TY - GEN
T1 - Improving Online Handwriting Recognition with Transfer Learning Using Out-of-Domain and Different-Dimensional Sources
AU - Lee, Jiseok
AU - Akiba, Masaki
AU - Iwana, Brian Kenji
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
PY - 2025
Y1 - 2025
N2 - Online handwriting recognition is a widely used technique in our daily lives. Furthermore, deep learning has become one of the most popular and influential methods for online handwriting recognition. However, artificial neural networks typically require massive datasets. Transfer learning is a standard method to overcome the problem of lack of data. Usually, transfer learning works by initiating a network with trained weights and fine-tuning with a smaller dataset. Still, obtaining large amounts of online handwriting can be difficult for pre-training networks. Therefore, we propose pre-training with data sources with dimensions different from handwriting. Namely, we propose using univariate or multivariate data as a source dataset for two-dimensional target data by embedding out-of-domain time series of different dimensions into two-dimensional space. We evaluated the proposed method with four handwritten character datasets: a numerical digit dataset, an uppercase alphabet dataset, a lowercase alphabet dataset, and a Chinese character dataset. Through the evaluation, we demonstrate that transfer learning from datasets with a different dimensionality as online handwriting is possible.
AB - Online handwriting recognition is a widely used technique in our daily lives. Furthermore, deep learning has become one of the most popular and influential methods for online handwriting recognition. However, artificial neural networks typically require massive datasets. Transfer learning is a standard method to overcome the problem of lack of data. Usually, transfer learning works by initiating a network with trained weights and fine-tuning with a smaller dataset. Still, obtaining large amounts of online handwriting can be difficult for pre-training networks. Therefore, we propose pre-training with data sources with dimensions different from handwriting. Namely, we propose using univariate or multivariate data as a source dataset for two-dimensional target data by embedding out-of-domain time series of different dimensions into two-dimensional space. We evaluated the proposed method with four handwritten character datasets: a numerical digit dataset, an uppercase alphabet dataset, a lowercase alphabet dataset, and a Chinese character dataset. Through the evaluation, we demonstrate that transfer learning from datasets with a different dimensionality as online handwriting is possible.
KW - Online Handwriting Recognition
KW - Time Seires Recognition
KW - Transfer Learning
UR - http://www.scopus.com/inward/record.url?scp=85212260174&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85212260174&partnerID=8YFLogxK
U2 - 10.1007/978-3-031-78119-3_5
DO - 10.1007/978-3-031-78119-3_5
M3 - Conference contribution
AN - SCOPUS:85212260174
SN - 9783031781186
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 61
EP - 75
BT - Pattern Recognition - 27th International Conference, ICPR 2024, Proceedings
A2 - Antonacopoulos, Apostolos
A2 - Chaudhuri, Subhasis
A2 - Chellappa, Rama
A2 - Liu, Cheng-Lin
A2 - Bhattacharya, Saumik
A2 - Pal, Umapada
PB - Springer Science and Business Media Deutschland GmbH
T2 - 27th International Conference on Pattern Recognition, ICPR 2024
Y2 - 1 December 2024 through 5 December 2024
ER -