Irrelevant Variability Normalization via Hierarchical Deep Neural Networks for Online Handwritten Chinese Character Recognition

This paper presents a novel irrelevant variability normalization (IVN) approach via hierarchical deep neural networks (HDNNs) and prototype-based classifier for online handwritten Chinese character recognition. The recent insight of deep neural network (DNN) is the deep architecture with large training data can bring the best performance in many research areas. The architecture design of our proposed hierarchical deep neural networks focuses on both "depth" and "width" of artificial neural network. Specifically for the multivariate regression, HDNN consists of multiple subnets, which is empirically more powerful than DNN. In this work, HDNN is adopted as a nonlinear feature transform to normalize the feature vector of handwritten samples with irrelevant variabilities to a target prototype. The effectiveness of proposed method is verified on a Chinese handwriting recognition task. Furthermore, we have an very interesting observation that DNN-based IVN can not even bring performance gain over the prototype-based classifier while HDNN-based IVN yields significant improvements of recognition accuracy.