A hand crafted bayesian network(trained with MAP inference) are efficient structure learning frameworks and can employed to learn structural similarity between two different sets of handwriting

Siamese twin networks are neural network architectures that learn the difference between a pair of inputs as an encoding(rank), with the help of contrastive-divergent loss, when trained with labelled pairs as inputs and are reasonably effective in similar tasks.

The project’s aim was to demonstrate the superiority of deep learning techniques in pattern recognition tasks, in terms of minimizing manual labour in hand designing features, provided the annotated data is of sufficient quantity.

https://github.com/AshVijay/Handwriting-Similarity-STN-PGM- —