Taha Karami
Eindhoven University of Technology
Electronic Systems Group
Canon Production Printing
EDL P16-25 P1: Deep Learning as a Service
Research assignment Deep Learning Based Real-Time Print Anomaly Detection Canon Production Printing as a global leader in the professional imaging industry develops high-end digital UVgel printer systems specialized for commercial and industrial professional graphic arts printing. Canon aims to be the “number one” in the printing industry. Therefore, the quality of its prints is of the highest importance. To develop an exceptional quality print, many factors play roles ranging from the quality of the used inks to the software level. In this project, we are addressing the artifacts that may appear on the prints due to many different reasons. The goal is to inspect the prints for any human-vison-visible artifacts, if detected any, corrective measures are taken to ensure that prints are delivered with the best possible quality and without any artifacts. To do so, we are deploying Deep Learning(DL) to develop a real-time print anomaly detection system that would be able to detect any sort of print artifacts. We are using DL as it has been proven to be the most promising solution when the problem changes over time depending on particular circumstances. Moreover, as the faulty prints do not happen frequently, acquiring a balanced annotated data set is challenging and laborious. Therefore, the focus is on unsupervised deep learning models to overcome this challenge as they do not require a balanced data set. Additionally, the requirement for the system to perform in real-time, in regard to the enormous amount of data that needs to be processed, is another formidable challenge that needs to be addressed throughout the development process. |
| Personal information |
| |