Binarization-free Text Line Extraction for Historical Manuscripts

Nowadays, large collections of old historical manuscripts, which contain valuable information about our cultural heritage, exist in libraries around the world. Recently, there has been much interest in their digitization for preservation reasons, since many of the available manuscripts’ quality has deteriorated from exposure to the environment. Digitization though is only the first step to make the information contained in manuscripts accessible to researchers and to the interested public. What we create after digitization is only a “digital image” of the page and further processing steps need to be applied during the handwriting recognition process, so that the manuscript’s content is transformed into a form that is interpretable by a computer. One important step in the handwriting recognition process is that of text line extraction, which aims at extracting individual text lines from the manuscript page. In this paper, we propose a binarization-free text line extraction method using seam carving [1]. The main idea is to compute an energy map of the input text blocks and determine minimum energy paths that pass through them. The energy map is constructed in a way so that gaps between text lines have low energy values. Therefore, a minimum energy path will pass only through these regions and will successfully separate two text lines. Our algorithm has the following two advantages: