defect pixel point S, the difference representation value, and the defect confidence are, the larger the abnormality degree representation value of the suspected defect pixel point S is. When the abnormality degree representation value of the suspected defect pixel point S is larger, the probability that the suspected defect pixel point S is a real defect pixel point is larger, and vice versa.

[0049] Therefore, the abnormality degree representation value of each suspected defect pixel point can be obtained through the above process, and the value of the abnormality degree representation value is 0 to 1.

[0050] In step S005, the surface of the to-be-detected printed metal plate is printed defect recognized according to the abnormality degree representation value.

[0051] After the abnormality degree representation value of each suspected defect pixel point on the surface image is obtained, it is determined whether the abnormality degree representation value of each suspected defect pixel point is greater than a preset abnormality threshold. If yes, the corresponding suspected defect pixel point is determined to be a real defect pixel point, and then the corresponding suspected defect pixel point is recorded as a defect pixel point. In a specific application, the implementer needs to set the preset abnormality threshold according to the value range of the abnormality degree representation value, experimental statistics, and other actual situations. For example, the preset abnormality threshold can be set to 0.7 in this embodiment.

[0052] After all the defect pixel points on the surface image are obtained, all the printing defect regions on the surface image are obtained by using the existing algorithm, and after the printing defect regions are obtained, the types of the printing defect regions can be obtained by using the classification model. In addition, the printing defect regions can be obtained by using the region growing algorithm and morphological processing, that is, the defect pixel points on the surface image are subjected to region growing, and the obtained regions are subjected to closing operation processing to obtain the printing defect regions. The process of obtaining the printing defect regions on the surface image by using the existing algorithm is a known technology, and therefore the embodiment will not be described in detail.

[0053] Thus, the detection of the printing defects on the surface of the metal plate is completed, and based on the difference representation value of the to-be-detected pixel point, the characteristics of the interference region and the defect region are analyzed, that is, the accurate identification of the defect pixel points on the surface image of the to-be-detected printing metal plate is completed by combining the suspected judgment index value of the to-be-detected pixel point, the gradient direction of the suspected defect pixel point in the clustering cluster to which the suspected defect pixel point belongs, and the standard deviation and mean value of the gray values of all the suspected defect pixel points in the clustering cluster, so as to achieve the accurate identification of the defect region on the surface image of the to-be-detected printing metal plate.

[0054] In summary, the surface image and the standard image of the to-be-detected printing metal plate are first obtained, then the difference representation value of each to-be-detected pixel point is obtained according to the difference between the to-be-detected pixel point and the standard pixel point at the same position, then the suspected judgment index value of the to-be-detected pixel point is obtained according to the difference representation value of the to-be-detected pixel point, the eigenvalue of the gray co-occurrence matrix of the to-be-detected pixel point, and the neighborhood to-be-detected pixel point of the to-be-detected pixel point, and the suspected defect pixel point is obtained according to the suspected judgment index value, then all the suspected defect pixel points are clustered according to the suspected judgment index value and the coordinate value of the suspected defect pixel point, and each clustering cluster is obtained, and the abnormality degree representation value of the suspected defect pixel point is obtained according to the difference representation value of the suspected defect pixel point, the gradient direction of the suspected defect pixel point in the clustering cluster to which the suspected defect pixel point belongs, and the standard deviation and mean value of the gray values of all the suspected defect pixel points in the clustering cluster, and finally the printing defect on the surface of the to-be-detected printing metal plate is identified according to the abnormality degree representation value. Based on the difference representation value of the to-be-detected pixel point, the characteristics of the interference region and the defect region are analyzed, that is, the abnormality degree representation value obtained by combining the suspected judgment index value of the to-be-detected pixel point, the gradient direction of the suspected defect pixel point in the clustering cluster to which the suspected defect pixel point belongs, and the standard deviation and mean value of the gray values of all the suspected defect pixel points in the clustering cluster, which can improve the accuracy of the identification of the printing defect region on the surface of the to-be-detected printing metal plate.

[0055] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for efficient identification and detection of printing defects on the surface of metal plates, characterized in that, The method includes the following steps: Acquire surface images and standard images of the printed metal plate to be inspected; Based on the difference between the pixel to be detected and the standard pixel at the same location, the difference characterization value of each pixel to be detected is obtained. The pixel to be detected belongs to the surface image, and the standard pixel belongs to the standard image. Based on the difference characterization value of the pixel to be detected, the gray-level co-occurrence matrix feature value of the pixel to be detected, and the neighboring pixels to be detected, the suspected judgment index value of the pixel to be detected is obtained, and the suspected defective pixel is obtained based on the suspected judgment index value. Based on the suspected defective pixel's suspected judgment index value and coordinate value, all suspected defective pixels are clustered to obtain each cluster. Based on the difference characterization value of the suspected defective pixel, the gradient direction of the suspected defective pixel in the cluster to which the suspected defective pixel belongs, and the standard deviation and mean of the gray value of all suspected defective pixels in the cluster, the abnormality characterization value of the suspected defective pixel is obtained. Based on the anomaly degree characterization value, printing defects are identified on the surface of the printed metal plate to be inspected; Methods for obtaining the difference representation values ​​of the pixels to be detected include: Centered on the pixel to be detected, a window with a length and width of a preset first length is established and recorded as the first local window corresponding to the pixel to be detected. Centered on the standard pixel, a window with a length and width of a preset first length is established and recorded as the first local window corresponding to the standard pixel. For any pixel to be detected on the surface image, the standard pixel on the standard image with the same coordinates as the pixel to be detected is recorded as the standard pixel at the same position as the pixel to be detected. A set of all grayscale value types appearing in the first local window of the pixel to be detected is obtained and recorded as the grayscale value type set corresponding to the pixel to be detected. The absolute value of the grayscale value difference between the pixel to be detected and the standard pixel at the same position is recorded as the grayscale feature difference of the pixel to be detected. Each grayscale value type in the grayscale value type set is then obtained. The corresponding probability difference is used, and the mean of the probability differences corresponding to all gray value types in the neighborhood gray value set is recorded as the probability feature difference of the pixel to be detected. The product of the gray value feature difference and the probability feature difference of the pixel to be detected is recorded as the difference characterization value of the pixel to be detected. The probability difference corresponding to the b-th gray value type in the gray value type set is the absolute value of the difference between the probability of the b-th gray value type appearing in the first local window of the pixel to be detected and the probability of the b-th gray value type appearing in the first local window of the standard pixel at the same position. The expression for calculating the difference representation value of the pixel A to be detected is: ； in, The difference representation value of the pixel A to be detected. Let A be the grayscale value of the pixel to be detected. Let B be the grayscale value of the standard pixel at the same position as pixel A to be detected, and let B be the total number of grayscale value types in the set of grayscale value types corresponding to pixel A to be detected. Let be the probability that the b-th grayscale value type in the set of grayscale value types corresponding to the pixel A to be detected appears in the first local window of the pixel A to be detected. Let b be the probability that the b-th gray value type in the set of gray value types corresponding to the pixel A to be detected appears in the first local window of the standard pixel at the same position as the pixel A to be detected; Methods for obtaining the suspected judgment index value of the pixel to be detected include: For any pixel to be detected on the surface image: A window with a preset second length and width is established centered on the pixel to be detected, and denoted as the second local window of the pixel to be detected. A window constructed from the grayscale feature differences of all pixels to be detected within the second local window of the pixel to be detected is denoted as the second feature window of the corresponding pixel to be detected. A window constructed from the grayscale feature differences of all pixels to be detected within the first local window of the pixel to be detected is denoted as the first feature window of the corresponding pixel to be detected. Based on the pixel to be detected within the first local window of the pixel to be detected... The grayscale value of a pixel is used to obtain the grayscale co-occurrence matrix of the corresponding pixel to be detected. Based on the inverse difference feature value of the grayscale co-occurrence matrix of the pixel to be detected and the difference between the mean of all grayscale feature differences in the first feature window of the pixel to be detected and the mean of all grayscale feature differences in the second feature window of the pixel to be detected, the defect characterization value of the pixel to be detected is obtained. The result of normalizing the product of the defect characterization value and the difference degree characterization value of the pixel to be detected is recorded as the suspected judgment index value of the pixel to be detected.

2. The method for efficient identification and detection of printing defects on the surface of metal plates as described in claim 1, characterized in that, The method for determining the defect characterization value of the pixel to be detected includes: The inverse difference feature value of the gray-level co-occurrence matrix of the pixel to be detected is mapped by an inverse function and recorded as the local texture representation value; the absolute value of the difference between the mean of all gray-level feature differences in the first feature window of the pixel to be detected and the mean of all gray-level feature differences in the second feature window of the pixel to be detected is recorded as the window mean difference value; the product of the local texture representation value and the window mean difference value is used as the defect representation value of the pixel to be detected.

3. The method for efficient identification and detection of printing defects on the surface of metal plates as described in claim 1, characterized in that, Methods for obtaining suspected defective pixels include: If the suspected defect index value of the pixel to be detected is greater than the preset suspected defect threshold, then the pixel to be detected is recorded as a suspected defect pixel.

4. The efficient identification and detection method for printing defects on the surface of a metal plate as described in claim 1, characterized in that, Methods for obtaining the anomaly characterization values ​​of suspected defective pixels include: For any suspected defective pixel: obtain the cluster to which the suspected defective pixel belongs and denote it as the cluster to be analyzed; obtain the gradient direction of each suspected defective pixel in the cluster to be analyzed; calculate the information entropy of the gradient direction of all suspected defective pixels in the cluster to be analyzed and denote it as the gray-level distribution complex characterization value; denote the mean of the gray-level values ​​of all suspected defective pixels in the cluster to be analyzed as the gray-level mean of the cluster to be analyzed; denote the standard deviation of the gray-level values ​​of all suspected defective pixels in the cluster to be analyzed as the standard deviation of the cluster to be analyzed; denote the mean of the gray-level values ​​of all pixels on the surface image as the global gray-level mean; and obtain the defect confidence of the suspected defective pixel based on the gray-level mean and standard deviation of the cluster to be analyzed, the global gray-level mean, and the gray-level distribution complex characterization value. Obtain a set of all pixels to be detected within the eight neighborhoods of the suspected defective pixel, and denote it as the neighborhood set of the suspected defective pixel. Obtain the label value of each pixel to be detected in the neighborhood set. Denote the mean and standard deviation of the label values ​​of all pixels to be detected in the neighborhood set as the mean and standard deviation of the neighborhood label value, respectively. Denote the result of normalizing the product of the ratio of the mean and standard deviation of the neighborhood label value, the difference characterization value of the suspected defective pixel, and the defect confidence of the suspected defective pixel as the abnormality characterization value of the suspected defective pixel.

5. The efficient identification and detection method for printing defects on the surface of a metal plate as described in claim 4, characterized in that, Methods for determining the defect confidence of suspected defective pixels include: The ratio of the gray-level mean of the cluster to be analyzed to the global gray-level mean is mapped by an inverse function and recorded as the regional brightness characterization value; the product of the standard deviation of the cluster to be analyzed, the regional brightness characterization value, and the gray-level distribution complexity characterization value is used as the defect confidence of the suspected defective pixel.

6. The efficient identification and detection method for printing defects on the surface of a metal plate as described in claim 4, characterized in that, Methods for obtaining the tag value include: For the j-th pixel to be detected in the neighborhood set, if the j-th pixel to be detected is a suspected defective pixel, then the difference characterization value of the j-th pixel to be detected is directly used as the label value of the j-th pixel to be detected; if the j-th pixel to be detected is not a suspected defective pixel, then a preset constant is used as the label value of the j-th pixel to be detected.

7. The efficient identification and detection method for printing defects on the surface of a metal plate as described in claim 1, characterized in that, A method for identifying printing defects on the surface of a printed metal plate to be inspected based on the aforementioned anomaly severity characterization value includes: All suspected defective pixels with an anomaly severity value greater than a preset anomaly threshold are recorded as defective pixels, and all printing defect areas on the surface image of the printed metal plate to be inspected are obtained based on the defective pixels on the surface image.

Images