Automatic detection and segmentation of offset lithography defects using full reference image comparison

Abstract

Offset printing has become the modern standard in high-volume industrial printing of magazines and newspapers. Therefore, the continuous improvement of the printing quality through the detection and removal of printing defects is an indispensable necessity. A vision-based inspection method that compares an inspected image to a reference image is proposed to detect and segment printing defects. The structural similarity image metric is used together with a Bayesian Classifier to detect defective regions in the inspected image. The methods are evaluated using a dataset of artificially simulated images based on defects samples from the printing facility. Results show that the proposed methods can detect 93% of the simulated defects.