基于四元数小波幅值相位表示及分块投票策略的人脸识别方法*

提出一种基于四元数小波幅值相位表示及分块投票策略的人脸识别方法。该方法首先对人脸图像进行预处理，利用四元数小波变换的四路小波提取多个角度方向的小波系数，并求取四元数幅值和三个相位，将这些幅值和相位特征组合并分成若干子块，对每个子块根据最近邻原则进行分类，对各子块分类结果进行投票以实现人脸图像最终识别。对五个人脸数据库的实验表明，该方法具有较高识别率和对表情及光照变化的鲁棒性。

Face recognition method based on quaternion wavelet magnitude/phase representation and block voting strategy

This paper proposed a face recognition method based on magnitude/phase representation of the QWT and block voting strategy. This method firstly preprocessed human face images, then used four wavelets of QWT to extract the wavelet coefficients of multi-orientation, then computed quaternion amplitude and three phases. After combining these quaternion amplitude and phase features and dividing them into several sub-blocks, and classified each sub-block according to the principle of the nearest neighbor. Finally voted these sub-block classification results to acquire the ultimate face recognition. Experimental results on five face databases show that the method has high recognition rate and robustness to expression and light changes.