Low frequency sub-band figures are extracted with wavelet transform

the optimal cast shadow space is found by using optimized linear discriminant analysis（LDA）algorithm

the optimal feature space is got.Face classification and identification are realized in the feature space.In this method

nonsingularity of within class scatter matrix became unnecessary

and the problem of edge overlap is also solved.So

it has better generalization ability comparing with traditional LDA algorithm.Experimental results show that this method is superior to the tranditional LDA and principle component analysis（PCA） method.