从增强图象的分类信息角度出发，对人脸图象做频域上的带通滤波预处理，来提高奇异值特征的分类性能。给出了一组圆形滤波函数，并使用遗传算法来选择可分性较大的频段和进行圆形滤波函数的参数估计，设计了染色体的表示方式和适应度的求法。实验证明，做频段选择预处理，奇异值特征的分类效果比处理前有较明显的提高，而做更精确的圆形滤波函数的参数估计，奇异值特征的分类效果比处理前有更明显的提高，计算量则相对较大。这种方法对其它代数特征也适用。

In this paper, in order to enhance the classification information of an image, we use band pass filtering to preprocess face images, improve the sorting performance of singular value (SV) feature . We give a group of circular filter functions, use genetic algorithm to select the frequency bands of larger divisibility value and to estimate the parameters of circular filter functions, and design the expression of chromosome and the solution of fitness. Experiments prove that by selecting the frequency bands the sort results of SV features can be improved apparently, and by estimating the parameters it can be improved much more apparently and computation complexity is relatively larger. Our method can also be adapted to the other algebraic features.