Abstract:
Paper grading is a repetitive and time-consuming task that can be automated to allow academics to focus on more productive activities like teaching and research. While digital MCQ quizzes can be graded instantly, most exams still rely on paper-based answer sheets. To automate the grading process, this paper presents an Automated Paper Grading System for Multiple-choice questions (MCQs) that combines a high-accuracy computer vision model with a user-friendly web application. This system enables users to upload answer scripts and model answers, automatically grade them, and ultimately visualize the results. After uploading the answer sheet, the system processes the image through several computer vision techniques to extract and evaluate responses. First, the image is converted to grayscale, followed by edge detection to identify the sheet's corners. Contour detection then locates the largest rectangles containing student responses. Warp perspective transformations are applied to each column to ensure accurate alignment. Finally, a pixel threshold method determines the marked answer by identifying the highest white pixel density within each section. The web application is built using PostgreSQL, Express, React, and Node (PERN stack) and deployed on Google Cloud. Further, it utilizes Python and OpenCV to process scanned MCQ sheets by identifying answers marked in colour using pixel value analysis, leveraging computer vision technology. The study demonstrates that this system achieved 100% grading accuracy with the guidelines followed under standard conditions, taking an average of around 30 seconds to grade 250 answer scripts. Hence, the system ensures precision and efficiency offering a reliable solution for managing large batches of assignments compared to manual grading. Future improvements include extending the system’s capabilities to evaluate textual and essay-type answers, further reducing manual intervention.
