can clstm recognize speical character well about clstm

[mittagessen]

In general an LSTM+CTC configuration is able to recognize anything from the training data including "special" symbols (doing ancient Greek and playing around with Arabic here). You have to ensure the input you want to handle is included in the training data which is the reason the default model of ocropy doesn't deal well with these inputs.

"Tricky" stuff right now is training models performing well (<1% error) on multiple fonts and RTL scripts will need some preprocessing to reorder the label sequence.