Preprints are manuscripts made publicly available before they have been submitted for formal peer review and publication. They might contain new research findings or data. Preprints can be a draft or final version of an author's research but must not have been accepted for publication at the time of submission.
posted on 16.10.2020by Besma Rabhi, Abdelkarim Elbaati, Houcine Boubaker, Adel M. Alimi
The online signal is rich in dynamic features such as trajectory chronology, velocity, pressure and pen up/down. Their offline counterpart consists of a set of pixels. Thus, the online handwriting recognition accuracy is generally better than the offline one. In this paper, we propose an original framework for recovering temporal order and pen velocity from offline handwriting. Our framework is based on sequence to sequence Gated Recurrent Unit (Seq2Seq GRU) model. The proposed system consists in extracting a hidden representation from an image using Convolutional Neural Network (CNN) and Bidirectional GRU (BGRU), and decoding the encoded vectors to generate dynamic information using BGRU. We validate our framework by an online recognition system applied on Latin, Arabic and Indian On/Off dual handwriting character database. To prove the performance of the proposed system, we achieve a low error rate of point coordinates and a high accuracy rate of the LSTM recognition system.