BRESSAY: A Brazilian Portuguese Dataset for Offline Handwritten Text Recognition

BRESSAY: A Brazilian Portuguese Dataset for Offline Handwritten Text Recognition

연구 분야: Databases

학회: International Conference on Document Analysis and Recognition

This work introduces the BRESSAY dataset, a novel contribution to the field of offline Handwritten Text Recognition (HTR), specifically targeting Brazilian Portuguese. Despite significant advancements in HTR, challenges remain due to the variability of human handwriting. This is particularly evident in academic contexts, where diverse handwriting styles are common, often influenced by external factors such as time constraints and pressure. In this context, the BRESSAY dataset provides a rich and complex environment that reflects these challenges, making it a valuable resource for developing robust optical models. It includes a unique collection of student essays with special annotations such as superscript and subscript texts, various forms of illegible or crossed-out text, erasures, and a broad range of writing styles. In our research, we conducted an exploration using well-established optical models for line-level recognition. The initial study was designed to evaluate the dataset’s utility in enhancing model performance, with the aim of establishing a baseline for future research in this area. The dataset is expected to significantly contribute to the HTR community by aiding in the automated transcription and analysis of handwritten educational materials. Consequently, our work not only adds a valuable resource to the HTR research domain but also opens up possibilities for innovations in the field of automated assessment based on handwritten documents. The BRESSAY dataset is available for download in the repository (https://github.com/arthurflor23/handwritten-text-recognition).