A Study on the Verification Precision Improvement of Mobile Hand-written Signature

• Main Authors: CHAN-TE LIN, 林璨德
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/7mjyr2

碩士 === 大同大學 === 資訊經營學系(所) === 107 === This paper proposes a method for signing and verifying on mobile phones, tablets and computers. The system is expected to verify whether the signer is the right person. Using the system, we can sign on mobile devices, e.g., after a consumption using credit card, which can leverage the difficulty of KYC (know your customer). 　　The experiment is to perform a signature test to find the optimal verification, for example, an experiment of accept/reject, to confirm the judgment of the system on the given signature. The mobile device collects data such as the dynamic attribute XY coordinates, duration, stroke number as well as pressure if the mobile device can also collect pressure data for more precise comparison. 　　Based on the Facet Analysis Method proposed by Chen, this study calculates the average similarity of signatures of basic data set (of 3 signatures), and then uses the Euclid distance to determine the threshold (range) of signature similarity. Future signatures will be verified based on this basic data set (and its subsequent verified signatures). After evaluation using FAM, if the result exceed a given threshold, it is a fake. Otherwise, the new signature is accepted and its data is integrated into the basic data set. The signature verification is fully automatically, without human intervention. At the same time, the signature data is quantized and its feature can be pre-calculated in the background to lower the calculation workload of the signature verification. 　　This study is an extension of the work of Li, Improvement of Verification of Hand-written Signatures on Mobile Devices Based on Facet Analysis Method. Our system enables cross-platform sharing of data and higher discrimination. An APP is developed for dynamic signature verification; this is a further development of the previous system. The experimental results show that the Type II Error of false acceptance is 9.7% while the Type I Error of false rejection is 44%. It leaves a room for improvement in the future.