method: Megvii-Image++2015-11-09

Authors: Chi Zhang, Shuchang Zhou, Zhimin Cao, Yin Qi

Description: First, the words are recognized with an LSTM recurrent neural network. Then, the results are adjusted with a dictionary.