A METHOD FOR AUTHENTICATION OF EXAMINEES DURING E-TESTING USING THEIR STYLES OF HANDWRITING

In this paper, we study an authentication method in e-test systems. E-Test is useful for examinees at distant location because it breaks down barriers of time and location. However, it is difficult to detect an unauthorized person taking the place of the examinee. In many current e-test systems, the examinees are authenticated at the beginning of the test by only an identifier and password. Then, systems cannot detect impersonation of an examinee during the test and e-test is not used for important examinations. To address this problem, we have proposed an authentication framework which authenticates the examinee when the test is being taken. In our framework, each examinee writes answers on a digital pen tablet. Dynamic information of writing of the answers obtained from the pen tablet can be used for the authentication without disturbing examinees during the test. The dynamic information is information on the writing process, such as x- and y-coordinates, pen pressure, and the angle and position of the digital pen. Most of online signature verification methods using dynamic information assume that the correct signature is registered in advance, and the methods compare the differences between the extracted characteristics of the registered and written signatures for authentication. In such comparison, the same characters are used. However, to develop an authentication method for any type of e-test, it is not realistic to collect dynamic information about all possible characters written by the examinees in advance, particularly in the case of Japanese, which has thousands of characters. Therefore, we have proposed an authentication method which decomposes each character to stroke segments and makes groups of the segments by the same stroke direction to reduce the number of characters to be registered.

In this study, we improve the authentication method using more detailed characteristics. We use three types of characteristics in each stroke of Japanese characters: stop, hook, and fade. Each examinee registers dynamic information of writing of some characters in advance, and after that the examinee takes an e-test and the information about the writing in answers is collected. Both the registered and answered data are broken down into sub-strokes and their characteristics in our method. We calculate the similarity between them. If the similarity is above the predetermined threshold, the examinee is judged to be the correct person. The calculation of the similarity is formulated as a mathematical programming model. The proposed method is evaluated by using answers given when examinees took an examination by e-testing. The results indicate that our method is applicable to the examinee authentication.