Unsupervised Optical Mark Recognition on Answer Sheets for Massive Printed Multiple-Choice Tests

Area of Science:

• Computer Science
• Image Processing
• Educational Technology

Background:

• Manual scoring of multiple-choice tests is time-consuming and prone to errors.
• Existing optical mark recognition (OMR) algorithms struggle with real-world variations in scanned answer sheets.
• Automating the scoring process for large-scale assessments is crucial for efficiency.

Purpose of the Study:

• To develop and evaluate a desktop application for optical mark recognition (OMR) of multiple-choice question (MCQ) answer sheets.
• To improve the speed and accuracy of scoring standardized tests.
• To provide a user-friendly interface for operators to manage and analyze exam results.

Main Methods:

• Compiled a dataset of 6029 scanned answer sheets (564,040 four-option answers) from exams administered in Tamaulipas, Mexico.
• Developed an image-processing module for answer extraction from digitized answer sheets.
• Created an operator interface for folder selection and results tabulation.

Main Results:

• The OMR system achieved 96.15% accuracy in grading entire exams without errors.
• The system correctly classified 99.95% of individual four-option answers.
• Automated scoring averaged 1.04 seconds per answer sheet, compared to 2 minutes for manual scoring.

Conclusions:

• The developed OMR desktop application offers a highly accurate and efficient solution for scoring multiple-choice tests.
• The system significantly outperforms manual grading in terms of speed and precision.
• This technology has the potential to streamline large-scale educational assessments.