CN105675626B - A kind of character defect inspection method of tire-mold - Google Patents

Abstract

The invention discloses a character defect detection method of a tire mold, comprising: S1, scanning and collecting a set of images of the membrane of the tire to be detected, and obtaining the outer circular arc profile of the tire; S2, fitting the outer circular arc of the tire After the center of circle and the radius of the shaped contour, locate the tire film image area as the ROI image to be tested; S3, classify the ROI image; S4, select different methods to process according to the classification of the ROI image, and obtain the image corresponding to each ROI The CAD image block that matches the image; S5, perform character recognition, and then perform defect judgment according to the character recognition result; S6, respond to the situation of judging that there is a character defect, return to execute steps S4 and S5 to perform defect judgment again, and select the character defect Less judgmental results as final results. The invention has high detection stability, low detection cost, high detection accuracy, low false alarm rate and wide application range, and can be widely used in the field of tire membrane detection.

Description

A character defect detection method for tire molds

technical field

The invention relates to the field of image processing, in particular to a character defect detection method of a tire mold.

Background technique

Glossary:

ROI: Region Of Interest, area of interest;

NCC: Normalized Cross Correlation normalized cross-correlation.

In the production of tire molds, quality inspection is particularly important, and character defect detection is an important work of quality inspection. At present, the defect detection of tire mold characters in the industry mainly relies on human eyes. However, the noisy factory environment, a large amount of inspection work and high requirements for quality inspection make it difficult to meet the demand solely by human eye detection. Machine vision is to use machines instead of human eyes to complete observation and judgment. It is often used in product quality inspection in the mass production process, or in dangerous environments that are not suitable for humans or in occasions where human vision is difficult to satisfy. Compared with manual visual inspection , can greatly improve the detection accuracy and speed, thereby improving production efficiency, and can also avoid deviations and errors caused by human visual inspection. Machine vision has been widely used in various fields of industrial production. Machine vision is widely used in the detection of character defects in industrial products. Character defects include character stroke defects, missing and misprinted characters. The character defect detection system is required to detect characters that do not meet the requirements, that is, to detect characters with stroke defects, missing or misprinted characters. Traditional machine vision-based character defect detection methods mainly use standard character images as templates, and establish templates by extracting some features of characters, such as shape features, and match the image to be tested with the standard template. If the matching result is lower than the threshold, it is considered Yes is flawed. This method is easy to implement and has a better detection effect. However, for tire molds, due to the small production volume, it is impossible to take a standard mold image as a template from the perspective of production efficiency. The application number is 201510437595.1, and the name is a Chinese patent application for the detection method of tire membrane surface character defects based on machine vision. The straight image is used as a template, and the flat CAD image is used as a target detection method. Specifically, in the process of obtaining an image block that matches the ROI image to be tested from a straight CAD image, thresholding is performed on each ROI image to be tested. Segmentation, and then classify the ROI image to be tested by morphological operations, and convert the ROI image to be a binary image after preprocessing according to the classification. Then obtain the flat image corresponding to the CAD design drawing of the tire mold to be tested, scale the ROI image to be tested according to the height ratio of the flat image and the ROI image to be tested, and then sequentially intercept image blocks with the same width as the ROI image block , calculate the correlation coefficient, the image block with the largest correlation coefficient is used as the image block matching with the ROI image to be tested, and character recognition is performed, so as to perform defect judgment according to the character recognition result. This method has low detection cost and a wide range of applications, and can be quickly detected. However, this method mainly uses the information of large characters in the upper part of the image for matching and positioning. When there are no characters or only small characters in the upper part of the image, the result of matching and positioning will be wrong, and it is impossible to guarantee that all ROI images to be tested are correct. Accurate acquisition of matching CAD image blocks may result in large detection errors. Moreover, during character segmentation, for some densely distributed and relatively small characters, individual characters may be disconnected or glued to other characters during segmentation, causing false positives and detection errors.

Contents of the invention

In order to solve the above technical problems, the object of the present invention is to provide a method for detecting character defects of tire molds.

The technical solution adopted by the present invention to solve its technical problems is:

A character defect detection method of a tire mold, comprising steps:

S1. Sequentially scan and collect a set of images from the membrane of the tire to be detected, and process each of the collected images to obtain the outer circular arc profile of the tire;

S2. After fitting the center and radius of the outer circular arc profile of the tire, convert the outer circular arc image of the tire to be tested into a straight image to be tested by polar coordinate transformation, and perform threshold segmentation on the straight image to be tested. After that, locate the tire membrane image area as the ROI image to be tested;

S3. Perform threshold segmentation on each ROI image respectively, and then classify the ROI images after the threshold segmentation through morphological operations, and simultaneously obtain a CAD flat image corresponding to the CAD design drawing of the tire membrane to be detected;

S4, according to the classification of the ROI image, select different methods to process the CAD flat image and the ROI image, and intercept and obtain CAD image blocks that match each ROI image after processing on the processed flat image ;

S5. Perform character recognition on each ROI image and its matched CAD image block, and then perform defect judgment according to the character recognition result;

S6. In response to judging that there are character defects, return to steps S4 and S5 to perform defect judgment again, and select a judgment result with fewer character defects as the final result.

Further, the step of thresholding each ROI image as described in step S3, and then classifying the thresholded ROI images through morphological operations, specifically includes:

S31. Calculate the initial area and initial height of each ROI image respectively;

S32. Perform threshold segmentation on the ROI image to divide the foreground area;

S33. Obtain preset morphological structural elements, corrode the foreground area and perform connectivity marking, and filter the connected domains according to the preset screening conditions, and classify the ROI images into categories A and B according to the number of connected domains selected. kind.

Further, after the connected domains are screened according to the preset screening conditions in the step S33, the step of dividing the ROI image into category A and category B according to the number of connected domains screened out is specifically as follows:

Screen the connected domains, and select the number of connected domains whose area is greater than 1/2 of the initial area and whose height is greater than 1/2 of the initial height. The number of connected domains of is greater than 0, and the ROI image is classified into class B.

Further, the step S4 includes:

S41. According to the classification of the ROI image, according to the iterations of the processing times, for the case where the ROI image is Class A, sequentially select the processing method 1 and the processing method 2 to process the CAD flat image and the ROI image, and the ROI image is Class B. case, select processing method 2 and processing method 3 in turn to process the CAD flat image and ROI image;

S42. Scale the ROI image according to the height ratio between the processed CAD flat image and the ROI image;

S43. On the processed CAD flat image, sequentially intercept image blocks with the same width as the zoomed ROI image, calculate the correlation coefficient between each image block and the zoomed ROI image, and then use the image block with the largest correlation coefficient as A CAD image block matching the ROI image;

Said processing method 1 is specifically: performing morphological processing on the ROI image after threshold segmentation, and according to the ratio of the area area after the morphological processing to the initial area, locate and obtain the character area containing the character, and then convert the ROI image into two value image, and at the same time convert the CAD flat image into a binary image after thresholding;

The second processing method is specifically: performing standard deviation filtering on the ROI image and the CAD flat image, and intercepting the lower half of the standard deviation filtered ROI image;

The third processing method specifically includes: using the canny operator to obtain the edge image of the ROI image and converting it into a binary image, and at the same time performing threshold segmentation on the CAD flat image and converting it into a binary image.

Further, the step S1 is specifically:

The tire membrane to be detected is scanned and collected in turn to obtain a set of images, and after image denoising and threshold segmentation are performed on each of the collected images, the contour of the tire membrane is obtained, and then the contour is broken according to the curvature of the contour, so that According to the direction, length and curvature of each segment of the profile, the outer circular arc profile of the tire is obtained.

Further, the step S5 includes:

S51. Carry out morphological operation after performing threshold segmentation on the CAD image block, perform connected domain screening according to a preset connected domain threshold, and divide the CAD image block into a small character area and a large character area;

S52. Perform a morphological operation on the ROI image according to the classification of the ROI image, and then perform connected domain screening according to a preset connected domain threshold, and divide the ROI image into a small character area and a large character area;

S53. According to the classification of the ROI image, after feature extraction and feature matching are performed on the ROI image and the large character area of the CAD image block, defect judgment is performed according to the matching result;

S54. After character recognition is performed on the small character area of the ROI image and the CAD image block, the character array obtained by the recognition is divided into multiple character strings, and then the character strings obtained by the ROI image and the CAD image block recognition are matched, and according to the matching As a result, a defect judgment is performed.

Further, the step S52 is specifically:

For the case that the ROI image is class A, after performing local threshold segmentation and region growth on the ROI image, morphological operations are performed, and then connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also divided into small character areas and large character areas. character area;

For the case that the ROI image is of type B, the ROI image is sequentially subjected to mean filtering, region growing segmentation, and binarization result inversion, followed by morphological operations, and then the connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also Divided into a small character area and a large character area.

Further, the step S53 is specifically:

For the case where the ROI image is type A, use the large character area of the ROI image as a template to perform NCC matching on the large character area of the CAD image block. If the matching degree is greater than the preset matching threshold, the ROI image and the CAD image block After performing morphological subtraction, difference operation, and morphological corrosion on the large character area in sequence, it is judged whether the area of the obtained area is smaller than the preset threshold, if so, it is judged that the character has a printing defect, otherwise, it is judged that the character is printed correctly;

For the case where the ROI image is type B, use the canny operator to obtain the edge of the large character area of the ROI image as a template, and search and match on the binary image of the large character area of the CAD image block. If the matching degree is less than the preset If it matches the threshold, it is determined that the character has a printing defect, and the center coordinates of the area where the wrong character is located are recorded.

Further, the step S54 includes:

S541. After character recognition is performed on the small character area of the ROI image and the CAD image block, two character arrays are respectively obtained, and each character array is divided into multiple character strings;

S542. For each character string of the ROI image, perform a matching search in the character array of the CAD image block. If the match is unsuccessful, perform step S543. Otherwise, it is judged that the character string is printed correctly;

S543. Re-divide the character area after adjusting the parameters, and then re-extract characters from the ROI image of the small character area corresponding to the character string, and then recognize and obtain a new character string;

S544, match and search this new character string in the character array of CAD image block, if match is unsuccessful, then return to step S543 to repeat character string identification, match operation, and judge whether to match successfully in the specified number of searches, if so, then Judging that the string is printed correctly, and entering it into the correct character array, otherwise, judging that the string has a printing defect, and recording the center coordinates of the area where the wrong string is located;

S545. For each character string of the CAD image block, perform a matching search in the correct character array, if the match is unsuccessful, then perform step S546, otherwise it is judged that the character string is printed correctly;

S546. After extracting characters on the ROI image of the region corresponding to the character string, identify and obtain a verification character array, and then perform a matching search on the character string in the verification character array. If the match is unsuccessful, it is determined that the character string exists Missing printing defects, otherwise it is judged that the string is printed correctly.

Further, the flat image corresponding to the CAD design drawing in the step S3 is obtained in the following manner:

Get the CAD design drawing and convert it into a binary image after threshold segmentation, use the engraved image in the CAD design drawing as the foreground image, and then fit the minimum circumscribed circle of the foreground image, and obtain the center and radius of the minimum circumscribed circle Finally, the polar coordinate transformation is carried out according to the obtained circle center and radius, and the flat image of the CAD design drawing is obtained.

The beneficial effects of the present invention are: a character defect detection method of a tire mold according to the present invention, comprising: S1, sequentially scan and collect a group of images of the tire film to be detected, and perform a separate image processing on each collected image After processing, obtain the outer circular arc profile of the tire; S2, after fitting the center and radius of the outer circular arc profile of the tire, convert the circular arc image outside the tire to be measured into a straight image to be measured by polar coordinate transformation, and After performing threshold segmentation on the straight image to be tested, locate the tire membrane image area as the ROI image to be tested; S3, perform threshold segmentation on each ROI image respectively, and then perform threshold segmentation on the ROI image after threshold segmentation through morphological operations. Classify, obtain the CAD straight image corresponding to the CAD design drawing of tire film to be detected simultaneously; S4, according to the classification of ROI image, select different methods to process CAD straight image and ROI image, and after processing Intercepting on the flat image to obtain a CAD image block that matches each ROI image after processing; S5, performing character recognition on each ROI image and its matching CAD image block, and then performing defect judgment according to the character recognition result; S6 1. In response to judging that there are character defects, return to steps S4 and S5 to perform defect judgment again, and select a judgment result with fewer character defects as the final result. The method can automatically detect the character defects of the tire membrane to be detected, has high detection stability, low detection cost, high detection accuracy, low false alarm rate and wide application range, and can quickly and effectively detect the character defects of the tire membrane .

Description of drawings

The present invention will be further described below in conjunction with drawings and embodiments.

Fig. 1 is a flowchart of a character defect detection method of a tire mold according to the present invention.

Detailed ways

With reference to Fig. 1, the present invention provides a kind of character defect detection method of tire mold, comprises steps:

S1. Sequentially scan and collect a set of images from the membrane of the tire to be detected, and process each of the collected images to obtain the outer circular arc profile of the tire;

S2. After fitting the center and radius of the outer circular arc profile of the tire, convert the outer circular arc image of the tire to be tested into a straight image to be tested by polar coordinate transformation, and perform threshold segmentation on the straight image to be tested. After that, locate the tire membrane image area as the ROI image to be tested;

S3. Perform threshold segmentation on each ROI image respectively, and then classify the ROI images after the threshold segmentation through morphological operations, and simultaneously obtain a CAD flat image corresponding to the CAD design drawing of the tire membrane to be detected;

S4, according to the classification of the ROI image, select different methods to process the CAD flat image and the ROI image, and intercept and obtain CAD image blocks that match each ROI image after processing on the processed flat image ;

S5. Perform character recognition on each ROI image and its matched CAD image block, and then perform defect judgment according to the character recognition result;

S6. In response to judging that there are character defects, return to steps S4 and S5 to perform defect judgment again, and select a judgment result with fewer character defects as the final result.

Further as a preferred embodiment, the step of performing threshold segmentation on each ROI image in step S3, and then classifying the thresholded ROI images through morphological operations, specifically includes:

S31. Calculate the initial area and initial height of each ROI image respectively;

S32. Perform threshold segmentation on the ROI image to divide the foreground area;

S33. Obtain preset morphological structural elements, corrode the foreground area and perform connectivity marking, and filter the connected domains according to the preset screening conditions, and classify the ROI images into categories A and B according to the number of connected domains selected. kind.

Further as a preferred embodiment, after the connected domains are screened according to the preset screening conditions in the step S33, the step of dividing the ROI image into the A category and the B category according to the number of the filtered connected domains is specifically as follows:

Screen the connected domains, and select the number of connected domains whose area is greater than 1/2 of the initial area and whose height is greater than 1/2 of the initial height. The number of connected domains of is greater than 0, and the ROI image is classified into class B.

Further as a preferred embodiment, the step S4 includes:

S41. According to the classification of the ROI image, according to the iterations of the processing times, for the case where the ROI image is Class A, sequentially select the processing method 1 and the processing method 2 to process the CAD flat image and the ROI image, and the ROI image is Class B. case, select processing method 2 and processing method 3 in turn to process the CAD flat image and ROI image;

S42. Scale the ROI image according to the height ratio between the processed CAD flat image and the ROI image;

S43. On the processed CAD flat image, sequentially intercept image blocks with the same width as the zoomed ROI image, calculate the correlation coefficient between each image block and the zoomed ROI image, and then use the image block with the largest correlation coefficient as A CAD image block matching the ROI image;

Said processing method 1 is specifically: performing morphological processing on the ROI image after threshold segmentation, and according to the ratio of the area area after the morphological processing to the initial area, locate and obtain the character area containing the character, and then convert the ROI image into two value image, and at the same time convert the CAD flat image into a binary image after thresholding;

The second processing method is specifically: performing standard deviation filtering on the ROI image and the CAD flat image, and intercepting the lower half of the standard deviation filtered ROI image;

The third processing method specifically includes: using the canny operator to obtain the edge image of the ROI image and converting it into a binary image, and at the same time performing threshold segmentation on the CAD flat image and converting it into a binary image.

Further as a preferred embodiment, the step S1 is specifically:

The tire membrane to be detected is scanned and collected in turn to obtain a set of images, and after image denoising and threshold segmentation are performed on each of the collected images, the contour of the tire membrane is obtained, and then the contour is broken according to the curvature of the contour, so that According to the direction, length and curvature of each segment of the profile, the outer circular arc profile of the tire is obtained.

Further as a preferred embodiment, the step S5 includes:

S51. Carry out morphological operation after performing threshold segmentation on the CAD image block, perform connected domain screening according to a preset connected domain threshold, and divide the CAD image block into a small character area and a large character area;

S52. Perform a morphological operation on the ROI image according to the classification of the ROI image, and then perform connected domain screening according to a preset connected domain threshold, and divide the ROI image into a small character area and a large character area;

S53. According to the classification of the ROI image, after feature extraction and feature matching are performed on the ROI image and the large character area of the CAD image block, defect judgment is performed according to the matching result;

S54. After character recognition is performed on the small character area of the ROI image and the CAD image block, the character array obtained by the recognition is divided into multiple character strings, and then the character strings obtained by the ROI image and the CAD image block recognition are matched, and according to the matching As a result, a defect judgment is performed.

Further as a preferred implementation manner, the step S52 is specifically:

For the case that the ROI image is class A, after performing local threshold segmentation and region growth on the ROI image, morphological operations are performed, and then connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also divided into small character areas and large character areas. character area;

For the case that the ROI image is of type B, the ROI image is sequentially subjected to mean filtering, region growing segmentation, and binarization result inversion, followed by morphological operations, and then the connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also Divided into a small character area and a large character area.

Further as a preferred implementation manner, the step S53 is specifically:

For the case where the ROI image is type A, use the large character area of the ROI image as a template to perform NCC matching on the large character area of the CAD image block. If the matching degree is greater than the preset matching threshold, the ROI image and the CAD image block After performing morphological subtraction, difference operation, and morphological corrosion on the large character area in sequence, it is judged whether the area of the obtained area is smaller than the preset threshold, if so, it is judged that the character has a printing defect, otherwise, it is judged that the character is printed correctly;

For the case where the ROI image is type B, use the canny operator to obtain the edge of the large character area of the ROI image as a template, and search and match on the binary image of the large character area of the CAD image block. If the matching degree is less than the preset If it matches the threshold, it is determined that the character has a printing defect, and the center coordinates of the area where the wrong character is located are recorded.

Further as a preferred implementation manner, the step S54 includes:

S541. After character recognition is performed on the small character area of the ROI image and the CAD image block, two character arrays are respectively obtained, and each character array is divided into multiple character strings;

S542. For each character string of the ROI image, perform a matching search in the character array of the CAD image block. If the match is unsuccessful, perform step S543. Otherwise, it is judged that the character string is printed correctly;

S543. Re-divide the character area after adjusting the parameters, and then re-extract characters from the ROI image of the small character area corresponding to the character string, and then recognize and obtain a new character string;

S544, match and search this new character string in the character array of CAD image block, if match is unsuccessful, then return to step S543 to repeat character string identification, match operation, and judge whether to match successfully in the specified number of searches, if so, then Judging that the string is printed correctly, and entering it into the correct character array, otherwise, judging that the string has a printing defect, and recording the center coordinates of the area where the wrong string is located;

S545. For each character string of the CAD image block, perform a matching search in the correct character array, if the match is unsuccessful, then perform step S546, otherwise it is judged that the character string is printed correctly;

S546. After extracting characters on the ROI image of the region corresponding to the character string, identify and obtain a verification character array, and then perform a matching search on the character string in the verification character array. If the match is unsuccessful, it is determined that the character string exists Missing printing defects, otherwise it is judged that the string is printed correctly.

Further as a preferred embodiment, in the step S2, the step of fitting the outer circular arc profile of the tire and converting it into a straight image to be tested and then segmenting it is specifically: fitting the outer circular arc of the tire After the center and radius of the contour, the circular arc image on the outside of the tire to be tested is converted into a straight image to be tested by polar coordinate transformation, and the straight image to be tested is segmented.

Further as a preferred embodiment, the straight image corresponding to the CAD design drawing in the step S3 is obtained in the following manner:

Get the CAD design drawing and convert it into a binary image after threshold segmentation, use the engraved image in the CAD design drawing as the foreground image, and then fit the minimum circumscribed circle of the foreground image, and obtain the center and radius of the minimum circumscribed circle Finally, the polar coordinate transformation is carried out according to the obtained circle center and radius, and the flat image of the CAD design drawing is obtained.

The present invention will be described in detail below in conjunction with specific embodiments.

With reference to Fig. 1, a kind of character defect detection method of tire mould, comprises steps:

S1. Sequentially scan the membrane of the tire to be detected and acquire a set of images, and process each of the collected images separately to obtain the outer circular arc profile of the tire, specifically: scan the membrane of the tire to be detected in sequence And collect a set of images, and perform image denoising and threshold segmentation on each of the collected images to obtain the contour of the tire membrane, and then break the contour according to the curvature of the contour, so that according to the direction and length of each contour and the curvature to obtain the outer circular arc profile of the tire.

The specific steps of breaking the contour according to the contour curvature are as follows: judge whether the point on the contour is on a straight line or an arc according to the contour area, if the curvature of a certain point is consistent with the curvature of a nearby point, it means that the two points are on the same arc or on the same straight line, otherwise, it means that the two points are not on the same arc or straight line, and the two points are disconnected. Through this method, the direction, length and curvature of each segment of the profile can be obtained, and according to the features of the circular arc profile, the outer circular arc profile of the tire can be obtained.

S2. After fitting the center and radius of the outer circular arc profile of the tire, convert the outer circular arc image of the tire to be tested into a straight image to be tested by polar coordinate transformation, and perform threshold segmentation on the straight image to be tested. After that, locate the tire membrane image area as the ROI image to be tested;

S3. Perform threshold segmentation on each ROI image respectively, and then classify the ROI images after the threshold segmentation through morphological operations, and simultaneously obtain a CAD flat image corresponding to the CAD design drawing of the tire membrane to be detected;

Wherein, the step of performing threshold segmentation on each ROI image respectively, and then classifying the ROI images after threshold segmentation through morphological operations, specifically includes steps S31-S33:

S31. Calculate the initial area S_area and initial height S_Height of each ROI image respectively;

S32. Perform threshold segmentation on the ROI image to divide the foreground area, for example, the selected threshold is 120, and the area higher than 120 is divided into the foreground area;

S33. Obtain preset morphological structural elements, corrode the foreground area and perform connectivity marking, and filter the connected domains according to the preset screening conditions, and classify the ROI images into categories A and B according to the number of connected domains selected. Class, specifically: obtain the preset morphological structure elements, corrode the foreground area and mark the connectivity, and filter the connected domain, and filter out the area greater than 1/2 of the initial area S_area and the height greater than 1/2 of the initial height S_Height The number of connected domains, if the number of connected domains selected is equal to 0, the ROI image will be classified into class A, if the number of connected domains selected is greater than 0, the ROI image will be classified into class B.

The flat image corresponding to the CAD design drawing is obtained by the following methods: obtain the CAD design drawing and convert it into a binary image after thresholding, and use the engraved image in the CAD design drawing as the foreground image, and then fit the foreground The minimum circumscribed circle of the image, and after obtaining the center and radius of the minimum circumscribed circle, perform polar coordinate transformation according to the obtained center and radius to obtain a flat image of the CAD design drawing.

More specifically, the steps of fitting the minimum circumscribed circle of the foreground image and obtaining the center and radius of the minimum circumscribed circle are as follows:

Obtained by nonlinear optimal iterative method fitting: fitting the circumscribed circle of the foreground image, accumulating and summing the square distances from all points on the edge to the fitted circumcircle according to the following formula, and then calculating the circumscribed circle with the smallest sum As the minimum circumscribed circle of the foreground image, and obtain the center and radius of the minimum circumscribed circle:

In the above formula, ε ² represents the cumulative sum of the square distances from all points on the edge to the fitted circumcircle, (α, β) represents the coordinates of the center of the circle, ρ represents the radius of the circle, and (r _i , c _i ) represents the edge The coordinates of the point on .

Preferably, the step of performing polar coordinate transformation according to the obtained circle center and radius is specifically:

Combining with the following formula, according to the obtained circle center and radius, the CAD design drawing is transformed into polar coordinates:

In the above formula, (α, β) represents the coordinates of the transformation center, Indicates the coordinates of the point on the CAD design drawing after polar coordinate transformation, d _i is the distance relative to the transformation center, is the vector angle, (r _i , c _i ) is the coordinates before polar coordinate transformation.

S4, according to the classification of the ROI image, select different methods to process the CAD flat image and the ROI image, and intercept and obtain CAD image blocks that match each ROI image after processing on the processed flat image , specifically including steps S41-S43:

S41. According to the classification of the ROI image, according to the iterations of the processing times, for the case where the ROI image is Class A, sequentially select the processing method 1 and the processing method 2 to process the CAD flat image and the ROI image, and the ROI image is Class B. For the situation, select processing method 2 and processing method 3 in turn to process CAD flat image and ROI image; according to the number of iterations of processing, select different methods to process CAD flat image and ROI image in sequence, such as ROI If the image is type A, choose method 1 for processing for the first time, if the detection result is wrong, choose method 2 for the second processing;

S42. Scale the ROI image according to the height ratio between the processed CAD flat image and the ROI image;

S43. On the processed CAD flat image, sequentially intercept image blocks with the same width as the zoomed ROI image, calculate the correlation coefficient between each image block and the zoomed ROI image, and then use the image block with the largest correlation coefficient as The CAD image block matching the ROI image; the correlation coefficient refers to the correlation coefficient between the image block and the matrix of the zoomed ROI image.

The first processing method is as follows: performing threshold segmentation on the ROI image and performing morphological processing, and according to the ratio of the area area after the morphological processing to the initial area S_area, locating and obtaining the character area containing the character, and then converting the ROI image into Binarize the image, and at the same time convert the CAD flat image into a binary image after thresholding;

The second processing method is specifically: performing standard deviation filtering on the ROI image and the CAD flat image, and intercepting the lower half of the standard deviation filtered ROI image;

The third processing method specifically includes: using the canny operator to obtain the edge image of the ROI image and converting it into a binary image, and at the same time performing threshold segmentation on the CAD flat image and converting it into a binary image.

S5. Perform character recognition on each ROI image and its matching CAD image block, and then perform defect judgment according to the character recognition result, specifically including S51-S54:

S51. Carry out morphological operation after performing threshold segmentation on the CAD image block, perform connected domain screening according to a preset connected domain threshold, and divide the CAD image block into a small character area and a large character area;

S52. According to the classification of the ROI image, perform a morphological operation on the ROI image, and then perform connected domain screening according to the preset connected domain threshold, and divide the ROI image into a small character area and a large character area, specifically: for the ROI image: In the case of category A, after local threshold segmentation and region growth are performed on the ROI image, morphological operations are performed, and then connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also divided into small character areas and large character areas;

For the case that the ROI image is of type B, the ROI image is sequentially subjected to mean filtering, region growing segmentation, and binarization result inversion, followed by morphological operations, and then the connected domain screening is performed according to the preset connected domain threshold, and the ROI image is also Divided into a small character area and a large character area.

S53. According to the classification of the ROI image, after feature extraction and feature matching are performed on the ROI image and the large character area of the CAD image block, defect judgment is performed according to the matching result, specifically:

For the case where the ROI image is type A, use the large character area of the ROI image as a template to perform NCC matching on the large character area of the CAD image block. If the matching degree is greater than the preset matching threshold, the ROI image and the CAD image block After performing morphological subtraction, difference operation, and morphological corrosion on the large character area in sequence, it is judged whether the area of the obtained area is smaller than the preset threshold, if so, it is judged that the character has a printing defect, otherwise, it is judged that the character is printed correctly;

For the case where the ROI image is type B, use the canny operator to obtain the sub-pixel precision edge of the large character area of the ROI image as a template, and search and match on the binary image of the large character area of the CAD image block, if the matching degree If it is less than the preset matching threshold, it is determined that the character has a printing defect, and the center coordinates of the area where the wrong character is located are recorded.

S54. After character recognition is performed on the small character area of the ROI image and the CAD image block, the character array obtained by the recognition is divided into multiple character strings, and then the character strings obtained by the ROI image and the CAD image block recognition are matched, and according to the matching As a result, the defect judgment is carried out, which specifically includes steps S541-S546:

S541. After character recognition is performed on the small character areas of the ROI image and the CAD image block, two character arrays are respectively obtained, and then each character array is divided into multiple character strings; the character arrays corresponding to the ROI image and the CAD image block are respectively Array_ROI and Array_CAD, the strings are String_ROI[j] and String_CAD[i] respectively, where i=1,2,3...N, N is the number of strings corresponding to the CAD image block, j=1,2, 3...M, M is the number of character strings corresponding to the ROI image;

S542, for each string String_ROI[j] of the ROI image, perform a matching search in the character array Array_CAD of the CAD image block, if the matching is unsuccessful, then perform step S543, otherwise it is judged that the string is printed correctly;

S543. Re-divide the character area after adjusting the parameters, and then re-extract characters from the ROI image of the small character area corresponding to the character string, and then recognize and obtain a new character string;

S544, match and search this new character string in the character array Array_CAD of the CAD image block, if the match is unsuccessful, then return to step S543 to repeat the character string identification and matching operations, and judge whether the match is successful within the specified search times, if so, Then it is judged that the string is printed correctly, and entered into the correct character array String_Re, otherwise, it is judged that the string has a printing defect, and the center coordinate of the area where the wrong string is located is recorded;

S545. For each string String_CAD[i] of the CAD image block, perform a matching search in the correct character array String_Re, if the matching is unsuccessful, then perform step S546, otherwise judge that the string is printed correctly;

S546. After extracting characters on the ROI image of the region corresponding to the character string, identify and obtain a verification character array String_Recheck, and then perform a matching search on the character string in the verification character array String_Recheck, and if the match is unsuccessful, determine the character If there is a missing printing defect in the string, otherwise it is judged that the string is printed correctly.

S6. In response to judging that there are character defects, return to steps S4 and S5 to perform defect judgment again, and select a judgment result with fewer character defects as the final result.

In the detection process of this method, a feedback mechanism is integrated. After the first character defect detection, if the detection fails, the character defect detection is performed again. After two defect detections, the result with fewer errors is taken as the final result to improve The stability of the detection is improved, the false alarm rate is reduced, the matching error caused by a single detection process is avoided, and the detection accuracy is enhanced. Moreover, for different character classifications, different detection methods are used for judgment. For character segments that are prone to errors and false positives, a feedback mechanism is incorporated. After the first character area division, the intercepted CAD image block and ROI image are detected and judged. , if the judgment is unqualified, feed back the wrong result to the previous step, change the method, perform the second character area division, and then perform the detection and judgment again, which improves the detection accuracy.

Preferably, the following steps are also included:

S7, sequentially splicing the ROI images of the collected images according to the collection sequence to obtain a straight spliced image of the tire membrane to be detected;

S8. Performing an inverse polar coordinate transformation on the flat spliced image to obtain an arc-shaped spliced image;

S9. For the characters that are judged to be defective, highlight them on the corresponding positions of the arc-shaped mosaic image. Characters with printing defects are marked with red circles on the corresponding positions of the arc-shaped mosaic image.

The method can automatically detect the character defect of the tire membrane to be detected, has high detection stability, low detection cost and wide application range, and can quickly and effectively detect the character defect of the tire membrane.

The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the described embodiments, those skilled in the art can also make various equivalent deformations or replacements without violating the spirit of the present invention , these equivalent modifications or replacements are all included within the scope defined by the claims of the present application.

Claims (7)

1. a kind of character defect inspection method of tire-mold, which is characterized in that including step：

S1, detection tire fetal membrane is treated successively be scanned and acquire and obtain one group of image, and respectively to acquired every figure As obtaining tire outer arc shape profile after being handled；

It, will be round on the outside of tire to be measured by polar coordinate transform after S2, the center of circle for being fitted tire outer arc shape profile and radius Arc image is converted to flat type testing image, and positions tire fetal membrane figure into after row threshold division to flat type testing image As region is as ROI image to be measured；

S3, respectively to each ROI image into row threshold division, and then by morphology operations by the ROI image after Threshold segmentation Classify, while the corresponding CAD flat types image of the CAD design figure for obtaining tire fetal membrane to be detected；

S4, the classification according to ROI image, select different methods to handle CAD flat types image and ROI image, and Interception obtains and treated CAD image block that each ROI image matches on treated flat type image；

S5, character recognition is carried out to each ROI image and matched CAD image block, and then according to character identification result Carry out defect dipoles；

S6, in response to judge there are character defects the case where, return to step S4 and S5 to again carry out defect dipoles after, The judging result for selecting character defect less is as final result；

Respectively to each ROI image into row threshold division described in the step S3, and then threshold value is divided by morphology operations The step of ROI image after cutting is classified, specifically include S31~S33：

S31, the initial area and elemental height for calculating separately each ROI image；

S32, to ROI image into row threshold division, divide foreground area；

S33, it obtains and presets Morphological Structuring Elements, corroded and carried out connectivity label to foreground area, and according to default sieve After selecting condition to screen connected domain, ROI image is divided by A classes and B classes according to the connected domain quantity filtered out；

After being screened to connected domain according to default screening conditions described in the step S33, according to the connected domain number filtered out The step of ROI image is divided into A classes and B classes by amount, it is specially：

Connected domain is screened, the connected domain number that area is more than 1/2 elemental height more than 1/2 initial area and height is filtered out ROI image is divided into A classes by amount if the connected domain quantity filtered out is equal to 0, if the connected domain quantity filtered out is more than 0, ROI image is divided into B classes；

The step S4, including S41~S43：

S41, the classification according to ROI image the case where for ROI image being A classes, are selected successively according to the iteration of number of processes Processing method one and processing method two handle CAD flat types image and ROI image, for the feelings that ROI image is B classes Condition selects processing method two and processing method three to handle CAD flat types image and ROI image successively；

S42, ROI image is zoomed in and out according to the height ratio of treated CAD flat types image and ROI image；

With the ROI image after scaling with wide image block, calculating is every for interception successively on S43, CAD flat type images after treatment A image block with scaling after ROI image related coefficient, and then using the maximum image block of related coefficient as with the ROI image The CAD image block to match；

The processing method one is specially：To ROI image into carrying out Morphological scale-space after row threshold division, and according to morphology at The ratio of region area and initial area after reason, positioning obtains the character zone for including character, and then ROI image is converted to Binary image, while by CAD flat types image into being converted to bianry image after row threshold division；

The processing method two is specially：Standard deviation filtering is carried out to ROI image and CAD flat type images, and intercepts standard deviation The lower half portion of filtered ROI image；

The processing method three is specially：The edge image of ROI image is obtained using canny operators and is converted to bianry image, Simultaneously by CAD flat types image into being converted to bianry image after row threshold division.

2. a kind of character defect inspection method of tire-mold according to claim 1, which is characterized in that the step S1 is specially：

Treat detection tire fetal membrane successively and be scanned and acquire and obtain one group of image, and respectively to every image being acquired into After row image denoising and Threshold segmentation processing, tire fetal membrane profile is obtained, and then profile is disconnected according to contour curvature, to basis Direction, length and the curvature of every section of profile obtain tire outer arc shape profile.

3. a kind of character defect inspection method of tire-mold according to claim 1, which is characterized in that the step S5, including：

S51, to CAD image block into morphology operations are carried out after row threshold division, be connected to according to preset connected domain threshold value Domain is screened, and CAD image block is divided into small characters region and big character zone；

S52, the classification according to ROI image, to ROI image carry out morphology operations, then according to preset connected domain threshold value into Row connected domain is screened, and ROI image is also divided into small characters region and big character zone；

S53, the classification according to ROI image carry out feature extraction and feature to the big character zone of ROI image and CAD image block After matching, defect dipoles are carried out according to matching result；

S54, for the small characters region of ROI image and CAD image block, after carrying out character recognition, number of characters that identification is obtained Group is divided into multiple character strings, and then after the character string that ROI image and the identification of CAD image block obtain is matched, according to matching As a result defect dipoles are carried out.

4. a kind of character defect inspection method of tire-mold according to claim 3, which is characterized in that the step S52 is specially：

The case where for ROI image being A classes, after carrying out Local threshold segmentation and region growing to ROI image, carry out form student movement It calculates, connected domain screening is then carried out according to preset connected domain threshold value, ROI image is also divided into small characters region and big character Region；

For ROI image be B classes the case where, carry out mean filter, Region growing segmentation and binaryzation knot successively to ROI image Fruit carries out morphology operations after negating, and then carries out connected domain screening according to preset connected domain threshold value, ROI image is also divided For small characters region and big character zone.

5. a kind of character defect inspection method of tire-mold according to claim 4, which is characterized in that the step S53 is specially：

For ROI image be A classes the case where, using the big character zone of ROI image as template, in the big character of CAD image block NCC matchings are carried out on region, if matching degree is more than preset matching threshold value, to the big character area of ROI image and CAD image block Domain carries out morphology and subtracts each other, after difference operation and morphological erosion successively, and it is default to judge whether the area in the region obtained is less than Threshold value, if so, judging the character, there are printing defects, otherwise, it is determined that the character printing is correct；

For ROI image be B classes the case where, using canny operators obtain ROI image big character zone edge after be used as mould Plate scans for matching on the binary image of the big character zone of CAD image block, if matching degree is less than preset matching threshold Value then judges the character there are printing defects, and the centre coordinate of misregistration character region.

6. a kind of character defect inspection method of tire-mold according to claim 4, which is characterized in that the step S54, including：

S541, two number of characters are obtained respectively after progress character recognition for the small characters region of ROI image and CAD image block Group, and then each character array is divided into multiple character strings；

S542, each character string to ROI image, carry out matching search in the character array of CAD image block, if matching not at Work(thens follow the steps S543, otherwise judges that character string printing is correct；

Character zone division is re-started after S543, adjusting parameter, and then the ROI in the corresponding small characters region of the character string is schemed After extracting character again, identification obtains a new character string；

S544, the new character string, if matching is unsuccessful, return to step are searched in matching in the character array of CAD image block The identification of S543 repeat character strings, matching operation, and judge whether the successful match in defined searching times, if so, judging Character string printing is correct, and is entered into correct characters array, conversely, judging the character string, there are printing defects, and record The centre coordinate of error character string region；

S545, each character string to CAD image block, carry out matching search in correct characters array, if matching is unsuccessful, Step S546 is executed, otherwise judges that character string printing is correct；

S546, it after extracting character on the ROI image of the character string corresponding region, identifies and obtains an array of checking character, into And matching search is carried out to the character string in array of checking character, if matching is unsuccessful, judges that the character string exists and bite Defect, on the contrary judge that character string printing is correct.

7. a kind of character defect inspection method of tire-mold according to claim 1, which is characterized in that the step S3 Described in the corresponding flat type image of CAD design figure obtain in the following manner：

Obtain CAD design figure and by it into being converted to binary map after row threshold division, using the marking image in CAD design figure as Foreground image, and then be fitted the minimum circumscribed circle of the foreground image, and after obtaining the center of circle and the radius of the minimum circumscribed circle, according to The center of circle of acquisition carries out polar coordinate transform with radius, obtains the flat type image of CAD design figure.