The Recognition And Implementation Of Handwritten Character Based On Deep Learning

Deep Learning is a network architecture based on deep multiple hidden layers and being processed to research some hot issues in the field of machine learning nowadays, such as image retrieval and identification. The nature of the deep learning is by building multiple hidden layer of neural network for training sample data, while the essence of training samples is to allow a network study sample characteristics independently, thus eliminating the artificial feature extraction steps based on people experience previously. Although there are more applications using deep learning technology to solve handwritten digit recognition problems, but in the field of handwritten Chinese character, there are more methods based on artificial feature extraction rather than deep learning. Since the Chinese character strokes are more complex compared with other common characters such as letters and Arabic numbers, and the Chinese character becomes a great variety because of the difference on personal writing style and habit. Therefore, the recognition of Chinese character has always been a hot issue in the field of machine learning.In this paper, we aim at the recognition of handwritten character especially handwritten Chinese character. We use different depth convolution neural network to research the recognition of datasets which contain handwritten digits and handwritten Chinese character, we compared the performance differences between same depth and different depth network. We got the network that can be used for recognition finally, and achieved a handwritten character system based on it.The accuracy of our network on MNIST dataset reached 99.18% while the accuracy on HWDB1.1 subset reached 92.02%, and with a great applicability on new data set. We proved that our network structure’s engineering application value by making a contrast with other character recognition methods, and finally we achieved a handwritten character system based on the model which we’ve trained to confirm the feasibility of our network.