线性合成的双粒度 RNN 集成系统

针对脱机文字识别,提出了一种基于线性合成的双粒度递归神经网络(Recurrent neural net work, RNN)集成系统.首先,使用单词RNN对未知图 像进行识别;然后,依据识别结果进行字符分割,使用字符RNN对分割后的字符进行识别,并利用查表法计算字符的后验概率;最后,综合两个RNN的识别结果决定最终单词输出.在CAPTCHA识别 和手写识别上的实验结果证明了该系统的有效性.

Ensemble System of Double Granularity RNN by Linear Combination

For offline text recognition, an ensemble system of double granularity RNN (recurrent neural network) by linear combination is proposed. Firstly, the unknown image is recognized by a word RNN. Secondly, based on the recognition result, connected characters are segmented, and then recognized by a character RNN. Characters' posterior probabilities are calculated by a table-looking method. Finally, the final output is decided by combining both RNNs' results. Experiment results based on CAPTCHA (completely automated public Turing test to tell computers and humans apart) recognition and handwritten recognition proved the effectiveness of this ensemble system.