New Preprocessing Techniques for Handwritten Word Recognition

The research described in this paper focuses on the presentation of two novel preprocessing techniques for the task of off-line handwritten word recognition. A technique for the identification of straight and skewed underline noise is described along with a novel algorithm for detecting skew in handwritten words. The latter identifies skew by detecting the center of mass in each half of a word image. By hypothesizing a line between the two centres and by measuring the angle it makes with the x-axis, an angle for skew may be estimated. The algorithms are tested on the CEDAR benchmark database of handwritten cursive words. Results above 96% are reported for skew detection and underline removal.