A Spiking Neural Network Model for Complex Handwriting Movements Generation Mahmoud Ltaief #1 , Hala Bezine #2 , Adel M. Alimi #3 # REGIM-Lab.: REsearch Groups in Intelligent Machines, University of Sfax ENIS, BP 1173, Sfax, 3038, Tunisia 1 mahmoud.ltaief@ieee.org 2 hala.bezine@ieee.org 3 Adel.Alimi@ieee.org Abstract—In this paper a spiking neural network model for online complex handwriting movement generation is proposed. Online handwriting is described as the superposition of strokes with the elliptical shape which is the result of the algebraic sum of the beta profiles. Handwriting can be partitioned into simple strokes. Each one is fully modeled by a set of ten parameters which characterize the handwriting in both the kinematics and the static fields. The network is composed of an input layer which uses a set of Beta-elliptic parameters as input, a hidden layer and an output layer dealing with the estimation of the script coordinates X(t) and Y (t). An additional input is used as a timing network to prepare the input parameters. This later, acts as starting pulse of each stroke belonging to a given handwriting script. The simulation results showed that the spiking neural network model could generate both Latin and Arabic handwriting scripts. Similarity degree is measured between original scripts and generated scripts to evaluate our model. The proposed spiking neural network model can be applied in new ways such as: signature verification and shape recognition. Index Terms—Spiking neural network, Beta-elliptic model, Handwriting generation, Similarity degree. I. I NTRODUCTION Writing is considered among the fastest motor activity and the most complex of our directory engine. This activity applies a coordination of several joints and muscles to generate a succession of graphic shapes quickly and suitably precise to be known [33]. The writer begins with the intention of writing a message (semantic level), that transforms into words (syntac- tical and lexical level). When the single letters (graphemes) are recognized, the writer chooses the specific letter shape (allograph). The choice is based on selection syntax of formal allograph, random selection or personal preference [31], [32]. Then, allographs are converted into movement patterns, that is the subject of this work. Handwriting models: We will now browse some of the prin- cipal model of handwriting, in particular focus on those which inspired the development of the suggested model. There are two major types of handwriting models in literature appearing to be adopted by researchers in the past [35]. The first one, called the ”bottom-up” models, also known as computational models concern peripheral and biophysical characteristics of the hand writing generation [26]. The main motivation of these models is to describe how the features are generated from motor regularities in handwriting movements and bio- physical systems effectors, they do not claim to be faithful to the underlying processes neuromotor writing process [26], [19]. The second method of modeling of writing focuses on psychologically descriptive models [9], [8]. These ”top-down” models, also called neurocognitive models aim to describe the cognitive processes that provide the engines and language goals for the training of handwriting trajectory [11]. This work is inspired from the category of “top-down”. Hollerbachs oscillation theory is one of the most important models of handwriting. This model is based on the principle that the stroke data can be solved into oscillatory elements by a decomposition of Fourier transform. This theory was appeared with Hollerbach [17] who presented a perceptive handwriting generation model in which the hand-pen system is described by two perpendicular pairs of opposed springs acting upon an inertial force. It was emphasized that the oscillations of the natural movements of this system look like real handwriting segments. Anatomical substantiation for such a simple system has also been investigated [17]. Schomaker [32] proposed a neuronal model; in this model an oscillatory network generates horizontal and vertical pen movement. For training of the network Schomaker uses a delta-rule variation, this variation has led to doubtful out- comes: the network performance has depended strongly on network settings. Despite of the gaps of the model perfor- mance, the work of Schomaker elucidated clearly certain re- lated questions with any model of handwriting. Consequently, the process of handwriting should have four fundamental steps at the same time in handwriting shaping and chaining [32]: 1. Configuration of the model: this step is differently known as coordinative structure gearing, engine programming, planning and preparation. 2. Startup of the script: after having configured the model for the task to achieve, there has to be a release signal of the script at the right time. 3. Script execution: the duration of this step and the measures which are taken depend on features like the distance from a International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 7, July 2016 319 https://sites.google.com/site/ijcsis/ ISSN 1947-5500