Inter-subspace distance: a new method for face recognition with multiple samples

In this paper, we develop a systematic method that can cope with multiple images simultaneously for face recognition. The proposed method, referred to as inter-subspace distance, employs the minimal distance between the two subspaces formed by training and test images, respectively. The advantages of our method are that it can use temporal information (image sequences) and multiple sampling (in scales or spatial positions) for face recognition. In addition, our method can ease the burdens of face detection by dealing with inaccuracies of positions and scales of the detected faces.