Research On Defect Detection Of Printed Matter Based On Image Processing

Due to printing technology or mechanical vibration or other reasons, somesurface defects often occur in the printing process. Traditional manual detectionmethods, due to subjective factors, has been unable to meet customerrequirements on product quality; and the speed can not meet the requirements ofmodern industrial production. Therefore, how to improve the speed and accuracyof detecting is becomes one research focus in the printing field. Recently, withthe development of science and technology, the use of machine vision and imageprocessing technology for print quality inspection has become feasible andpracticable.The main content of this dissertation is to study the defect detectiontechnology and propose its own solutions. And then an experimental platformwhich simulates the real print production is designed and built. At last, thealgorithms proposed by this paper are applied to the presswork image on thedesigned experimental platform and their feasibility is validated.First of all, because real-time image can not be completely aligned with thestandard image in space, it is very necessary to finish image registration betweenthem before defect detection. Therefore, according to the in-depth comparativestudy for the commonly used image registration algorithms and the demands ofthe system, this thesis explores a suitable registration algorithm for inspectionsystem. For the text image, the combination of mathematical morphology andhough transform is implemented to acquire the skew angle of the real-time image,and then the fourier phase correlation principle-based approach is applied toobtain offset parameters. And for the screen image, the improved dual-templatematching image registration algorithm is proposed.Secondly, image registration is followed by defect detection. The design ofdefect detection algorithm is also an important aspect of the thesis. After sometypical defect detection algorithms are discussed and analyzed, a novel imagedifferencing based on a glide rectangle window is proposed in this dissertationaccording to the inspection system functional requirments. After a difference image is generated, the defect which has a certain practical significance and acertain area is searched and marked. At last, whether the number of marks isgreater than zero decides whether the printed matter has defects.Finally, an online defect detection experimental system for printed matter isdesigned and built, which is consist of a computer, a matrix scan CCD, a cameralen, a frame grabber, a light source, a single-chip microcomputer, an inverter, amotor and a conveyor belt. This system can provide a experimental platform forstudying online printing defect detection algorithms. The proposed imageregistration and defect detection algorithms are performed with the pressworkimage on the designed experimental platform. The experimental resultsdemonstrate that these methods not only have high accuracy, but also have highrate, which can meet the requirements of the inspection system. It is also provedthat the designed experimental system can detect the basic defects of the printedmatter automatically at real-time.