| المستخلص: |
Nowadays, many computer-based grading of multiple-choice exams are available worldwide. Despite this strive towards automating multiple-choice exams, pencils and papers are still widely used especially when the number of examinees submitting the exam exceeds the number of computers. Manual-grading for a large number of multiple-choice exams is a tiresome and an error prone process. Therefore, this is a field where machines would prove more capable than humans. Employing an automated system for grading reduces grading time, human effort, as well as reducing the probability of grading errors. In this work, a novel approach for developing an automatic, efficient, fast and offline fixable system is introduced that can be used for grading scanned multiple-choice exams. This system guarantees high accuracy based on a similarity matching algorithm. The scanned answer sheets are processed by robust algorithms for noise removal and feature extraction. After which they are passed through edge and line detection, right answer detection, and scores accumulation. We focused on multiple marked cases of irregularities that are observed by exam takers and accounted for them in our system; whereas existing automatic exam grading systems failed to address them. The proposed grading system relies on four voting functions for similarity measurement. In addition to providing an automatic grading for each answer sheet, a confidence ratio is also provided as well as list the questions that require manual inspection. This proposed grading system does not require any special hardware or any special requirements. The provided experimental results are based on actual exam sheets from several semesters and various graduate and undergraduate courses and diverse student levels showing superior performance compared to existing solutions.
|
