Subspace method is one of the classical methods in face recognition, which assumes that face images lie in a low-rank subspace. However, due to illumination variation, shadows, occlusion, specularities and corruption, real face images seldom reveal such low-rank structure. We propose a face subspace recovery method based on the Robust Principal Component Analysis. The face image matrix is modeled as the sum of a low-rank matrix and a deviation matrix, in which the low-rank matrix reveals the ideal subspace structure and the deviation matrix accounts for the illumination variation, shadows, occlusion, specularities and corruption. By using the robust principal component analysis, the low-rank matrix and deviation matrix can be recovered efficiently. The experimental results show that this method is efficient in recovering the low-rank face subspaces.%子空间方法是人脸识别中的经典方法,其基本假设是人脸图像处于高维图像空间的低维子空间中.但是,由于光照变化、阴影、遮挡、局部镜面反射、图像噪声等因素的影响,使得子空间假设难以满足.为此,提出一种基于鲁棒主成分分析的人脸子空间重构方法.该方法将人脸图像数据矩阵表示为满足子空间假设的低秩矩阵和表征光照变化、阴影、遮挡、局部镜面反射、图像噪声等因素的误差矩阵之和,利用鲁棒主成分分析法求解低秩矩阵和误差矩阵.实验结果表明,文中方法能够有效地重构人脸图像的低维子空间.
展开▼
