Historical Typewritten Document Recognition Using Minimal User Interaction

Recognition of low-quality historical typewritten documents can still be considered as a challenging and difficult task due to several issues i.e. the existence of faint and degraded characters, stains, tears, punch holes etc. In this paper, we exploit the unique characteristics of historical typewritten documents in order to propose an efficient recognition methodology that requires minimum user interaction. It is based on a pre-processing stage in order to enhance the quality and extract connected components, on a semi-supervised clustering for detecting the most representative character samples and on a segmentation-free recognition stage based on a template matching and cross-correlation technique. Experimental results prove that even with minimum user interaction, the proposed method can lead to promising accuracy results.