Local structure preservation based discriminant projection method for feature reduction

To cope with the high dimensionality of data and small sample size problem, a supervised feature extraction method is proposed in this paper which increases the class discrimination and preserves the local structure of data. The proposed method maximizes the between-class differences, minimizes the within-class differences, and minimizes the reconstruction errors simultaneously. The experimental results on two real hyperspectral images show the preference of proposed method compared to some popular feature extraction methods from classification accuracy point of view in small sample size situation.