Handwriting recognition using sinusoidal model parameters

• Published in: Pattern recognition letters Vol. 121; pp. 87 - 96
• Main Authors: Choudhury, Himakshi, Prasanna, S.R. Mahadeva
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 15-04-2019; Elsevier Science Ltd
• Subjects: Character recognition; Feature extraction; Handwriting; Handwriting recognition; Internet; Language; Markov chains; Modelling; Online handwriting recognition; Oscillations; Parameters; Pattern recognition; Sinusoidal model; Support vector machines; UNIPEN; Velocity; Velocity distribution; Word recognition; Online handwriting recognition; UNIPEN; Word recognition; Sinusoidal model
• ISSN: 0167-8655; 1872-7344
• DOI: 10.1016/j.patrec.2018.05.012

•Analysis of sinusoidal model parameters for handwriting recognition.•Modeling of acceleration and position information using sinusoidal oscillations.•Proposal of sinusoidal features for character and word recognition system.
Handwriting is produced by the oscillatory motion of the hand in both horizontal and vertical directions, with a constant drift velocity along its writing direction. The velocity profiles of handwriting in these orthogonal directions have an invariant bell-shaped nature with zero velocities at the change in trajectory directions. According to recent studies, online handwriting can be suitably characterized by modeling the velocity profiles with sinusoidal oscillations. In this paper, we propose a set of features derived from sinusoidal modeling of handwriting velocities for online handwriting recognition (HR) task. Although the predominant information of handwriting is modeled in the velocity profiles, the first derivatives of the velocity profiles (i.e. acceleration) and the x- and y-coordinates are also important in characterizing the handwriting. Accordingly, these signals are also modeled by sinusoidal oscillations, and the parameters are utilized as features to develop the HR system. As these parameters are extracted directly using the hand movement generation theory, therefore it may also contain additional information describing the generation of the pattern along with its spatial shape information. The efficacy of the proposed features is shown for character and word recognition task employing hidden Markov model (HMM) and support vector machine (SVM) classifiers. The experiments are conducted on three online handwritten databases: Assamese digit database, UNIPEN English character database and UNIPEN ICROW-03 English word database. The results obtained are promising over the prior works for these databases.