A new method of mark detection for software-based optical mark recognition

Area of Science:

• Computer Science
• Data Science
• Survey Methodology

Background:

• Software optical mark recognition (SOMR) converts survey data into machine-readable formats.
• Existing SOMR methods struggle with inaccurate mark detection due to incomplete darkening, internal labels, soft pencils, and print/scan artifacts.
• High error rates in SOMR limit its application in scientific research.

Purpose of the Study:

• To introduce a novel mark detection method for SOMR.
• To improve the accuracy and robustness of SOMR, particularly in the presence of form imperfections.
• To provide a SOMR solution suitable for scientific applications requiring high precision.

Main Methods:

• Developed a new mark detection algorithm that surpasses traditional pixel counting and simple thresholding techniques.
• Evaluated the method's performance using field testing with both trained and untrained respondents.
• Compared the new method against existing techniques on both clean and artifact-laden forms.

Main Results:

• The new SOMR method demonstrated exceptional accuracy, with sensitivity, specificity, and accuracy rates of 99.73%, 99.98%, and 99.94% respectively in field tests.
• Achieved superior performance over pixel counting and simple thresholding, especially on forms with print and scan artifacts.
• Showcased near-perfect detection rates (e.g., 100% sensitivity, specificity, and accuracy) on forms with print artifacts, outperforming other methods significantly.

Conclusions:

• The novel SOMR method offers a significant improvement in mark detection accuracy and reliability.
• This technique is robust against common form imperfections, making it ideal for scientific data collection.
• The method is specifically designed for bubble and box fields, providing a reliable tool for accurate survey data processing.