A behavioral model of writing

In this paper we propose a new model to generate handwriting based on behavioral patterns we believe is to be found in humans when imitating a written character. The proposed algorithm has a hierarchical structure. It is consisted of two main levels. At the first level the graphical features of the written letter to be imitated are extracted. These features are the directions of movement for each stroke. To extract the strokes that shape a letter, zero crossings of vertical and horizontal spatial velocity profiles are determined. At the second part a given letter is regenerated by using the extracted directions. At this level, the letter is linearly divided into several subdivisions. After that the excursion length (step) and the slope of line-segments producing the subdivisions are estimated. In each trial of learning, trajectory points are chosen with the estimated step and a random slope. The final trajectory is produced by successive arrangement of the strokes. In each trial the slope and step of the target point has a distance less than a specified threshold from the actual path is stored in memory and others are generated again. This process will continue during different trials until the trajectory can be generated using only memory. The resulted trajectories for different letters written by different subjects are qualitatively and quantitatively similar to the actual trajectories generated by human.