Handwritten Hindi Word Generation to enable Few Instance Learning of Hindi Documents

Handwritten Text Recognition (HTR) of Hindi Documents is a challenging research problem of interest which could enable digitization of millions of official documents. Due to challenges in character segmentation, Segmentation-free Word Recognition is the preferred approach. Lack of a large, diverse Hindi Handwritten Word dataset for pre-training deep learning architectures is a pressing issue. In this paper, we propose a novel way of generating diverse Handwritten Hindi Word images using only Handwritten Hindi Characters and further analyze its effectiveness in enabling Few Instance Learning of Handwritten Hindi Documents.