Chinese Character CAPTCHA Recognition Based on Convolutional Neural Network

The goal of this paper is to achieve effective recognition of Chinese character CAPTCHA, we propose a convolutional neural network model with reference to LeNet-5, the number of convolution kernels is increased to enable more efficient extraction of features, while adding dropout layers to prevent overfitting and adding normalized layers to prevent gradient explosions. The model takes the grayscale, binarization, and segmented CAPTCHA pictures as input, and outputs the vector of 3,500 dimensions which indicate the probability of each Chinese character. After training, the model can achieve a recognition rate of 99.6%. The experiment also compares the model with existing model, the results show that the model can identify Chinese character CAPTCHA more effectively.