Line Segmentation of Handwritten Kannada Text

Abstract

Text line segmentation is an essential pre-processing
stage for off-line handwriting recognition in many Optical Character Recognition (OCR) systems. It is an important step, as inaccurately segmented text lines will cause errors in the subsequent recognition stage. Text line segmentation of handwritten documents is still one of the most complicated problems in developing a reliable OCR. The nature of handwriting styles makes the process of text line segmentation very challenging. This paper provides a means of text line segmentation using Vertical Strip Projection based method. In this paper, the document is divided into vertical strips of appropriate width. Strip-wise horizontal histograms are then computed and the relationship of the peak valley points is used for line segmentation.
Though the global horizontal projection method works efficiently for printed documents, it becomes highly unreliable when used in case of unconstrained handwritten text because of many factors like variable skew, touching and overlapping lines. The proposed method provides much improved and reliable results when compared to global horizontal profile method in case of skewed and variable sloping
lines. Segmentation accuracy of 83% is achieved for handwritten Kannada text using the proposed method.