Optical character recognition with fast training neural network

Optical character recognition has been extensively investigated in the past few years. Many existing techniques are able to provide high recognition rate, but at the cost of long training time. In this work, we present a neural network based approach to reduce the training time while maintain the high recognition rate. The main idea is to perform a preprocessing stage to partition the training data prior to the training stage. A multi-stage approach is then used to deal with various types of input source. Our experiments on real image datasets have demonstrated that the balance between the training time and recognition time can be achieved using the proposed method.