Textline Detection Dataset

In document layout analysis (DLA), the detection and classification of multiple classes of layout elements such as headings, images, tables etc. is required in order to extract and process them in subsequent steps. This proves to be a challenging task in digitised cultural heritage items such as books, newspapers and archival records. In order to train a convolutional neural network capable of recognising text lines, a dataset was created. The machine learning task for which this dataset was collected falls into the domain of image segmentation and, more generally, of computer vision. The dataset was compiled by Vahid Rezanezhad within the research project “Mensch.Maschine.Kultur – Künstliche Intelligenz für das Digitale Kulturelle Erbe” at the Staatsbibliothek zu Berlin – Berlin State Library (SBB). The research project was funded by the Federal Government Commissioner for Culture and the Media (BKM), project grant no. 2522DIG002. The Minister of State for Culture and the Media is part of the German Federal Government.