The second step focuses on making copies of single characters, with the challenge being the interference from "noise" like scratches on the bones, which can make it more difficult to recognize the white characters against their dark bone background.

"AI model generation algorithms have been used to identify and remove this 'noise' and to produce clear copies of dark characters on a white background. This process clears the way for scholars to conduct further recognition and matching work," Wang says.

Then, the copies are stored, contributing to the formation of a database.

In particular, copies of a character that appears multiple times or in different rubbings have been made to facilitate their study.

"Just like reading comprehension, we can connect the different contexts for interpretation, potentially inspiring insight into the relationships between recognized and unrecognized characters," Wang says.

The database contains 1.43 million single-character rubbings along with electronic reproductions. Each character is also mapped to its specific location in the rubbings, he adds.