uTHCD: A New Benchmarking for Tamil Handwritten OCR

Handwritten character recognition is a challenging research in the field of document image analysis over many decades due to numerous reasons such as large writing styles variation, inherent noise data, expansive applications it offers, non-availability benchmark databases etc. There has been considerable work reported literature about creation database for several Indic scripts but Tamil script still its infancy only one [5]. In this paper, we present done an exhaustive and unconstrained Character Database (uTHCD). consists around 91000 samples with nearly 600 each 156 classes. The unified collection both online offline samples. Offline were collected by asking volunteers write on form inside specified grid. For samples, made similar grid using digital pad. encompass vast variety styles, distortions arising from scanning process viz stroke discontinuity, variable thickness stroke, distortion Algorithms which are resilient data can be practically deployed real time applications. generated 650 native including school going kids, homemakers, university students faculty. isolated will publicly available raw images Hierarchical Data File (HDF) compressed file. With database, expect set new handwritten serve launchpad avenues domain. Paper also presents ideal experimental set-up convolutional neural networks (CNN) baseline accuracy 88% test data.