Robust Heartbeat-based Line Segmentation Methods for Regular Texts and Paratextual Elements
Mathias Seuret, Daniel Stökl Ben Ezra, Marcus Liwicki · 2017
We have developed a simple, but powerful extension for two well known line segmentation methods which makes them more robust when working on historical manuscripts with almost regular line spacing. Against the intuitive impression that such manuscripts are easy to be handled, existing methods and tools fail to correctly segment some columns, mainly because of empty or nearly empty lines. Since historical documents frequently do have a regular occurrence of lines it is advisable to take this knowledge into account. From a literature review, our method seems to be the only one allowing to detect willingly empty lines between text lines, i. e., lines skipped by the scribe. Such paratextual information can contain immense importance for the understanding of the layout of documents. This heartbeat can be used for filtering out irregular candidates and finally bridging eventually resulting gaps. We tested our approach using the appearance heartbeat of the text lines on two well known line segmentation methods and show that it improves significantly the result quality and simplifies parameter tuning.