Research On Surface Defects Detection System Of Refill Ball Based On Machine Vision

Ballpoint pen is widely used in daily life and ball as the key part of refills while its surface defects directly affect the writing performance of refills.Therefore,there is a important theoretical and practical significance to research surface defect detection.This thesis has developed the design of detector,refill ball surface defect image extraction,the defect image mosaic of different attitude,refill ball surface defect pattern classification and others technical issues in-depth research,and it has built a prototype.The main content and contributions are as follows:According to the requirements of system parameters,make the design of the key parts of detector such as the automatic feeding and blanking module,the spherical unfloding module and the image acquisition module.And refill ball detector prototype based on machine vision has been built,and realizes the image acquisition of the ball surface.The technology of refill ball surface defect image exraction is done.Firstiy,image preprocessing method of image acquisition and image filtering end enhancing of refill ball surface are researched.According to the contrast of the filtering effect,Gauss filter is selected as the filtering method of the image.And image contrast is improved by the gray scale transformation method.The size and shape of the defect region varies with the location of defects while the ball ratating.Thus the image coordinates of the spherical projection transformation to spherical arc length coordinate system,this thesis established a calibration model of spherical image and resuming defect to the responding surface size and shape of a refill ball.The recognition mistake of the shadow caused by the detector device and the light source is avoided by setting an image mask.Otsu method is applied to realize the automatic threshold segmentation of ball surface images.According to the segmentation results,the seeded points of the defect image are determined.Finally,the whole defect image is extracted from the single image by region growing method.For a single ball surface image may contain only part of the defect image.The extracted defect image from the 8 images obtained from a single ball is mosaiced based on contour.And this thesis presented a new corner detection algorithm based on the area ratio of two regions.Specifically,the algorithm firstly defined the ratio between the smaller area and the larger area,which from Circular mold plate diveded by contour,as region area ratio.Then search the point of the minimum value of region area ratio on the contour line and compared with the threshold,and the point is considered as contour corner if the value is less than the threshold.By compared with the RJ contour corner detection algorithm and Teh-Chin contour corner detection algorithm,the proposed algorithm is more accurate and has better noise immunity.The shortest distance matching method is used for rough matching of the feature points,and RANSAC algorithm is used to match the image precisely.Finally,the linear weighting method is used for image fusion.Fifteen statistical features that combination of geometric and texture features were extracted as effective characteristics of three kinds of surface defects.KNN classification algorithm was used to analyze shape characterystics,moment features and statistical texture features as the chassifacation described characteristics.And effective characteristics combination was selected according to the results,which improves the system accuracy of chassifacation defects.