Binarization and isolation of historical manuscripts’ symbols

The problem of historical manuscripts’ binarization with the purpose of original symbol graphics’ isolation is considered. Correctness of historical documents’ decryption largely depends upon valid and proper binarization of the text. Historical shorthand records of the XIX century are taken as objects of research. The analysis of various binarization methods (Otsu, Bernsen’s methods, Eykvil, Niblek, and various threshold methods) was carried out. The conducted research proved that the proposed modified threshold method based on F-measure is more effective than other methods. The method is applied in the software complex aimed at getting original graphical representation of the symbols. The use of the elaborated software complex was instrumental in the reading of 29 manuscript sheets. More than 6,800 graphic symbols were isolated during decoding process.