A robust hybrid approach for text line segmentation in historical documents

Large-scale digitisation of historical documents demands robust methods that cope with the presence of frequent distortions and noisy artefacts. This paper presents a hybrid text line segmentation method that uses a novel data structure and a rule base to combine the strengths of top-down and bottom-up approaches while minimising their weaknesses. The effectiveness of the proposed approach has been methodically evaluated in the context of large-scale digitisation using a standardised framework. Results on a diverse dataset show improved performance over top-down and bottom-up approaches as well as over a leading commercially available system.