With the advancement of intelligent educational assessment technologies, automated paper grading systems are playing an increasingly important role in improving the efficiency and accuracy of exam evaluation. However, traditional grading methods still face challenges in objective question recognition, subjective question assessment, and data management. To address these issues, this study proposes an intelligent paper grading method based on computer vision. This method enables the recognition and result output of objective question answer sheets, the identification of grading traces and score determination for subjective questions, and provides functions for exam data statistics and management. Innovatively, Faster R-CNN is combined for score recognition in subjective questions to enhance the intelligence level of grading. The experimental results show that this method achieves a recognition accuracy of over 99% in automatic grading tasks for objective questions, and an mAP@0.5 of above 0.85 for score recognition in subjective questions. This significantly improves grading efficiency and reliability, providing strong support for the development of intelligent educational assessment systems.