摘要： 在进行降维时数据集合的多样性要求降维算法求解问题具有灵活性。为此，利用加入指数 来调节局部保持映射算法的约束条件，通过实验观察该指数的引入对降维以及识别率的影响，并总结指数 的范围和设计经验。实验结果表明，指数 可以影响降维效果，使维数降得更低，通过调节提高人脸识别率，在加入高斯白噪声后通过调节指数p也可改善识别的效果。

关键词: 局部保持映射, 流形学习, 邻接图, 约束条件, 噪声

Abstract: As the diversity of the data collection, it is necessary to enhance the flexibility of the algorithm when reduce the dimensions of the data sets. The article changes the constraints of the Local Preserving Projection(LPP) algorithm by adding an exponential parameter p. Then can see the result of dimension reduction and look over the recognition rate through different face databases. The article also attempts to summarize the scope of p and design experience. Experimental show that the exponential parameter does influent the dimension reduction results. The dimension can be lower and if select the proper p the result will be better. If add gussian white noise, the consequent is still better by adjusting the exponential parameter p.

Key words: Local Preserving Projection(LPP), manifold learning, adjaceny graph, constraints, noise

中图分类号: 

• TP391