【摘要】 Maximum Variance Unfolding(MVU)是一种基于流形学习的非线性降维方法。该算法中近邻点的选取对MVU的降维效果影响很大。利用样本点聚类后的类别信息构造密度系数,提出了一种MVU改进的算法。所提出算法根据近邻点分布的不同,挖掘数据局部的密度信息,有效的保持了高维数据中的流形结构。人脸表情和图像检索实验证实了所提出方法的有效性。更多还原

【Abstract】 Maximum Variance Unfolding(MVU) is manifold learning nonlinear dimensionality reduction method. In MVU algorithm, the effective of dimension reduction is affected by the distribution of the neighborhood points. By using classified information of sample points to construct density coefficient, this paper proposes an improved method of MVU algorithm. The new proposed algorithm mines local density information, can be effectively to keep the manifold structure of high dimension data. Face expression and image retrieve experiments show the effectiveness of the proposed algorithm.更多还原