Similarity Chinese character recognition based on deep neural network in large data

Similarity Chinese character recognition based on deep neural network in large data

Charles Tau Zhang Shuye Jin Lianwen

Aiming at the limitation of traditional handwritten Chinese character recognition system (SHCCR) by feature extraction method, this paper proposes to use the depth neural network (DNN) to recognize the characters of similar Chinese characters automatically, and introduces the method of similar character set generation and the specific structure of deep neural network for similar Chinese character recognition, The effects of different training data scale on recognition performance were studied. The experiments show that DNN can effectively carry out characteristic learning, avoid the shortage of artificial design features, and compare with the traditional support vector machine (SVM) based on gradient feature and nearest neighbor classifier (1-nn) method, the recognition rate is greatly improved, and as the training sample increases, DNN is more outstanding in improving the recognition performance, the large data training has obvious effect on the recognition rate of the elevation neural network.

Similarity Chinese character recognition based on deep neural network in large data