Recognition of on-line handwritten digits by neural networks using circular and beta approaches

On-line handwritten digit recognition based on trajectory and velocity modeling

“…The strokes number was based only on velocity signal extremum. Local extremum were not considered in (Bezine et al, 2003band Kherallah et al, 2002; consequently, the weakness of this approach is that it cannot detect the smallest handwritten trajectory curvature or the catastrophe variability of the handwritten trajectory caused by involuntary psychological behavior of the writer. However, in this paper we proceed to locate the inflexion points of the handwritten trajectory.…”

“…Previously, modeling systems were mainly based on either trajectory or velocity features. Kherallah et al (2002Kherallah et al ( , 2004 present the first idea of combining the kinematics and geometry in the trajectory modeling. These studies considered neither the overlap of Beta signal nor the inflexion points which present a key role for the strokes number determination.…”

“…In (Kherallah et al, 2002) the modeling system was based on Beta-circular approach. The circular strokes are superposed but there are not of overlapping shapes and cannot perfectly reconstruct the handwritten trajectory.…”

“…The Beta-elliptic representation is consisting of a combination between geometry and kinematics in handwriting generation movements (Bezine et al, 2003b;Kherallah et al, 2002Kherallah et al, , 2004Viviani and Schneider, 1991). According to this method of modeling, the number of features per character can reach 63 -dimensional feature vector.…”

“…These models are probabilistic and need powerful calculators and a considerable calculation time. However, in the structural method, a set of basic strokes are usually selected as primitives (El-Sheik and El-Taweel, 1990;Hafsa et al, 2004;Heutte et al, 1998;Kherallah et al, 2002Kherallah et al, , 2004Morasso et al, 1993;Simard et al, 1993), and stroke recognition is based on the use of certain geometrical features like line segment directions, strokes length, strokes order, strokes number, strokes relation, etc. These are also found to be useful for character recognition.…”