【摘要】 针对流形学习算法普遍存在对噪声敏感的问题,提出一种克服噪声的鲁棒Laplacian特征映射算法。该算法从Laplacian特征映射出发,在降维过程中,对样本点的邻域范围采用局部PCA的方法,以识别和剔除包含的噪声点,并在重构低维嵌入坐标的同时保持流形光滑连续的整体性,较好地改善了算法的特征提取性能。实验结果表明,所提算法有效地提高了对噪声的鲁棒性。更多还原

【Abstract】 Aiming at the sensitivity of the manifold learning algorithms to noise,a robust Laplacian Eigenmaps algorithm to overcome the noise is proposed in the paper.The algorithm is mapped from the Laplacian Eigenmaps itself,in the process of dimensionality reduction,using local PCA from the neighborhood of sample points,to identify and eliminate the noise,and reconstruct low dimensional embedding coordinates while maintaining manifold smooth and continuous integration,so as to improve the performance of the feature extraction.The experimental results show that the new algorithm can effectively improve the robustness against noise.更多还原