A Neuro-beta-Elliptic Model for Handwriting Generation Movements

Abstract

A neural network model for handwritten script generation is proposed, in which curvilinear velocity signals are approximated by the Beta profiles. For each Beta profile we associate an elliptic arc to fit the initial stroke in the trajectory domain. The network architecture consists of an input layer which uploads the set of Beta-elliptic characteristics as input, hidden layers and the output layer where script coordinates X(t) and Y(t) are estimated. A separate timing network prepares the input data. This latter involves the time-index starting time of each simple stroke for an appropriate handwriting movement signal. The experiments showed that the neural network model could be applied for the case of Latin handwriting scripts as well as Arabic handwriting scripts. New ways are proposed for the application of the neural network model such as: generation of complex handwriting movements, shape and character recognition.

A model-based approach to offline text-independent Arabic writer identification and verification

“…This model was exploited in the literature with good results for handwriting description [23,24], regeneration [25,26] and more recently OCR [27,28,29]. To the best of our knowledge, however, it has not been yet used for writer identification.…”

Online Arabic writer identification based on Beta-elliptic model

“…The Beta-elliptic model used strongly in many field of research for online handwriting, such as in Optical Character Recognition [24] and in regeneration of handwriting [25], [26]. The Beta-elliptic model has not yet used in online writer identification.…”

Calligraphic Stylisation Learning with a Physiologically Plausible Model of Movement and Recurrent Neural Networks

“…A number of methods have used neural inspired approaches for the task of handwriting trajectory formation. With a preprocessing step similar to ours, Ltaief et al [28] use a neural network to learn the mapping between the parameters of a handwriting model [6] and the corresponding trajectory. e network is then used instead of the model as a trajectory generator.…”