Research On Visual Method Of Plastic Mask Printingquality Detection

With the continuous development of printing technology, a new generation of printing machines improve a lot in speed and quality, but now printing technology still has all kinds of printing quality defects, unable to meet higher and higher social needs in printing quality. Therefore, detection systems of printing quality were born in this background. Detection systems of printing quality greatly improve the detecting efficiency of printing quality, using machine instead of human eyes and using computer hardware and image processing software instead of human brain for defect recognition and information feedback. In this paper, the detection system of printing quality is mainly composed of distortion correction of collected images, feature extraction, image matching and defect detection. A brief discusses for four parts is as follows.For detection of printing quality of plastic film, this paper studies the calibration method in both geometric distortion and lens distortion correction, and this method represents the corresponding relation between plane coordinate of calibration plate and image coordinate using polynomial. Polynomial coefficients implicitly contains camera parameters and distortion coefficients and are estimated by the least square method. For the sorting problem in extracting feature points of calibration plate images, a method of comparing the slope of a straight line between two points is proposed. Experimental results show that this method can obtain higher correction accuracy.After the distortion correction, feature extraction of corrected images is needed. The existing detection methods of feature point can be roughly classified into three categories: the method based on contour, the method based on parameter model and the method based on gray value of image. Harris, Fast and SIFT feature point detecting algorithms are commonly used in the method based on gray value of image. In this paper, through comparing the performance and real-time property of these detection algorithms with a lot of experiments, finally Fast with best real-time is chosen.The next need to match distortion corrected images. Image matching makes overlapping on object images collected in different time and environment, aiming at improving the matching accuracy. The ORB feature point matching method based on the binary string descriptor is adopted in this paper, and feature points of plastic film to be detected are matched with standard feature points, using RANSAC to eliminate the mismatch points. The experimental results show that the method has high efficiency.According to the pairs of feature points of plastic film images to be detected and standard images to estimate the model parameters of geometrical transformation, images approximately overlapped with standard plastic film are acquired. Difference operation and morphology processing method are adopted to detect the defects in images. Experimental results show that this method has higher precision for detecting defects on plastic film.