Lp norm spectral regression for feature extraction in outlier conditions

Spectral regression is a newly proposed method which is widely used in signal processing and feature extraction. However, like most methods based on regression analysis, it is prone to outlier artifacts with large norm. In this paper, a novel regression function for SR is constructed in the Lp (p ≤ 1) norm space with the aim at compressing the outlier effects on pattern recognition. The quantitative evaluation using simulated outliers demonstrates the proposed method can effectively deal with the outliers introduced in the features.