Recognition of Assay Markings in Precious Metals Using Smartphones

The validation of the authenticity of precious metals has always posed a considerable challenge for non-expert buyers. In addressing this issue for the Portuguese Mint and Official Printing Office (INCM), the VIS Team of the Institute of Systems and Robotics (ISR) of the University of Coimbra (UC) has proposed a novel authentication framework in the scope of the UniqueMark project. It consists of carving unique laser markings on metals, with posterior identification using deep learning techniques. By employing this approach, the project aims to significantly improve the recognition and verification process, providing a reliable and efficient means for ensuring the legitimacy of precious metals. Such a solution has the potential to significantly impact the field, offering enhanced security and confidence for both buyers and sellers in the realm of precious metal transactions. Following the previous contributions, the proposed goal is to study the possibility of correctly identifying and authenticating small laser markings on precious metals using only a smartphone and its native camera. To overcome this challenge, Deep Learning techniques based on CNN’s were used. Given the time and cost to construct a big enough dataset, Transfer Learning techniques are also proposed, with weights from ImageNet dataset. The previously available dataset was expanded, focusing now on captures made with smartphones, studying the effect of the lens, the Macro feature and ambient conditions such as light.