Periodical Examinees Identification in e-Test Systems Using the Localized Arc Pattern Method

In this paper, we study an identification method for use in e-Test systems. The e-Test is a tool for assessing knowledge of examinees over the World Wide Web. This type of test is useful for examinees at distant locations. However, some drawbacks of the e-Test have been pointed out, especially when used for strict tests such as the entrance examination tests of universities. One of the most critical drawbacks is the difficulty in confirming examinee identification. Many e-Test systems only ask for the password to confirm identification at the start of the test. Nevertheless, it is difficult to determine impersonations of examinees during the test at remote locations. To overcome this drawback, we propose here an e-Test system that confirms the examinee identification periodically during the test. Considering that examinees write their answers using pen tablets in e-Test systems, we can use writer identification of the answers for confirmation without disturbing the examinees during the test. We apply the localized arc pattern method (Yoshimura & Yoshimura 1991), conventionally used for off-line identification based on the arc of an individual stroke of a certain sentence on paper, for examinee identification when using pen tablets. The outline of the original method has three steps. First, each character on paper is scanned into a computer and converted to binary images. Next, we count the numbers of matching binary patterns of the images with predetermined numbers. Finally, the writer is identified by comparing the numbers statistically with those of the registered data. This method can identify writers based on signatures in most cases. Since the original method was investigated in terms of characters written on paper, we first examined whether the method can be applied to characters written on pen tablets. We used characters written by 4 persons, where each person writes 48 sentences with 25 characters each. We take the first 8 sentences as the registered data set for each person and the other sentences remain as the test data set. The two data sets are compared in order to identify the writer. Our experimental results show that approximately 80% of the test data accurately identified the writer. To improve the percentage, we consider a combination of the proposed method and writing characteristics such as pen pressure.