HABFD: Houma Alliance Book Facsimiles Database

The Houma Alliance Book contains official characters from the Jin State (approximately 550 BC) and is also the earliest discovered calligraphy using writing brushes in the archaeological community. As an ancient handwriting system, HAB stores a wealth of information that helps us understand the world, the history of the Jin State in the Spring and Autumn Period, the evolution of calligraphy, and more. However, HAB was only discovered approximately sixty years ago, and there is little research on it. Due to environmental corrosion, the HAB has become less legible, making the technology for automatic detection and recognition of characters an important solution. This paper aims to make an Houma Alliance Book facsimiles database namely FADB, and to carry out detection and recognition. The database contains 655 facsimiles with 327 classes of Chinese characters. More specifically, it has the individual characters from the 655 facsimiles. Then, this paper uses a deep learning network to recognize these samples. Experiments are performed on the new database. The experimental results provide new baseline results of detection and recognition for this new database and demonstrate the feasibility of our dataset.