Datasets in OCR for Handwriting Recognition: A Survey

Languages across the globe have been used in expressing and writing thoughts, information, and discoveries. Aiming to store and preserve these contents in digitized form is served by the Optical Character Recognition mechanisms. The performances of these OCR engines depend a lot on the type of data given as input to learn, the model used for training, and the various evaluation matrices used for the purpose. Though literature is flooded with reviews and surveys on tools, models, and evaluation methods used for the purpose, very less is witnessed about the datasets available for the same. With this thought, we proceeded to present a survey of the various datasets created and used to recognize handwritten texts. This survey will definitely enlighten fellow researchers in this field to understand the characteristics and forms of the various datasets available to date. It will also motivate them in developing new and complex datasets which will eventually contribute to the creation of accurate and efficient learning models helpful for various OCR tasks.