Convolutional Neural Network for English Character Recognition

Abstract:
In the subject of image recognition, character recognition is a very active branch. On the one hand, the difficulty of the problem has made it a tough subject to study. Character recognition, on the other hand, is not a stand-alone application technology; it has some of the same fundamental and common issues as other fields of pattern recognition. Pattern recognition and image analysis have grown into a mature scientific discipline as a result of the rapid development of character recognition technology. In this paper, the recognition of handwritten English characters will be experimented with convolutional neural network (CNN) in order to achieve higher recognition accuracy. In the experiment, the preprocessing of NIST dataset will be completed first, and then the CNN model used in the experiment will be continuously optimized. By debugging the hyperparameters and observing the influence of different parameters on the experimental results (character prediction accuracy), the more suitable model parameters will be determined. The results of the experiment reveal that CNN is better at extracting visual features. When using a three-layer convolutional neural network to train a CNN model, the computational complexity of the model is reduced, and the final prediction accuracy is increased.