Outcome summary
We delivered five inspection machines in 2022 for a government security-printing organisation in Asia. They solved the problem that had stopped other suppliers: too many false positives.
Starting situation
A government security-printing organisation in Asia needed every national passport checked for print defects in the photo. The difficulty was false positives. Skin marks, glasses, earrings and particular print textures looked like defects to a vision system, and every false alarm meant a good passport flagged for nothing.
Several other companies had tried this project and failed, precisely because they could not solve the false-positive problem.
The system
- Five inspection machines, built and delivered by us.
- Deep-learning software that finds print defects in the photo printed in the passport.
- A comparison step: each defect found is compared with the original photo. If the same feature is already in the original photo, it is excluded from the report.
Implementation
The key was the algorithm. Deep learning finds candidate defects in the printed photo. The system then compares each one with the original photograph. A skin feature, a pair of glasses, an earring or a texture that is already in the original is not a print defect, so it is left out of the report. Only what the printing added remains.
Results and evidence
The false positives that stopped other suppliers are solved. The report lists only genuine print defects. Five machines were delivered in 2022.