基于推广形式的K-L变换的人脸识别方法

Face Recognition Based on Generalized K-L Transform

提出了一种图像预处理方法,将不同光照条件下的图像甚至是负像处理成亮度、对比度与参考图像基本相同的图像,且调整后的图像与原图像保持较高的相关性,从而有效降低了光照对人脸识别结果的影响.根据随机矩阵的主行列分析法给出了推广形式的K-L变换,提出了进行人脸识别的一种方法.同基于K-L变换的人脸识别方法相比,大幅度缩减了协方差矩阵的维数,从而大大降低了计算特征值的运算量,提高了运算速度.理论分析和实际识别证明了该方法的有效性.

A new method of image preprocessing is presented. An image under different illumination condition,even a negative image,is changed into an image keeping higher correlation with original image. As a result, the influence of illumination to face recognition can be weaken effectively. According to principal column and row analysis of random matrix,the generalized K-L transform is obtained and a new method of face recognition is proposed. The computational amount to find eigenvalue and eigenvector is reduced considerably and the computational speed can be accelerated because the dimension of covariance matrix is lessened greatly. The superiority is shown remarkably by theoretical analysis and practical recognition.