Auto-Generation of Captions for Ancient Indian Historical Images using Deep Learning Networks

Automatic image annotation is the process through which a machine can automatically describe an input digital image in the form of keywords or captions. In this paper we focus on generating captions for various sculptures carved in the ancient past by the Indian kingdoms in various heritage sites such as temples, caves and other places. Towards this direction, a new dataset along with annotations is created. This dataset is used on various deep learning models such as VGG16-LSTM, ResNet50-LSTM, VGG19-LSTM, Xception-LSTM, InceptionResNetV2-LSTM for the auto generation of captions and the performance measure for each of the models is evaluated.