Preliminary Results on Ancient Cham Glyph Recognition from Cham Inscription images

摘要

This paper presents an original work on ancient Cham glyph which is a language of Cham people in the Southeast Asia from the 6th to 15th century. Unfortunately it is in danger of being destroyed by times as well as of being ignored when the specialists of ancient Cham disappear. Motivated by this fact, we contribute to build a corpus of ancient Cham glyph that contains of 1607 images of Cham inscriptions. These images have been carefully pre-processed then annotated into 37 classes by a specialist/historian in ancient Cham. This digitized version of Cham glyph could be stored anywhere as long as wanted and available to wide audience for studying and exploration. As the corpus is the first introduced world wide, no work on automatic Cham recognition has been considered. We then investigate computer vision and machine learning techniques to show how effective a machine learning technique could be for this case study on a still limited amount of data. The best recognition F1-score is 86.3% with features extracted from GoogleNet and K-NN classifier, showing very promising performance.