Developing A Web Application That Uses Image Processing To Automatically Score Multiple-Choice Test Answer Sheets

Abstract

This research develops a web application that uses image processing methods to automatically read and score standardized multiple-choice test answer sheets that are used by high schools in Kamphaeng Phet Province, Thailand.  This tool was built in five stages consisting of data collection, image resizing, image segmentation, morphological image analysis, and answer identification. The performance of the application was evaluated in two ways.  First there was a test to check the application’s ability to distinguish between types of answers on the answer sheet.   This old-style answer sheet requires students to write an X in a box corresponding to their chosen answer.  The four possible kinds of answers are:  crossing only one choice, crossing more than one choice, no cross at all, and crossing only one choice with a wrong choice hidden with correction fluid. The accuracy of the application was evaluated using 30 test sheets that were actual answer sheets produced by students.  These were visually checked and scored in advance.  In all cases where only one choice was crossed, the application accurately assessed all answers as correct or incorrect. The 30 test sheets contained 5 total instances of crossing more than one choice and 11 total instances of crossing no choice.  The application accurately assessed all these responses as incorrect.  There were also 23 total instances of liquid paper covering a choice.  In all 23 cases, the application accurately ignored the choice covered in liquid paper and correctly evaluated any other markings present.  In summary, the test showed that the app always evaluated responses correctly and was never confused by non-standard conditions, thus performing with an accuracy of 100%.  This web application can be relied on to correctly score standardized multiple-choice test answer sheets.