A Hybrid Approach Combining YOLO and Visual Image Transformers for Automated Mark Digitization of Exam Scripts

In many educational institutions, manually entering marks into spreadsheets is a tedious process prone to errors, leading to inefficiencies and inaccuracies. This paper presents a hybrid approach to automate the digitization of examination marks directly from scanned exam scripts, reducing manual effort and improving accuracy. The system integrates YOLOv9 for precise table detection and localization and Visual Transformers for robust digit recognition, ensuring structured mark extraction. The retrieved marks are automatically stored in a question-wise organized Excel sheet, enabling seamless academic record management. To handle variations in handwriting styles, paper quality, and lighting conditions, the model is trained on a custom dataset with data augmentation techniques to enhance robustness. Experimental results demonstrate that this system significantly improves efficiency and accuracy, offering a practical and scalable solution for automating examination evaluation in educational institutions.