Feature Dimension Reduction of NaXi Pictographs Characters Recognition based on LDA
Hai Guo, Jinghua Yin, Jingying Zhao · 2012
As a kind of pictographic character, Naxi pictographs character has received little academic attention. Proposing dimension reduction method of Naxi pictographs characters on the basis of LDA (Linear Discriminant Analysis), this paper thus makes an in-depth study of feature dimension reduction, an important issue in the recognition of Naxi pictographs characters. By constructing a recognition sample library involving four features of grid feature, permeability number feature, moment invariant feature, and directional element feature (DEF), 50% of data are selected from sample library as training set and testing set respectively. Two dimension reduction methods of LDA and FA (Factor Analysis) are applied to dimension reduction experiment of features of Naxi pictographs characters. The experiment result proves LDA method to be significantly superior to FA method, as LDA method could still maintain a 99% recognition precision when the dimension is reduced to 10% of the original dimension.