Research On Dimensionality Reduction Based On Neighborhood Preserving Embedding And Sparse Representation

With the rapid development of the internet, the storage and analysis of large-scale data is getting more and more important.So these years,lots of technologies for storing and computing of large-scale data emerge.However,we can also use some methods of dimensionality reduction to reduce the high-dimensional data effectively, this reduces the complexity of computing from another perspective.In real world, high-dimensional data are everywhere, but the nature structure behind them is always featured by only a few parameters. With the fast development of computer vision and pattern recognition, in lots of applications, more and more data dimensionality reduction problems are involved. For example, in face recognition, a face image usually consists of tens of thousands of features, but it includes redundant informations and important relations among features hide behind it. This will increase the difficulty of training classifier and computing. It's so called "curse of dimensionality". All of the challenges above largely premote the development of dimensionality reduction.Linear method such as LPP (Local Preserving Projection), NPE(Neighborhood Preserving Embedding) and SPP(Sparsity Preserving Projections) come up with recently, nolinear method such as LLE(Local Linear Embedding) and improvement version kernel NPE.One particularly simple but effective assumption in face recognition is that the samples from the same class lie on a linear subspace, so lots of nolinear methods only perform well on some artificial data sets. This paper emphasizes on NPE and the representation of sparsity come up with recently, and combines these two methods, the experiments show the effect of new method outperform some classic unsupervised methods.