Similarity Queries on Script Image Databases

Abstract

For a long time, researchers and archaeologists have studied evolution and similarity among ancient scripts. The vast image libraries which have recently become available allow for data mining to enable the study of evolution of these ancient scripts. In particular, the origin of the Indus Valley script is highly debated and is considered an undeciphered script. In this paper, we use convolutional neural networks to test which alphabets/symbols from various languages may be related to the Indus Valley script. The languages focused on include the Proto-Elamite script and Sumerian.