Recognition of handwritten Chinese postal address using neural networks

An automatic mail-sorting machine based on the development of a real time OCR system is proposed. The system, being the first of its kind in handling Chinese mail, is capable of distinguishing the city/county names in Taiwan, on handwritten/machine-printed standard Chinese style envelopes. The identification of the Chinese postal addresses is achieved by implementing an address recognition strategy that consist of a number of stages, including preprocessing, address block location, address segmentation, character recognition, contextual postprocessing, and result arbitration stages. Some empirical approaches are adopted to improve both the speed and effectiveness of postal address processing, in order to put the system to work in a real-time situation. The system architecture and related strategies are reported. Experimental results demonstrated that the system is capable of sorting 5400 mail pieces per hour with a correct rate of 75.6% and an error rate of 0.92%