基于全局距离和类别信息的邻域保持嵌入算法

提出一种基于全局距离和类别信息的邻域保持嵌入算法。该方法在使用欧氏距离构造邻域图中,加入表征全局距离的全局因子和表示类别信息的函数项,全局因子可以使分布不均匀的样本变得平滑均匀,类别信息可以使同类样本点紧凑异类样本点疏离,通过提高所选邻近点的质量,优化数据的局部邻域,使降维后的数据具有更好的可分性。试验结果表明,该算法具有较高的准确率,优于传统的邻域保持嵌入算法。

A neighborhood preserving embedding algorithm based on global distance and label information

An algorithm of neighborhood preserving embedding based on global distance and label information was proposed. A global factor that characterized the global distance and a function term that characterized the label information were added in the traditional Euclidean distance formula of adjacent graph. Global factor could make unevenly dirtibuted samples smooth and uniform, label information could make intra-class compact and inter-class separable, which improved quality of neighborhood and constructed an optimal adjacency graph, and improved classification accuracy. Experimental results showed that the proposed algorithm had higher accuracy and performed more effective than traditional neighborhood preserving embedding algorithm.