[4L3-GS-10] AI application: machine learning

[4L3-GS-10-04] Online handwriting recognition based on LSTM using signature prediction stroke feature

〇Yuri Yoshima¹, Jun Rokui¹ (1. University of Shizuoka)

Keywords:Long short-term memory, Online Handwriting Recognition, Stroke Signature Authentication

Recently, information technology such as Internet, computer, and portable telephone has rapidly developed. The biometric authentication which does not require the possession and memory of keys, passwords, cards, etc. is highly secure. In particular, handwriting authentication is highly acceptable and is expected to become widespread from the point that the registration information can be changed. Traditionally, in handwriting authentication using handwriting stroke, it is necessary to write off the signature of the identification object to the end, which is a delay cause in real-time processing. In this research, we propose a method that can be identified in the middle of process without going through the process of all handwritten strokes using long-term memory (LSTM), which is widely used for time-series prediction. This research can hope to improve both convenience and authentication accuracy in handwriting authentication.