HIT-MW Dataset for Offline Chinese Handwritten Text Recognition.pdf

HIT-MW Dataset for Offline Chinese Handwritten Text Recognition Tonghua Su Tianwen Zhang Dejun Guan School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China tonghuasu@ School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China twzhang@ Heilongjiang Mobile, Harbin, China guandejun@ Abstract A Chinese handwritten text dataset, HIT-MW, is presented to facilitate the offline Chinese handwritten text recognition. Texts for handcopying are sampled from China Daily corpus with a stratified random manner. To collect naturally written handwriting, forms are distributed by postal mail or middleman instead of face to face. The current version of HIT-MW includes 853 forms and 186,444 characters that are written by more than 780 participants under an unconstrained condition without preprinted character boxes. Its lexical coverage of 3,041 characters is about 99.33% measured on China Daily corpus with about 80 million characters. Handwritten texts of HIT-MW mainly written by college students follow a balanced distribution both in sex and in department. It can be used to conduct Chinese text- line segmentation, segmentation-free recognition, and to verify the effect of statistical language model in a real handwriting situation. Keywords: Standardization, Data acquisition, Optical character recognition, Handwritten Chinese text 1. Introduction Standard datasets play crucial roles in handwriting recognition research. On the one hand, they provide a large number of training and testing data, resulting in high model fit and reliable confidence in statistic. On the other, they offer a means by which evaluation among different recognition algorithms can be performed. More and more handwriting researchers begin to pay much attention to the dataset standardization and evaluate their work using standard datasets. Dozens of handwriting datasets have been published since 1990s. I