Research on Feature Extraction for Character Recognition of NaXi Pictograph

Character recognition is one of important applications for pattern recognition. Feature extraction is the most important problem in character recognition. In this paper, we have researched on feature extraction for character recognition of NaXi pictograph. The characteristics of NaXi pictograph are firstly analyzed. Four kinds of features, including permeability number, coarse grid, directional line element, and invariant moments, and 6 kinds of computation distances, including Lidean distance, Standardized Euclidean distance, Minkowski metric, Chebychey distance, City distance, and Correlation distance will be applied to classify characters of NaXi pictograph, based on its characteristics. We have built a sample database containing 21000 samples for 2120 NaXi characters. The results of experiments suggest that coarse grid and directional element are more suitable for character recognition of NaXi pictograph, while invariant moments is not advisable for this task.