The Research Of Online Automatic Printing Quality Detection System Based On Machine Vision

With the increasingly stringent requirements for printing’s quality, because of the inefficiencies and high costs of traditional manual post-detection methods, it has been unable to meet the needs of modern printing production. In contrast, printing flaws detection system which was based on machine vision technology has the advantage of online, automatically and cost-effective. Therefore, based on the study of existing correlation algorithms of automatically prints inspection system which was based on machine vision, we proposed improved algorithms. The main work of this paper include the following:(1)At first, we proposed the overall design of the system, which included the hardware architecture and software architecture of the system, also gives the complete dectection process.(2)Secondly, this paper presents an algorithm that use PCA algorithms decomposed template color image into the remain component and the main component,Then we respectively used twice watershed segmentation and clustering segmentation on the two components.Finally we fused the segmentation results of different image components to obtain various color area, achieving the segmentation of template color image.(3)Thirdly, we preclude the use of quadrilateral detection and AKAZE algorithms to detect the feature points of target image and template image, and used M-LDB descriptors to describe those feature points. And then we used the fast nearest neighbor matching algorithm and RANSAC algorithm to realize the fine registration of template image and target image;(4)Then, after the pyramid decomposition of image, we respectively used statistical methods and invariant contours moment contrast to detect color and shape flaws.(5)Finally, we use OpenCV to realize the designed algorithms, and test their performance by compared with other algorithms. Proved the superiority of those designed algorithms.By the theoretical analysize and experimental results, we proved that the proposed image segmentation, image registration and flaws detection algorithms were feasible and effective, with high efficiency and reliability.