Abstract: In order to resolve the shapcbased face recognition method's descriptor shortcoming that is not stable and the variance with respect to transform, scaling, and rotation, multi scale edge images and image's characterized descripfor were got through wavelet transform, classification feature was extracted by improved linear discriminant analysis in wavelet transform domain, at last geting the recognition result. The PIE face database was used to test the proposed method, the experiment results show that new method which is based on wavelet transform has better recognizing effect, the new descriptor which can describe image's shape and spatial distribution message very well has the excellence result of stable and the respect to transform, scaling, and rotation.

Key words: Face recognition, Wavelet transform, Shape feature, Linear discriminant analysis