基于球面调和函数的人脸识别

根据在任意光照下朗伯凸表面的图像可以由一个低维线性子空间近似这一原理，提出了一种新的人脸识别方法．首先使用共形变换原理把训练样本图像投影到球面和圆柱面的一个区域上，然后对球面和圆柱面上的图像用多项式插值方法进行扫描和插值，得到离散函数图像表达式．接着利用球面调和函数估计球面上图像的9个低频谐波系数，对待定的阶数求范数后得到三维的向量．最后依据最近邻决策规则进行识别．实验表明与PCA方法、Fisherface方法以及SVD方法相比，该方法的识别率和对光照、姿态的鲁棒性均有明显的提高．

Face Recognition Based on Spherical Harmonic Functions

Under a wide variety of lighting conditions, the set of images of a convex Lambertian object obtained can be approximated by a low-dimensional linear subspace using spherical harmonics representation. Using this theory a novel approach is proposed this theory. It consists of three steps： the image is projected firstly to a spherical surface and cylindrical surface by conformal mapping. By the polynomial interpolation techniques, the images of spherical and cylindrical surface are scanned and interpolated. Discrete function expressions are obtained through this process. Then, based on the spherical harmonic function theory, a three-dimension recognition feature can be extracted. Finally, the nearest neighbor decision rule is used as recognition rule. Experimental results show that the recognition rate of the algorithm is higher than other methods like Eigenface, Fisherface and SVD.