Deep Neural Networks for Handwritten Chinese Character Recognition
Renan Guedes Maidana, Juarez Monteiro dos Santos, Roger Granada, Alexandre M. Amory, Rodrigo Coelho Barros · 2017
Automatic handwriting recognition is an important task since it can be used to replace human beings in various activities such as identifying postal addresses on envelopes, information in bank checks, and several other tedious tasks that humans need to perform. Convolutional Neural Networks are a power machine learning method for computer vision tasks, having achieved state-of-the-art results in the recognition of handwritten Arabic digits and also in multiple distinct alphabets. In this work, we extensively explore the performance of those networks for handwritten Chinese characters recognition (HCCR). For such, we have trained several models based on popular convolutional neural networks architectures that are commonly used for large-scale image recognition, and we also employ several distinct architectural fusion methods, resulting in more than 18 classification approaches. We report the results of all 18 configurations in the well-known HWDB and ICDAR2013 datasets.