Optimization of Romanian Identity Documents Processing

Abstract

This paper presents the development of an application based on computer vision and deep learning aimed at the automatic recognition of data from Romanian national identity documents. Its primary goal is to optimize the processing of such documents within public and private institutions. Developing such a system involves several challenges. These include the variability in the visual appearance of documents caused by wear and tear, scanning angles, camera quality, and ambient lighting conditions; the difficulty of accurately recognizing Romanian diacritics; and the correct identification of textual fields based on their positioning within the document. The system must be sufficiently robust to adapt to these variations while maintaining a high level of accuracy.