Information Discriminant Feature Extraction Based on Mutual Information Gradient Optimal Computation

A linear feature extraction method is present with information discriminant analysis, it is based on a feasible computationally feature extraction matrix used mutual information gradient. Firstly, this paper analyzes the limitation for current linear discriminant, and constructs a information discriminant analysis model which facilitates the maximization of the mutual information under the parametric class-conditional PDF. Then, it is proved that the mutual information is linear transformation invariance and optimal in the sense of Bayes, and the algorithm is present for computing feature extraction matrix with mutual information gradient. Finally, the good performance of the method is proved on real-world data set.