面向人脸识别的矩阵摄动法快速特征重建

FAST FEATURES RECONSTRUCTION WITH FACE RECOGNITION-ORIENTED MATRIX PERTURBATION THEORY

视频人脸识别是当今生物识别领域的研究热点。介绍一种新的面向人脸识别的特征重建算法。该算法在开放性视频实时监控系统中,能够快速有效地处理新图像数据库的特征重建问题。算法基于人脸识别的2DPCA方法,利用矩阵摄动和Rayleigh商理论,通过已知数据快速估算出具有很高精度的特征子空间;通过仿真实验,该算法在保证精度的情况下,能够节省大量训练时间,效果较好。

Video face recognition is the research focus in today＇ s biometric field. A new face recognition-oriented feature reconstruction algorithm is introduced. In real-time monitoring system of the open video, the algorithm can effectively and quickly deal with features reconstruction of new image database： The algorithm is based on 2DPCA method of face-recognition, by employing the theory of matrix perturbation and Rayleigh quotient, the algorithm quickly estimates the feature subspace in very high precision by known data. Through simulative experiments, the algorithm can save a lot of training time in condition of ensuring accuracy with fair effect.