Optical Mark Recognition Techniques for Multiple-Choice Tests: A Comprehensive Review

Optical Mark Recognition (OMR) still plays a vital role in the automation of grading multiple-choice tests, providing efficiency and reliability in educational assessment all over the world. A detailed review follows the progress of OMR systems from the earliest hardware-dominated solutions, which required very strict conditions to work, up to current ones that utilize the power of deep learning. Conventional image processing techniques—such as thresholding, projection profiles, and template matching–have typically supplied fast and accurate results; however, their performance significantly worsens due to the imperfections of the real world including misalignment, noise, and ambiguous markings. The rise of classical computer vision techniques, particularly edge detection, contour analysis, and geometric corrections, has considerably enhanced the capabilities of OMR systems to deal with such problems. Today, modern deep learning–especially object detection architectures such as YOLO, and advanced convolutional networks-has brought a new era of flexibility and accuracy, even though it requires more annotated data and computing power. Nonetheless, the new challenges remain the same, regarding ambiguous marks, variability in form layouts, lighting inconsistencies, and the lack of large, publicly available datasets for benchmarking. This article, which critically synthesizes 25 representative papers, gives a detailed insight into the current OMR methods and their strengths and weaknesses, as well as outlining research gaps. This paper concludes by indicating promise for future research, including hybrid pipelines, domain adaptation, and the creation of robust, diverse datasets to support fair and reproducible benchmarking.