基于正弦变换的人脸姿态校正及识别研究

姿态变化是影响人脸识别率的一个至关重要的因素，也是人脸识别问题中一个待解决的难题。当测试样本具有一定的姿态变化后，识别率会急剧下降。针对此问题，提出了利用正弦变换（Sine Transform，ST）对待识别的姿态图像进行姿态校正，虚拟出对应的正面人脸的方法。使用经典算法进行特征提取、最近邻分类器进行分类识别验证，得到了较好的结果。在FERET人脸库上的实验表明，该方法能够在一定程度上克服姿态变化的影响，平均识别率最高可提高17%。

Face pose correction for facial recognition based on sine transform

Pose variation is one of the crucial factors and also is a difficult issue that stands in the way of a complete solu- tion to the face recognition problem.The recognition rate will decrease drastically when the probe samples change.This paper proposes a strategy to make pose corrections to the probe samples before recognition,by using Sine Transform（ST） to gener- ate virtual frontal faces, classical algorithms to extract features and the nearest neighbor classifier to classify and recognize. The experiment results on FERET face database demonstrate that the recognition rates increase significantly by adding pose correction,and the best recognition rate has improved by 17%.