基于谱域特征提取与线性回归分类的智能人脸识别算法

针对人脸识别中由于光线、表情变化和遮挡导致人脸图像变化的问题，提出了一种谱域特征提取与线性回归分类算法相结合的智能人脸识别方法。为了实现特征提取的目的，首先使用Viola-Jones算法从原始图像中提取初始人脸部分，并将其转换为120×120像素大小的灰度图像；然后提出了一种计算极坐标傅里叶变换（FFT）以获得预处理人脸图像主要幅度谱特征的新框架，进一步在预处理的图像上执行2D-DFT，并表示为1D P-FFT。特征值是1D P-FFT幅值中的最大值，提取的特征值用于构造表示人脸图像的符号对象。最后利用快速有效的线性回归分类算法实现分类。在AR和GT数据库上进行了各种实验，分别取得了97.51％和98.02％的准确率，与最近报道的一些人脸识别技术相比，提出的方法识别准确率更高。

Intelligent face recognition algorithm based on spectral feature extraction and linear regression classification

Aiming at the problem of face image change caused by light, expression change and occlusion in face recognition, this paper proposed a smart face recognition method combining spectral domain feature extraction and linear regression classification algorithm. For the purpose of feature extraction, this method first extracted the initial human face part from the original image using Viola-Jones algorithm and converted into a grayscale image with a size of 120×120 pixels. Then, it proposed a new framework to compute the polar amplitude Fourier transform(FFT) to obtain the main amplitude spectral features of the preprocessed face image. It further performed the 2D-DFT on the preprocessed image and expressed as a 1D P-FFT. The eigenvalue was the maximum value of the 1D P-FFT amplitude, and it used the extracted eigenvalue to construct a symbol object representing a face image. Finally, it implemented classification using a fast and efficient linear regression classification algorithm. Various experiments were performed on the AR and GT databases, this method achieved 97.51% and 98.02% accuracy, respectively. Compared with some recently reported face recognition techniques, the proposed method has higher recognition accuracy.