Representation of online handwriting using multi-component sinusoidal model. (July 2019)

• Record Type: Journal Article
• Authors: Choudhury, Himakshi; Prasanna, S.R. Mahadeva
• Is Part Of: Pattern recognition. Volume 91(2019:Jul.)
• Page Start: 200
• Page End: 215
• Publication Date: 2019-07
• Subjects: Sinusoidal/oscillatory model -- Handwriting synthesis -- Sinusoidal features -- Handwriting recognition; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2019.02.013
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Analysis of sinusoidal model for handwriting generation. Proposal of multi-component sinusoidal model for representing online handwriting. Exploration of sinusoidal parameters as features in handwriting recognition system. Proposal is evaluated for signature reconstruction and its synthetic variation. Abstract: The representation of online handwriting is an important aspect of handwriting applications, which involves the extraction of various spatial and temporal attributes for analysis and individualization of handwritten patterns. In this work, a model based representation is proposed for online handwriting using a multi-component sinusoidal model. The method extracts sinusoidal parameters from handwriting by modeling its horizontal and vertical velocities between each successive pair of zero crossing points with a half period of the sine function. Thus, each velocity profile is represented by the sinusoidal oscillations whose parameters are modulated at the zero-crossing points. The use of multiple oscillations to model the velocities results in a better representation of the complex trajectories. The parameters of the proposed model are computed iteratively from its residual signals. We hypothesize that the analysis of the sinusoidal components and its parameters may provide added dynamic information about the handwriting. The efficacy of the proposal is demonstrated for online signature representation and synthetic variability generation by modifying the extracted parameters. Further, the proposed feature set is also employed for online handwriting recognition task. It is observed that the multi-component sinusoidal representation combined with existing point-based features provide an improvement in the recognition performance.