抗噪的有监督局部保留投影降维算法

Anti-noise supervised locality preserving projection algorithm for dimensionality reduction based on L1-norm

针对数据降维中的噪声干扰问题,提出基于L1-norm有监督局部保留投影算法SLPP-L1。SLPP-L1利用L1-norm替代了L2-norm;因为欧式距离比绝对值距离对噪声更加敏感,使得SLPP-L1抗噪性方面非常有效。实验结果表明,该方法可以有效地剔除噪声的影响并且提高分类的识别率。

To solve the noise disturbance of dimensionality reduction,this paper proposed a L1-norm based supervised locality preserving projection algorithm（termed SLPP-L1）.It proposed SLPP-L1 by replacing L2-norm with L1-norm.Because Euclidean distance was more sensitive to outlier than absolute distance,SLPP-L1 became robust to outlier effectively.The experiments show that the proposed method can eliminate noise effectively and improve rate of recognition in classification.