Face Recognition across Poses Using Transformed Features

In this paper we developed a subspace analysis technology for pose-insensitive face recognition system using the proposed transformation matrix. Generic training image sets are used to train view-dependent subspaces, transformation matrix is calculated by solving linear matrix equations. Utilizing transformation matrix, side-view features could be converted into frontal view, recognition is implemented within same subspace by matching converted features with enrolled features. Experiments on FERET database proved that the proposed transformation matrix is effective for face recognition across poses. Popular feature extraction methods were also compared