Text Line Extraction in Historical Documents Using Mask R-CNN

Text line extraction is an essential preprocessing step in many handwritten document image analysis tasks. It includes detecting text lines a and segmenting the regions of each detected line. Deep learning-based methods are frequently used for detection. However, only limited number tackle problems detection segmentation together. This paper proposes holistic method that applies Mask R-CNN extraction. A model trained to extract fractions from patches, which further merged form entire page. The presented was evaluated on two well-known datasets historical documents, DIVA-HisDB ICDAR 2015-HTR, achieved state-of-the-art results. In addition, we introduce new challenging dataset Arabic manuscripts, VML-AHTE, where numerous diacritics present. We show R-CNN-based can successfully segment lines, even such scenario.