Application of autoregressive models to the study of the temporal structure of a handwritten text

Abstract

Variations of the slope of handwriting in several hundreds of lines are predicted by using autoregressive models. Most of the unpredictable lines correspond to features such as the lack or excess of ink, the beginning of pages or paragraphs. The temporal succession of models is statistically analyzed in order to define descriptors, such as the transition matrix, that account for the building of a further state-space model for handwriting.