The mechanics of the system had to be following: The software checks scanned copies of documents in jpg, png, and pdf. After a visual analysis of the specification, the system puts a red box around elements of dubious authenticity. It also provides a quantitative estimate of the counterfeiting likelihood (in percentage).
The following markers were defined to solve the task:
- Print area (a stamp inserted from another document).
- Text (a copied fragment, a text typed over, rotation of the text, etc.).
- Tables (a table is rotated relative to the text).
The Eastwind team was able to successfully implement the project. Now, when a fake data sheet is detected on the client’s side, it is sent to a special service for verification, which finds out why the counterparty has submitted such a document and decides how to handle it next.
Work continues on our part to improve the statistical indicators of the developed service. One of the closest milestones will be automating the data collection and choosing a convenient storage. This will allow analysts to access the data more quickly and easily to fine-tune the analytical model.
