频谱脸方法是一种利用小波变换和Fourier变换有效地提取人像的位移不变特征和表情相对不变特征的方法。该文着重讨论了频谱脸方法系统化的预处理方法和相似性度量选择这两个关键性问题。其中，矩的方法被用于人像进行预处理，因为它能有效地对人像的伸缩和平面旋转进行矫正；通过对最近邻法、平均法、Hausdroff距离法和修正的Hausdroff距离法等4种典型的相似性度量方法中进行比较和分析的结果表明，最近邻法、平均法和修正的Hausdroff距离法都是频谱脸方法进行相似性度量的有效方法，其中，最近邻法是最有效的方法，它对诸如位移、伸缩、平面旋转、少许遮掩及少许姿势、表情和光照条件的变化多种影响人像识别的因素均具有最佳的容错性，并在Yale和Olivetti人 像数值库上进行了识别试验，分别取得了97％和99％的识别率。

Spectroface is a face representation method using wavelet transform and Fourier transform and have been prove to be invariant to translation and tolerant to expression variety. In this paper, the two important issues on Spectroface system is studied. One is how to preprocess system, another is the selection of similarity measurement. The moment is employed to preprocess system that it is good method to normalizing the scale and rotation of human face. The similarity measurement has been selected by comparing four typical kinds of similarity measurement, i.e., nearest neighbor method, averaging method, Hausdorff distance method and modified Hausdorff distance method(MH). Nearest neighbor method, averaging method and modified Hausdorff distance method are good for Spectroface. Nearest neighbor method is the most effective method in the recognition of frontal faces with translation, scale, rotation, different facial expressions, small pose, small occlusion and different illumination condition. It gives high accuracy as 97% and 99% in Yale and Olivetti face image databases respectively.