Margin Maximum Embedding Discriminant (MMED) for Feature Extraction and Classification

This paper develops a supervised discriminant technique, called margin maximum embedding discriminant (MMED), for dimensionality reduction of high-dimensional data. In graph embedding, our objective is to find a linear transform matrix to make the samples in the same class as compact as possible and the samples belong to the different classes as dispersed as possible. The proposed method effectively avoids the singularity problem frequently encountered in the classical linear discriminant analysis due to the small sample size (SSS) and overcomes the limitations of the traditional linear discriminant analysis algorithm (LDA) due to data distribution assumptions and available projection directions. Experimental results on ORL and AR face databases show the effectiveness of the proposed method.