Authors: BRTRS-Team

Description: A robust detector with FCN is used to find all candidate characters. Then, text-lines are generated by a graph-based grouping method. Finally, a Bi-LSTM model is trained to split the text-line into several words.