LPP/QR for under-sampled image recognition

In this paper, we propose a dimension reduction method of locality preserving projections based on QR-decomposition of training data matrix, namely LPP/QR. It is efficient and effective in under-sampled recognition of image and text data, especially when the number of dimension of data is greater than the number of training samples. Its theoretical foundation is presented. The equivalence between LPP/QR and generalized LPP is induced although LPP/QR is faster than generalized LPP. Several experiments are conducted on Yale face database. High recognition rates show that the algorithm performs better in under-sampled situations.