The researchers at China’s Southwest University and Capital Normal University of Beijing together have recently applied a multi-regional convolutional neural network (CNN) to study oracle bones.

Oracle bones are the earliest evidences of China’s recorded history in form of the oldest known ancient Chinese scripts. The oracle bones are typically made from ox scapulae or tortoise/turtle shells and were mainly used for divination. However, most surviving oracle bones are chipped and broken pieces that are preserved at the Yinxu Museum in Henan Province. To study and interpret these oracle pieces, written in Yi characters, is like putting a complex puzzle together from thousands of oracle rubbings. Apart from the inscription itself, the material also matters to ensure that wrong pieces such as ox bones and tortoise shells aren’t assembled together as it is vital for the recovery of a complete oracle bone script. Currently, such classification is an arduous task, and completely depends on the expert’s experience. The classification of tortoise shells and animal bones is ascertained mainly by the two unique characteristics to be found on tortoise shells - shield grain and tooth grain. The task is extremely challenging because the thousands of years’ worth of wear and tear on the bones. Also, to become an oracle expert one needs to do long-term studying and a lot of professional knowledge.

The team’s lead, China’s Southwest University Associate Professor Shanxiong Chen, told Synced, “Our team, as part of the Southwest University’s Database and Artificial Intelligence Laboratory, is particularly interested in using deep learning theories and methods in the field of image processing to identify, classify and hopefully assist in repairing ancient Chinese books and documents.” Chen’s team has developed a CNN framework that recognises Yi characters, even when the scripts are smudged or incomplete. The researchers use CNN’s classification and recognition model to study the oracle bone rubbing’s scanned images by diving the data into three local regions - shield grain region, tooth grain region and non-shield grain region. These corresponded to a feature extraction subnet consisting of two Conv-Pooling-ReLU layers and two fully connected layers, that extracts the function of each local region. Since the features of each local region are composed of vectors, the researchers then applied a multi-feature fusion subnet made up of four Auto-Encoding layers to fuse these features and reduce the region size to obtain fusion features before the final output.

To study the surfaces, the researchers used over 1400 tortoise shells and 300 ox bone rubbings as a dataset - two-third were used as training sets and one-third as a test-sets. So far, the results show accuracy similar to that of the oracle experts. “As I said, classification is the first step,” Chen explained. “This study specifically focused on telling between animal bones and tortoise shells, and we’re continuously working with Capital Normal University’s Center for Oracle Bone Studies on further classifying different types of animal bones.”

The team has future plans. They intend to build models for oracle bone conjugation that will help present complete oracle bone scripts for experts to interpret messages.