An optimal handwriting recognition system has many essential applications in the present day world - they can be used for the automatic deposit of bank cheques, digitization of documents, and even in post offices to scan parcels. Despite the significant contributions in this sector in the English language, much work still remains in the domain of Bengali, the national language of Bangladesh. This is particularly challenging since the curves and variations of Bengali characters (especially compound ones) are far more complex than they are in the English language.
In order to solve this problem, experiments were conducted with four state of the art deep learning models and their classification performances compared. A hypertuned, shallow CNN network was used as the baseline. The dataset of handwritten Bengali characters used in this project was collected from: https://shahariarrabby.github.io/ekush/#download.