Handwritten Bangla digit recognition using hierarchical Bayesian network

A hierarchical Bayesian network is used for handwritten Bangla digit recognition. Rather than extracted feature vectors, original digit images are used as the network input directly. The network is trained on handwritten samples. And then it’s tested on untrained images and hand-drawn digits. An average recognition accuracy of 87.5 is achieved. The system exhibits robust invariant recognition with respect to considerable object noise which is quite common in handwritten digits.