The digital documents quality control workflow at the BnF (operation, issue, improvement)

摘要

"Gallica" the digital library of the French national library is one of the most important digital library in the world. Today there are more than three million digital documents in Gallica. The indexation of its digital documents is done through its textual content obtained thanks to service providers that use Optical Character Recognition softwares (OCR). The robustness of the OCR systems is not always guaranteed since we find in its outputs some defects in the detection results of the document structure (segmentation results) and the recognition of the characters recognition results. One of the frequent errors in OCR outputs is the missed text components. The presence of such defects may leads to inconsistency in the digital libraries. We present in this paper the current document quality control workflow. Next, we propose a method that aims to detect the presence of missed text components without using ground truth. Our verification method uses the local information inside pages based on Radom transform descriptors and Local binary Patterns (LBP) to verify their OCR quality. The proposed approach has show a good performance since it can detect 84.15% of missed text components with a good precision rate 94.73%.