Novelty detection with instance-based learning for optical character quality control

Novelty detection involves modeling the normal behavior of a system and detecting any divergence from normality which may indicate onset of damage or faults. Using instance-based learning, a novelty detection approach for optical characters quality control in machine vision inspection application is given in the paper. A normal characters information pattern adapted to special application can be established by training and products information can be effectively inspected with no delay for the print error can be automatically distinguished from print quality in the process, which has been verified by the experiment.