张量局部Fisher判别分析的人脸识别

摘要

One of the key issues of face recognition is to extract the subspace features of face images. A new subspace dimensionality reduction method is proposed named as tensor local Fisher discriminant analysis (TLFDA), which benefits from two techniques, i.e; tensor based method and local Fisher discriminant analysis. Firstly, local Fisher discriminant analysis is improved for better recognition performance and reduced time complexity. Secondly, tensor based method employs two-sided transformation rather than single-sided one, and yields a higher compression ratio. Finally, TLFDA uses an iterative procedure to calculate the optimal solution of two transformation matrices. Experiment results on the ORL and PIE face databases show the effectiveness of the proposed method.%子空间特征提取是人脸识别中的关键技术之一,结合局部Fisher判别分析技术和张量子空间分析技术的优点,本文提出了一种新的张量局部Fisher判别分析(Tensor local Fisher discriminant analysis,TLFDA)了空间降维技术.首先,通过对局部Fisher判别技术进行分析,调整了其类间散度目标泛函,使算法的识别性能更高且时间复杂度更低；其次,引入张量型降维技术对输入数据进行双边投影变换而非单边投影,获得了更高的数据压缩率；最后,采用迭代更新的方法计算最优的变换矩阵.通过ORL和PIE两个人脸库验证了所提算法的有效性.