A kernel orthogonal isometric projection algorithm for face recognition
Guanqun Liu · Journal of Optoelectronics·laser · 2010
In this paper, a new feature extraction method for face recognition named as kernel orthogonal isometric projection (KOIsoP) is proposed.First,the face images are mapped into a high-dimensional nonlinear space by the kernel method to obtain the nonlinear information,which helps to extract the nonlinear manifold structure information,and then a linear transformation is performed by the isoprojection to get the basis vectors.Finally the basis vectors are orthogonalized,which can preserve the nonlinear manifold subspace in faces and the distance-based structure information,as well as reconstructed datasets,so the proposed algorithm can gain higher recognition rate with lower operation compexity.Experiments on ORL and PIE face database demonstrate the effectiveness of the proposed algorithm.