Development of High-Precision Recognition Technology for Low-Quality Handwriting Chinese Character in the 6th National Population Census

Abstract

We have conducted mid-and long- term R&D on the OCR technology concerning low-quality Chinese handwriting character recognition. This technology is highly praised by Chinese government. We competed with the local companies as a foreign company and won the tender. Our technology was adopted in the 6th National Population Census project conducted by Chinese government in Sep, 2010.

Technical points

1. Similar character recognition based on modified linear discriminant analysis (MLDA)
2. Adaptive recognition rejection method based on mutual information conditional entropy

Effect

The National Bureau of Statistics of China (NBSC) assessed this OCR technology based on 1.14 million real census samples in Apr, 2012. The assessment results indicated that our recognition precision reaches 99.94%. After the census was completed smoothly, our technology highly convinced the NBSC.

The successful development of this technology contributed to successful business negotiation (1700 scanners) for FHK’s scanners and OCR sales in China. It proves that the high technology can contribute to the real business even in complicated market like in China. Moreover, the Nikkei Computer Magazine (Oct, 2010) also introduced the case of OCR R&D and the leading position of the research of Fujitsu R&D Center in China.

---

Contact information:

Yu Hao: yu@cn.fujitsu.com

Sun Jun: sunjun@cn.fujitsu.com

He Yuan: heyuan@cn.fujitsu.com