基于高次曲线拟合的人脸特征表示

基于数字化曲线拟合技术, 提出一种新的人脸图像不变特征表示方法. 该方法先将人脸边缘图像分割为高次多项式及抛物线段构成的特征曲线段集合, 再运用极惯性矩度量不同人脸图像在特征曲线段两侧的局部纹理及形态差异. 实验结果表明， 该方法与直线边缘图法和抛物线边缘图法相比, 能更好地描述人脸图像的分布特征.

Face Eigen Expression Based on High Order Curve Fitting

We proposed a new method to extract invariant features from human face image. With the help of digital curve fitting, we represented face binary edge map with high order polynomial and parabola segments. Then, the polar moment of the inertia of pixel block beside the curve segments was used to measure the difference between local texture and shape. From the analysis of the experimental result, it is concluded that compared with line edge map and parabola edge map, this method can give a better way to describe the feature distribution of face images.