著者

今野 慎介

出版者

函館工業高等専門学校

雑誌

函館工業高等専門学校紀要 (ISSN:02865491)

巻号頁・発行日

vol.50, pp.35-42, 2016

---

Recently, to reduce the inconvenience caused by authentication operations in portable terminals such as smartphones, various authentication methods based on behavior characteristics have been studied. We proposed authentication method which identified individuals based on walking motions measured by wearable sensors such as acceleration sensors, and can be implemented on a smartphone (as smartphones are equipped with sensors). In the many case of multimodal biometrics authentication with machine learning, the common classifier for all users is generated. One of the reasons for this is that these authentication systems are installed into the equipment shared by users, such as the server computer. This server computer identifies the all authorized persons and the non-authorized persons. However, each smartphone is occupied by each user and is not shared by the multiple users. Therefore, the classifier to each user for fusing the features can be generated for each user in their smartphones. The purpose of this paper is to reveal the influence by generating the individual classifier for each user and a common classifier for all users. We evaluated the authentication accuracy by employing both the classifiers generation methods.