Properly set the parameter of a state-of-art LPP solution scheme 3

Locality preserving projection (LPP) not only has solid theoretical foundation but also has dominant experimental result. Especially, improved LPP solution scheme 3 can achieve a very apparent recognition result for face recognition. However, how to properly set the parameter of improved LPP solution scheme 3 is not yet well solved. As a consequence, it is very important to explore the issue of how to properly set the parameter. With this paper, we aim at addressing this problem through experimental analysis. We also show the difference in the classification right rate(CRR) of face recognition experiments that use different numbers of transform axes. Moreover, we have an interesting finding that the parameter can also be set to a negative value and in some cases the negative parameter can lead to a satisfied CRR.