Recognition of East Asian Language Character - a CS230 project
Authors: Yi-Ting Chen and Po-Nan Li
This project addresses the image classification of East Asian characters by using a convolutional neural network.
The trained network reaches the classification accuracy of 95.62%, and up to 99.75% if Traditional and Simplified Chinese are seen as a single label.
We further performed feature visualization, deconvolution and class model analyses, which reveal that our network correctly learned most of the frequent characters that can distinguish the four systems under consideration.