基于频谱脸和不变矩人脸识别新方法的研究

Research on face recognition system based on spectroface and moment invariants

提出了基于频谱脸和不变矩的人脸识别方法。频谱脸是一种人脸的特征表达方式,可以有效地消除人脸表情变化、少许遮掩对人脸识别的影响。而不变矩是图像的一种统计特征,具有的平移不变性、旋转不变性和比例不变性,可以有效减少少许姿态和光照条件变化所带来的识别误差。因此首先对原始图像进行适当级数的小波变换及傅立叶变换得到人脸图像的频谱脸表达,然后利用频谱脸图像矩阵的不变矩作为识别特征,建立人脸识别模型。利用ORL人脸数据库进行仿真实验,结果表明,该方法识别率较高,识别速度较快,便于实时实现。

This paper proposes a technique based on Spectroface and moment invariants for face recognition. Spectroface is a kind of face characteristic expression method which can effectively eliminate the influence caused by change of face expression and a few covers. Moment invariant is a kind of image statistical characteristic which is invariable under shifting, scaling and rotation. It can effectively reduce recognition errors caused by a few gestures and change of illumination. Therefore, we can get the Spectroface from the original face imagecombining wavelet transform and Fourier transform. And the face recognition model can be established using Spectroface image matrix as feature. Experiment results based on ORL face database demonstrate the method has high accuracy, high recognition rate and convenience for real - time realization.