曲波变换检测关键点的多模式三维人脸识别

Multi-mode 3D Face Recognition Using Curvelet Transform to Detect Key Points

针对基于曲波变换的人脸识别技术仍受限于整个面部二维全局特征匹配的问题,提出一种利用曲波变换检测关键点的多模式3D纹理人脸识别方法。首先,通过快速曲波变换获得不同尺度的人脸图像;然后,利用曲波系数均值选择关键点,并通过关键点检测算法去除冗余点;最后,利用余弦规则进行特征匹配。实验结果显示,FRGC v2和BU-3DFE数据集上的验证率分别高达99.1%、95.3%,在中性和中性、非中性和中性、所有和中性三种模式下均优于其他几种较新方法。

As face recognition technology based on curvelet transform is still limited to the two-dimensional global feature matching of the entire face, a multi-mode 3 D texture recognition method using curvelet transform to detect key points is proposed. Firstly, different scale images are obtained by fast curvelet transform. Then, the key points are selected by curvelet coefficient, and redundant points are removed by the algorithm of key point detection. Finally, the cosine rule is used to finish the feature matching. The experimental results show that the recognition rate on the two databases can achieve at 99.1 % and 95.3 % respectively. It is higher than several other advanced methods at all the three patterns： neutral and neutral, neutral and neutral, all and neuter.