A Printing Defect Recognition Method Based on Class-imbalanced Learning

Printing product is very common in people's daily life, such as newspapers, books, advertisements, product packaging and even banknotes. Because of various printing defects occurring in printing process, the quality of printing product is directly affected. During the high-speed printing, it is necessary to detect various printing defects quickly and accurately for eliminating defective printing products from qualified ones. The printing defect recognition method based on machine vision is replacing manual defect recognition in printing industry because of its automation, high efficiency and recognition accuracy. We use deep learning method to detect printing defects. The number of printing defects in different categories is often uneven. The quantity of normal samples and dirty points is significantly larger than other defect types. In this paper, a printing defect recognition method based on class-imbalanced learning is proposed. In addition, a warp-based bundled homography model is also proposed to improve the registration accuracy, which is beneficial to improve the efficiency of defect recognition. Experimental results demonstrate that the proposed method can detect printing defects quickly and effectively.