四元数小波和AdaBoost在人脸识别中的应用

提出一种基于四元数小波变换（QWT）幅值相位表示及AdaBoost的人脸识别方法。四元数小波变换具有近似的移不变特性，可以同时支持1个幅值和3个相位，其中两个相位编码局部图像移动，而第三个相位蕴含纹理信息。方法对人脸图像进行预处理，进行四元数小波变换并计算四元数幅值和相位特征，将这些幅值和相位组合并应用AdaBoost分类器进行分类，以实现人脸图像的最终识别。对Yale、ORL和FERET三个人脸数据库应用此方法的实验结果表明，该方法在识别率上优于AdaBoost和Gabor+AdaBoost。特别是在FERET数据上精度提高更为明显，而且在计算复杂度上QWT特征提取明显低于Gabor特征提取。

Application of quaternion wavelet and AdaBoost in face recognition

This paper proposes a face recognition method based on the Quatemion Wavelet Transform（QWT） magnitude/ phase representation and AdaBoost.This recent transform is a near shifl-invariant tight frame representation whose coefficients support a magnitude and three phases.The first two QWT phases encode the shifts of image features in the absolute horizon- tai/vertical coordinate system, while the third phase encodes edge orientation mixtures and textural information.The method preprocesses human face images, uses QWT to extract the wavelet coefficients of multi-orientation, and quaternion amplitude and three phases are computed.These quaternion amplitude and phase features are combined and AdaBoost is used to realize recognition.Experimental results on three face databases including Yale, ORL and FERET show that the method has higher recognition rates than AdaBoost and Gabor＋AdaBoost.This method has much better recognizing result on FERET especially. Compared with Gabor wavelet features,QWT features are superior both in accuracy and computation complexity.