Unsupervised Optical Mark Localization for Answer Sheet Based on Energy Optimization
Chenbo Shi, Junsheng Zhang, Jie Zhang, Chun Zhang, Xiangteng Zang, Lei Wang, Changsheng Zhu · 2023
Optical Mark Recognition (OMR) is a crucial automated data entry technique in industries such as education, finance, and consulting. However, the localization of optical marks in diverse scenarios remains challenging. Aiming at the problems of diverse types of optical marks, irregular printing quality, and irregular array layout in various examinations, questionnaires, and homework scenarios, this paper presents the first proposal of the energy optimization algorithm for accurate optical mark localization. The algorithm introduces a Single Mark Localization Model (SMLM) within an unsupervised framework. By considering pixel distribution, regional pixel proportion, and aspect ratio, the SMLM achieves precise localization of optical marks. In the experimental evaluation of over 18,000 answer sheets across 82 different marking symbols and layouts, a precision rate of 99.82% and a recall rate of 99.03% were achieved. The results show that the algorithm has strong robustness and versatility. Thus, our study significantly advances the field of optical mark recognition by providing a practical solution to the optical mark localization problem.