Line segmentation of handwritten text using histograms and tensor voting

Abstract

There are a large number of historical documents in libraries and other archives throughout the world. Most of them are written by hand. In many cases they exist in only one specimen and are hard to reach. Digitization of such artifacts can make them available to the community. But even digitized, they remain unsearchable, and an important task is to draw the contents in the computer readable form. One of the first steps in this direction is to recognize where the lines of the text are. Computational intelligence algorithms can be used to solve this problem. In the present paper, two groups of algorithms, namely, projection-based and tensor voting-based, are compared. The performance is evaluated on a data set and with the procedure proposed by the organizers of the ICDAR 2009 competition.

Direct Tensor Voting in line segmentation of handwritten documents

“…Later contests, such as those at ICFHR 2010 [45] and ICDAR 2013 [46], were also conducted, but they used more challenging data-sets for segmentation. In this study, we chose the simplest data-set to facilitate comparison with various methods, including the approach outlined by [19] and our earlier work [10].…”

“…Horizontal or vertical projection is commonly employed for segmenting lines or words. Both older publications like [7], [8] and newer papers like [9], [10] employ these methods. Some algorithms also utilize piece-wise projection profiles, as seen in [11], [12], which can effectively handle instances of overlapping or slanted lines in handwritten documents.…”