Abstract: For the research of face recognition, the discrete wavelet transform has the disadvantage that the resolution is low. S transform has high resolution and multi-scale direction, but it has high computational complexity and large redundant information. Two dimensional fast generalized S transform is proposed to overcome the shortcomings. Firstly, the amplitude and phase spectra of the two dimensional fast generalized S domain are analysed by programming, and the original signal is reconstructed. It is proven that two dimensional fast generalized S transform has multi resolution and multi orientation. Secondly, the direction based on the fast generalized S domain for two dimensional signals is combined with Support Vector Machine（SVM）.Low frequency direction of S domain is used as the input feature of SVM and face recognition and classification are realized in ORL standard face database. Experimental results show that compared with the discrete wavelet transform, the two dimensional fast generalized S transform, applied to face recognition by combining it with the support vector machine, has higher recognition rate.

Key words: redundant, resolution, orientation, two dimensional fast generalized S transform, recognition rate

摘要： 针对人脸识别研究中，离散小波变换分辨率低的缺点及S变换虽分辨率高、方向性多，但计算复杂度高且冗余信息量大的现状，提出了二维快速广义S变换，克服了这些缺点。首先通过编程分析了二维快速广义S域幅度谱及相位谱，并重构原始信号，证明了二维快速广义S变换具有多分辨率和多方向性。其次，将二维快速广义S域的方向性与支持向量机（SVM）结合，将S域低频方向作为SVM的输入特征，实现对ORL标准人脸库中的人脸的识别与分类。实验结果表明：与离散小波变换相比，二维快速广义S变换与SVM相结合应用于人脸识别中，具有更高的识别率。

关键词: 冗余, 分辨率, 方向性, 二维快速广义S变换, 识别率