A saliency-based multiscale method for on-line cursive handwriting shape description

We propose a method derived from an analogy with the primate visual system for selecting the best scale at which the electronic ink of the handwriting should be described. According to this analogy, the method computes a multiscale features maps by evaluating the curvature along the ink at different levels of resolution and arranges them into a pyramidal structure. Then, feature values extracted at different scales are combined in such a way that values that locally stand out from their surrounds are enhanced, while those comparable with their neighbours are suppressed. A saliency map is eventually obtained by combining those features value across all possible scales. Such a map is then used to select a representation that is largely invariant with respect to the shape variations encountered in handwriting. Experiments on two data sets have shown that simple algorithms adopting the proposed representation lead to very stable stroke segmentation and feature matching.