Finger-Written Chinese Characters Recognition Using Hierarchical LDA

In our camera based user interface, user inputs Chinese characters by moving fingertip with his/her fingers. Therefore, we call those characters "finger-written Chinese characters" (FWCCs). To recognize FWCCS with high recognition accuracy, we propose a new method to discriminate similar FWCCs based on a two stages linear discriminant analysis (LDA) architecture. Experimental results show the proposed method is useful for the improvements of recognition rate and outperforms our previous hybrid recognition method. Comparing with MQDF-based method, the proposed method can reach similar classification performance but with much less storage requirement.