A Model Based Text Line Segmentation Method for Off-line Handwritten Documents

Abstract

Text line segmentation is one of the important steps for offline handwritten text / handwriting recognition. This paper describes a novel method of text line segmentation based on the physical process of writing. The basic concept behind this segmentation method is: two successive handwritten text-line are always non-intersecting. The proof of the theory is explained with a model of pen-tip movement. The line segmentation is done by arranging the centroids of connected components present in the given text document. Experiments show that the proposed method achieves high accuracy for detecting unconstrained text line of handwritten documents.