A Dataset of Vietnamese Documents for Text Detection

Document analysis and recognition is a crucial technique for automating the input process of forms, receipts, documents at banks, governments, companies. With demands in both research and industry, there are available datasets for Document Analysis and Recognition in English, Chinese, Arabic, and Indic. However, there is no publicly datasets for Vietnamese Document Analysis and Recognition. In this paper, we introduce a new dataset for Vietnamese Document analysis named VNDoc, which aims to set up a standard dataset for researching and developing Vietnamese Document Analysis Systems. The dataset contains 226 documents scanned from mobile phones and scan machines. The documents are collected from diverse categories such as legal and administrations, invoices, resumes, handwriting forms, and so on, which target various applications. At the first stage, we provide ground truth for text lines, which allow performing research in text detection and layout analysis. Moreover, we describe a statistical analysis of text length and bounding box in the dataset and initial experiments for the existing methods for text detection. We are going to provide text transcriptions and available for research communities.