Abstract: A face recognition algorithm based on invariant feature of trace transform is presented to improve the recognition rate of single training sample in variations of pose and expression. With application of one order Scharr operator and second order scale-adapted Laplacian of Gaussian （LoG） & Harris filter, feature points are detected. Proper functionals are chosen to perform trace transform in the circular region around each feature point so as to achieve feature descriptors which are invariant to rotation and scaling. A coarse-to-fine matching strategy is conducted by applying the descriptor’s feature vectors and coordinates. It guarantees the stability as no problem of parameter selection. The experimental results demonstrate that this method decreases the effect caused by the variations of pose and expression and also reduces the running time.

Key words: face recognition, single training samples, trace transform, feature extraction, feature matching

摘要： 提出了基于trace变换不变性特征的人脸识别算法，提高了单训练样本下姿势和表情变化后的识别率。应用一阶Scharr算子、二阶尺度适应的高斯型拉普拉斯算子（LOG）和Harris滤波器定位特征点，选择合适的泛函在特征点的邻域内进行trace变换得到具旋转和尺度不变性的特征描述子。根据特征描述子的特征向量和坐标值实现由粗到精的匹配，整个过程不涉及参数选择问题，保证了算法的稳定性。实验结果证明该算法降低了姿势和表情变化时识别率低的影响，并减少了算法运行时间。

关键词: 人脸识别, 单训练样本, trace变换, 特征提取, 特征匹配