Globd: Geometric And Learned Logic Algorithm For Straight Or Curved Handwriting Baseline Detection

This paper presents a developed geometric and logic algorithm of on-line Arabic handwriting baseline detection. It consists of two stages: the geometric first stage detects sets of nearly aligned points candidates to support the baseline by considering the accordance between the alignment of the trajectory points and their tangents directions. While the logic second stage uses topologic conditions and rules specific to the Arabic handwritten script in order to evaluate the relevance of each one of the three most extended sets of points from the extracted groups to be recognized as a baseline and then to correct the first stage detection result which is based only on the size of the group of points. The system is also designed to be able to extract the baseline of inclined and/or irregular aligned short handwritten sentence thanks to the flexibility of the used method for the constitution of sets of nearly aligned points. The iterative application of this last method in a relatively short neighborhood window sliding on a long and curved handwritten line script permits to extract its curved baseline.