| Nüshu is the world's most gender character, and has an important value of the intangible cultural heritage protection. So far, the inheritance of Nüshu documents mainly relies on the way of manual transcription. With the inheritance persons of the Nüshu died successively, the culture of Nüshu is endangered as collecting and compiling the literature of Nüshu become more difficult. In order to solve this problem, handwritten character recognition technology is used in this dissertation to informationize the literature of Nüshu, so as to protect and promote Nüshu which is precious Chinese civilization.An off-line handwritten Nüshu character recognition program for the features of Nüshu is provided in this paper, based on the detailed analysis on the current off-line handwritten character recognition algorithm. The study started from the design of the scheme. The work processes of the off-line handwritten Nüshu character recognition, functions of each part and the frequently-used methods are carefully analyzed. The peripheral direction contribution feature extraction algorithm is applied to the feature extraction of Nüshu. An improved stroke density feature extraction algorithm and a three-level classification algorithm are proposed as well as a practical system of Nüshu is designed and implemented.The main work and the features are as follows:1) According to the samples of Nüshu, smoothing algorithms and binary algorithm are used to remove the background and noises in the sample images. According to the distribution characteristics of Nüshu characters which is in samples, used the combined with the line segmentation algorithm. Finally the size of the Nüshu characters is normalized.2) The features and the defects of using in Nüshu characters of the two stroke density feature extraction algorithm are analyzed.The peripheral direction contribution feature extraction algorithm is applied to the feature extraction of Nüshu as well as an improved stroke density feature extraction algorithm is proposed according to the tilt features of Nüshu characters.3) The current multi-level classifier is analyzed, a three-level distance classifier is designed according to the defects of Euclidean distance in the recognition process. One-level classifier uses Manhattan distance classifier as well as two-level and three-level use error balanced distance. This method has the advantage of the Manhattan distance classifier of high speed and the error balancing distance classifier can make the stable characteristics of Nüshu characters highlighted and the instable parts inhibit.4) An off-line handwritten Nüshu character recognition system is designed and implemented based on the improve stroke density feature extraction method and three-level distance classifier algorithms. Simulation experiment of the system is carried out and the experiment results are analyzed and compared. |
