Deep Learning based Ancient Literature Recognition and Preservation

Abstract

This paper introduces an deep learning based ancient literature recognition method for culture heritage preservation by deep learning. The model of Single Shot Multibox Detector is used for detecting and recognizing the ancient characters. This paper only focuses on the recognition of Oracle Bone Inscriptions which is the one of the oldest and most mysterious ancient characters, used about 3000 years ago in china. The experimental results show that Precision achieves and Recall achieves 0.86 and 0.97, respectively, and prove the effectiveness of Single Shot Multibox Detector in ancient characters recognition. By analyzing the experimental results, we found that in the case of heavy blurred image or tilted characters, the Single Shot Multibox Detector can not achieve a exciting results. Hence, we apply the pre-processing of binarization, changing brightness and contrast, and rotation in the test images, and then achieve accuracy improvement of recognition by using pre-processed images.