A method of image circle hole counting and classification

Abstract

The present application belongs to the technical field of image processing, and specifically relates to an image circle hole counting and classification method, which comprises the following steps: collecting images by using an image collection device; cropping the collected images; performing denoising processing according to the features of the collected images; performing threshold processing according to the features of the collected images; performing contour extraction on the images after threshold processing, and filling the contours, each contour being composed of pixel points, and calculating the first moment of each extracted contour; removing abnormal contours that do not meet the requirements; dividing large holes and small holes, and calculating the number of large holes and small holes. Compared with the method of distinguishing abnormal points by calculating variance and mean, the present method distinguishes abnormal points by using quartiles, does not need to determine the number of abnormal points in advance and then find abnormal points point by point, reduces the randomness of preset values, and is faster and more accurate.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to an image round hole counting and classification method. BACKGROUND

[0002] In the process of manufacturing an airplane, due to the process requirement such as riveting, a round hole needs to be left on the surface of a part during manufacturing, and there are two types of large holes and small holes. In order to ensure the riveting quality and reduce the rework cycle, the round holes on the surface of the part need to be counted and classified. In the prior art, a kind of round object detection method and system is disclosed in Chinese patent application No. CN202111570608.4, which comprises the following steps: S1, putting the edge points of a round to be detected into an edge point set D, and dividing the edge point set D into at least four equal parts according to an angle, and each equal part corresponds to a region; S2, randomly selecting three fitting points d1, d2 and d3 from the three regions, detecting the parameters of a candidate circle based on the three fitting points d1, d2 and d3; S3, randomly selecting a verification point d4, and substituting the verification point d4 into the candidate circle P1; S4, if the verification point is located within the allowable deviation range of the candidate circle, then the count score of the candidate circle P1 is increased by 1, the verification number k2=k2+1, and the step S3 is returned, and when the verification number reaches the maximum cycle number k2max, if the count score of the candidate circle P1 reaches the set threshold Nt, then the candidate circle P1 is a true circle.

[0003] The focus of the comparison file is on the recognition and fitting of a circle, and the fitting of a true circle depends on the imaging of a picture and the accuracy of a fitting algorithm. If the circle is stretched and deformed during image shooting, the true circle cannot be accurately found according to the file. The comparison file has the following differences and deficiencies:

[0004] (1) The application scenario of the comparison file is strongly dependent on the image acquisition level. Once the image is stretched due to the change of the shooting angle, the circle cannot be accurately fitted according to three points, and the true circle cannot be recognized.

[0005] (2) The comparison file focuses on the fitting of a circle and does not involve image acquisition, denoising, feature extraction and classification and other image pre-processing and post-processing. The entire process from image acquisition to feature recognition and classification cannot be completed end to end, and the problem faced by the comparison file cannot be solved by the method of the comparison file. SUMMARY

[0006] To solve the above problems, the application provides an image round hole counting and classification method for counting and classifying the round holes on the surface of a part.

[0007] In order to achieve the above-mentioned application purposes, the technical scheme provided by the present application is as follows:

[0008] A method for counting and classifying circular holes in an image, comprising the following steps:

[0009] S1. Use image acquisition equipment to collect images;

[0010] S2. Crop the collected image;

[0011] S3. Perform denoising based on the collected image features;

[0012] S4. threshold processing based on the collected image features;

[0013] S5. Extract the contours of the image after threshold processing and fill the contours. Each contour is composed of the pixels where it is located. Assuming that an image has The outline, The outline is ,in For each pixel's coordinates, is the number of all pixels of the contour;

[0014] S6. Calculate the first-order moment of each extracted contour :

[0015]

[0016] in Pixel When calculating the pixel value, all non-zero pixel values ​​are recorded as 1;

[0017] S7. Remove abnormal contours that do not meet the requirements;

[0018] S8. Divide the holes into large and small holes, and calculate the number of large and small holes.

[0019] Furthermore, the S1 specifically includes:

[0020] S11. Collection Preparation: For each collection, maintain a fixed light source position and intensity. Ensure a clear contrast between the background and the object being collected, and keep the background clean and free of unwanted objects. Place the object flat on the table, with the camera parallel to the object.

[0021] S12. Collection process: During collection, the entire surface of the object to be collected is collected at one time. If there are circular hole information on multiple surfaces of the object to be collected, each surface is collected to ensure that only one surface is collected each time.

[0022] Furthermore, the S2 specifically includes:

[0023] The image is read on the computer as a grayscale image. If it is a color image, it is converted to a grayscale image before cropping;

[0024] S21. If there are more than one background, then cropping is performed, and arbitrary shape cropping method is adopted, only the object to be collected is kept, and the redundant complex background is removed;

[0025] S22. If the background has only one color and the contrast with the color of the object to be collected is large, then the background is kept, and no cropping operation is needed.

[0026] Further, the S3 specifically comprises:

[0027] S31. If the noise in the image is similar to snowflake point scattered distribution or a small circular hole with a diameter less than one tenth of the minimum circular hole diameter appears, then pepper and salt denoising is performed, the pixel value is sorted by selecting the kernel size, the middle value is selected as the pixel value of the center pixel, and the kernel size is determined according to the maximum size of the minimum circular hole without blurring.

[0028] S32. If the image has no other impurities except the single color background and the circular hole, then Gaussian filtering is performed, the kernel size is selected, the pixel value of the middle value is calculated according to the two-dimensional Gaussian function, and the kernel size is determined according to the maximum size of the minimum circular hole without blurring.

[0029] Further, the S4 specifically comprises:

[0030] S41. If the background is single, the adaptive threshold algorithm is adopted to select the threshold by maximizing the inter-class variance.

[0031] S42. If the background is not single, the adaptive threshold algorithm is adjusted, and the threshold is floated up and down by 40 values. First, a coarse selection step of 10 is performed from the optimal threshold center, a coarse threshold is selected, then a fine selection step of 1 is performed, and the optimal threshold is selected to accurately divide the circular hole and the background.

[0032] S43: If the background is not single, the adaptive threshold algorithm is adopted, the image background boundary gray value is consistent with the circular hole gray value, the gray distribution graph is drawn, the 3 to 4 points with the largest change in the image valley are selected by the gray distribution graph, and the threshold segmentation is performed.

[0033] Further, the S7 specifically comprises:

[0034] S71. The contour with the first moment of 0 is deleted;

[0035] S72. The contour rectangle is drawn, each contour is composed of points, and the longest distance of the contour on the x-axis is the width of the rectangle, and the longest distance of the contour on the y-axis is the height of the rectangle. ​

[0036] ,

[0037] wherein , is the abscissa of the i-th point constituting the profile, , is the ordinate of the i-th point constituting the profile;

[0038] S73. Calculate the aspect ratio of each profile :

[0039]

[0040] S74. Draw a box plot of the aspect ratio and check for outliers;

[0041] S75. Sort the aspect ratio from small to large, and extract the upper quartile point of the sorted sequence, i.e. the number at the sequence position , and the position of the number is ; extract the lower quartile point of the sorted sequence, i.e. the number at the sequence position , and the position of the number is , and calculate the interquartile range , and the upper and lower limits of the outliers are: wherein k is a selected threshold value;

[0042]

[0043] wherein k is a selected threshold value;

[0044] S76: Remove the outlier profiles outside the upper and lower limits.

[0045] Further, the S8 specifically comprises:

[0046] S81. Recalculate the first moment of the profiles after removing the outliers, and sort the first moment sequence , and L is the number of all profiles after removing the outliers.

[0047] S82. Calculate the first difference :

[0048]

[0049] wherein the first difference is supplemented

[0050] S83. Extract the difference sequence ​​​​​The subscript of the maximum value in .

[0051]

[0052] S84. The number of small holes is , the number of large holes is .

[0053] The beneficial effects of the present invention are:

[0054] 1. This method uses quartiles to identify outliers. Compared with calculating variance and mean to identify outliers, it does not need to determine the number of outliers in advance and then search for outliers point by point, which reduces the randomness of the preset value and is faster and more accurate.

[0055] 2. This method selects the first-order moment and aspect ratio as features for post-processing. Compared with deep learning, it does not require a lot of work such as collecting thousands of images, labeling images, and training models, and is more efficient and faster.

[0056] 3. This method focuses on the extraction of circles and the distinction between large and small circles. For images with a single circular feature, image features are extracted through standard shooting, image denoising, threshold processing and other methods. Then, outliers are removed and large and small circles are distinguished based on statistical methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is a flow chart of the present invention.

[0058] Figure 2 This is the black and white image after threshold processing.

[0059] Figure 3 Box plot of the first moment.

[0060] Figure 4 A line chart of first-order differences. DETAILED DESCRIPTION

[0061] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are intended to explain the present invention rather than to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0062] The specific implementation method of the present invention is described below with reference to the accompanying drawings and examples, but the present invention is not limited to this embodiment.

[0063] Example 1

[0064] like Figure 1As shown, a method for counting and classifying circular holes in an image includes the following steps:

[0065] S1. Use a camera or other image acquisition device to collect images;

[0066] S2. Crop the collected image;

[0067] S3. Perform denoising based on the collected image features;

[0068] S4. threshold processing based on the collected image features;

[0069] S5. Extract the contours of the image after threshold processing and fill the contours. Each contour is composed of the pixels where it is located. Assuming that an image has The outline, The outline is ,in For each pixel's coordinates, is the number of all pixels of the contour;

[0070] S6. Calculate the first-order moment of each extracted contour :

[0071]

[0072] in Pixel When calculating the pixel value, all non-zero pixel values ​​are recorded as 1;

[0073] S7. Remove abnormal contours that do not meet the requirements;

[0074] S8. Divide the holes into large and small holes, and calculate the number of large and small holes.

[0075] Example 2

[0076] like Figure 1 As shown, a method for counting and classifying circular holes in an image includes the following steps:

[0077] S1. Use a camera or other image acquisition device to collect images;

[0078] S2. Crop the collected image;

[0079] S3. Perform denoising based on the collected image features;

[0080] S4. threshold processing based on the collected image features;

[0081] S5. Extract the contours of the image after threshold processing and fill the contours. Each contour is composed of the pixels where it is located. Assuming that an image has The outline, The contour is wherein is the coordinate of each pixel point, is the number of all pixel points of the contour;

[0082] S6. Calculate the first moment of each extracted contour :

[0083]

[0084] wherein is the pixel value of the pixel point, and all non-zero pixel values are recorded as value 1 during calculation;

[0085] S7. Remove abnormal contours that do not meet the requirements;

[0086] S8. Divide the large holes and the small holes, and calculate the number of the large holes and the small holes.

[0087] S1 specifically includes:

[0088] S11. Collection preparation: fix the light source position and the illumination intensity during each collection, and the background and the color of the object to be collected have obvious contrast, such as silver-white object to be collected and black background, black object to be collected and white background, etc., and the background is kept clean and no extra objects appear, the object to be collected is placed flat on the table top during collection, and the camera is parallel to the object to be collected;

[0089] S12. Collection process: pay attention to collect the complete surface of the object to be collected at one time, if the object to be collected has circular hole information on multiple surfaces, collect each surface, and ensure to collect only one surface at a time.

[0090] S2 specifically includes:

[0091] The image is read in the form of a gray-scale image on the computer, and if it is a color image, it is converted into a gray-scale image before being cropped;

[0092] S21. If the background is relatively complex, such as more than one color, cropping is performed, and an arbitrary shape cropping method is adopted during cropping to only keep the object to be collected and remove the redundant complex background;

[0093] S22. If the background is single, i.e., only one color and has large contrast with the color of the object to be collected, such as black and white, the background is kept and no cropping operation is needed.

[0094] S3 specifically includes:

[0095] ​S31. If the noise in the image is similar to snowflakes scattered or there are small circular holes with a diameter less than one-tenth of the minimum circular hole diameter, then pepper-salt denoising is performed. The size of the pixels in the selected kernel is sorted by pixel value, and the middle value is selected as the pixel value of the center pixel. The kernel size is determined by the maximum size of the smallest circular hole without blurring;

[0096] S32. If the image has no other impurities except the single-color background and circular hole, then Gaussian filtering is performed. The kernel size is selected, and the pixel value of the middle value is calculated according to the two-dimensional Gaussian function. The kernel size is determined by the maximum size of the smallest circular hole without blurring.

[0097] S4 specifically includes:

[0098] S41. If the background is single, i.e., there is only one color, an adaptive threshold algorithm is used to select the threshold by maximizing the inter-class variance.

[0099] S42. If the background is not single, after using the adaptive threshold algorithm, adjust the threshold value by floating up and down by 40 values. First, select a coarse threshold value with a step size of 10 from the optimal threshold center, then select a fine threshold value with a step size of 1, and select the optimal threshold value to accurately separate the circular hole from the background.

[0100] S43: If the background is not single, the adaptive threshold algorithm is used to draw a gray scale distribution graph. Select 3 to 4 points with the largest change in the image valley through the gray scale distribution graph, and perform threshold segmentation.

[0101] S7 specifically includes:

[0102] S71. Delete the contour with a first moment of 0;

[0103] S72. Draw a contour rectangle. Assuming that each contour is composed of points, the longest distance of the contour on the x-axis is the width of the rectangle, and the longest distance on the y-axis is the height of the rectangle:

[0104] ,

[0105] where , are the horizontal coordinates of the points constituting the contour, , are the vertical coordinates of the points constituting the contour;

[0106] ​S73. Calculate the aspect ratio of each contour :

[0107]

[0108] S74. Aspect Ratio Draw a box plot to view abnormal points;

[0109] S75. Sort from small to large and extract the upper quartile of the sorted sequence , that is, the sequence position Numbers, The location is ; Extract the lower quartile of the sorted sequence , that is, the sequence position Numbers, The location is , calculate the interquartile range , the upper and lower limits of distinguishing abnormal points are:

[0110]

[0111] Where k is the selected threshold. The larger k is, the fewer points are classified as outliers, and the smaller k is, the more points are classified as outliers. The default setting of k for the box plot is 1.5 to remove very extreme abnormal contours and select a threshold of about 10.

[0112] S76: Remove abnormal contours outside the upper and lower limits.

[0113] S8 specifically includes:

[0114] S81. Recalculate the first-order moment after removing the abnormal contour and sort it to obtain the sorted first-order moment sequence , L is the number of all contours after removing abnormal contours, that is, the number of all holes.

[0115] S82. Calculate first-order differences :

[0116]

[0117] Among them, the supplement

[0118] S83. Extract differential sequence The subscript of the maximum value in .

[0119]

[0120] S84. The number of small holes is , the number of large holes is .

[0121] Example 3

[0122] Based on Example 1 and Example 2, a method for counting and classifying circular holes in an image includes the following steps:

[0123] S1: Use image acquisition equipment such as cameras to acquire images.

[0124] S11: Collection Preparation. Place the part on a dark tabletop, clear of any extraneous items. Use a 9-megapixel camera, parallel to the surface of the part being captured. Use overhead lighting to ensure even light distribution across the part, eliminating any reflections.

[0125] S12: Collection process: The entire surface of the part to be processed is collected in one go.

[0126] S2: Crop the captured image. Read the image in grayscale format and crop the effective part surface into a rectangular shape.

[0127] S3: Perform denoising based on the acquired image features. Select salt and pepper denoising and set the kernel size to 11.

[0128] S4: Threshold processing based on the collected image features.

[0129] S41: Adopting the adaptive threshold algorithm, the threshold is 89.

[0130] S42: The adaptive threshold (89) is floated up and down by 40 values. First, a rough selection step of 10 is performed from the center of the optimal threshold to select a rough threshold (69). Then, a fine selection step of 1 is performed to select the optimal threshold 70 to accurately separate the circular hole from the background. Figure 2 shown.

[0131] S5: Perform contour extraction on the image after threshold processing and fill the contours. At this time, there are 108 contours.

[0132] S6: Calculate the first-order moment of each extracted contour :

[0133]

[0134] in Pixel When calculating the pixel value, all non-zero pixel values ​​are recorded as 1.

[0135] S7: Remove abnormal contours that do not meet the requirements.

[0136] S71: Delete first moment The profile with 0, remaining 107 profiles

[0137] S72: Draw the profile rectangle, the i-th profile is in The longest distance on the x-axis is the width of the rectangle The longest distance on the y-axis is the height of the rectangle :

[0138] ,

[0139] where , is the horizontal coordinate of the i-th point constituting the profile , , is the vertical coordinate of the i-th point constituting the profile .

[0140] S73: Calculate the aspect ratio of each profile :

[0141]

[0142] S74: Draw a box plot of the aspect ratio . Roughly check for outliers.

[0143] S75: Sort the aspect ratio from small to large, extract the upper quartile point of the sorted sequence, i.e. the number at the sequence position , , the position is ; extract the lower quartile point of the sorted sequence, i.e. the number at the sequence position , , the position is , calculate the interquartile range , and the upper and lower limits of outliers are:

[0144] where k is set to 10. The specific calculation is 1.43,

[0145] 0.55. As shown in . The horizontal coordinate is the aspect ratio sequence, and the label is 1; the vertical coordinate is the specific value of the aspect ratio sequence. Figure 3 Figure 3 S76: Remove the outliers at the bottom of the box plot. At this time, there are 106 profiles remaining.

[0146] S8: Divide large holes and small holes, and calculate the number of large holes and small holes.

[0147] S8: Divide large holes and small holes, and calculate the number of large holes and small holes. ​

[0148] S81: Recalculate the first order moments with the abnormal profiles removed and sort them. Get the sorted first order moments sequence ;

[0149] S82: Calculate the first order difference :

[0150]

[0151] Where the sequence is supplemented . The sequence is shown as Figure 4 . Figure 4 The horizontal coordinate is the index, and the vertical coordinate is the difference value.

[0152] S83: Extract the index of the maximum value in the difference sequence .

[0153]

[0154] The specific calculation is 54.

[0155] S84: Finally, the number of small holes is 53, and the number of large holes is 53.

[0156] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present application shall be included in the protection scope of the present application.​

Claims

1. A method for counting and classifying circular holes in an image, characterized in that: The following steps are involved: S1. Use image acquisition equipment to collect images; S2. Crop the collected image; S3. Perform denoising based on the collected image features; S4. threshold processing based on the collected image features; S5. Extract the contours of the image after threshold processing and fill the contours. Each contour is composed of the pixels where it is located. An image has a total of The outline, The outline is ,in For each pixel's coordinates, is the number of all pixels of the contour; S6. Calculate the first-order moment of each extracted contour : in Pixel When calculating the pixel value, all non-zero pixel values ​​are recorded as 1; S7. Remove abnormal contours that do not meet the requirements; S8. Divide the large holes and small holes, and calculate the number of large holes and small holes; The S7 specifically includes: S71. Delete the first moment The contour is 0; S72. Draw outline rectangles, each outline consists of points, then The outline in The longest distance on the axis is the width of the rectangle , the longest distance on the y-axis is the height of the rectangle : ， in 、 To form the outline The horizontal coordinate of the point, 、 To form the outline The vertical coordinate of a point; S73. Calculate the aspect ratio of each contour : S74. Aspect Ratio Draw a box plot to check for abnormal points; S75. Sort from small to large and extract the upper quartile of the sorted sequence , that is, the sequence position Numbers, The location is ; Extract the lower quartile of the sorted sequence , that is, the sequence position Numbers, The location is , calculate the interquartile range , the upper and lower limits of distinguishing abnormal points are: Where k is the selected threshold; S76: Remove abnormal contours outside the upper and lower limits.

2. The method for counting and classifying circular holes in an image according to claim 1, wherein: Said S1 specifically includes: S11. Collection Preparation: For each collection, maintain a fixed light source position and intensity. Ensure a clear contrast between the background and the object being collected, and keep the background clean and free of unwanted objects. Place the object flat on the table, with the camera parallel to the object. S12. Collection process: During collection, the entire surface of the object to be collected is collected at one time. If there are circular hole information on multiple surfaces of the object to be collected, each surface is collected to ensure that only one surface is collected each time.

3. The method for counting and classifying circular holes in an image according to claim 1, wherein: In S2, the image is read on a computer in grayscale format. If it is a color image, it is converted into a grayscale image before cropping.

4. The method for counting and classifying circular holes in an image according to claim 3, wherein: The S2 specifically includes: S21. If there is more than one background, cropping is performed. When cropping, an arbitrary shape cropping method is used to retain only the object to be collected and remove redundant complex backgrounds; S22. If the background has only one color and has a large contrast with the color of the object to be collected, the background is retained and no cropping operation is performed.

5. The method for counting and classifying circular holes in an image according to claim 1, wherein: The S3 specifically includes: S31. If the image contains scattered noise or contains tiny circular holes with diameters less than one-tenth the diameter of the smallest circular hole, perform salt and pepper denoising. Pixels within the selected kernel size are sorted by pixel value, and the middle value is selected as the pixel value of the center pixel. The kernel size is the largest size that does not blur the smallest circular hole. S32. If the image has no impurities other than the single-color background and the circular hole, perform Gaussian filtering, select a kernel size, and calculate the pixel value of the intermediate value based on the two-dimensional Gaussian function. The kernel size is based on the maximum size that does not blur the smallest circular hole.

6. The method for counting and classifying circular holes in an image according to claim 1, wherein: The S4 specifically includes: If the background is simple, an adaptive thresholding algorithm is used to select the threshold by maximizing the between-class variance. If the background is not uniform, the adaptive threshold algorithm is used and then adjusted. The adaptive threshold is adjusted by 40 values. First, a coarse threshold is selected with a step size of 10 from the center of the optimal threshold. Then, a fine threshold is selected with a step size of 1 to select the optimal threshold to accurately separate the circular hole from the background. S43: If the background is not uniform, an adaptive threshold algorithm is used to ensure that the grayscale value of the image background boundary is consistent with the grayscale value of the circular hole, and a grayscale distribution map is drawn. The three to four points with the largest changes in the image valley are selected through the grayscale distribution map for threshold segmentation.

7. The method for counting and classifying circular holes in an image according to claim 1, wherein: The S8 specifically includes: S81. Recalculate the first-order moment after removing the abnormal contour and sort it to obtain the sorted first-order moment sequence , L is the number of all contours after removing abnormal contours; S82. Calculate first-order differences : Among them, the supplement S83. Extract differential sequence The subscript of the maximum value in S84. The number of small holes is , the number of large holes is .

Images