A proposal of face identification using non-orthogonal transform sign

A face identification method using the sign of non-orthogonal transform is proposed. In our previous work, the sign of orthogonal transform is exploited for face identification. The orthogonal transform has a good characteristic to concentrate the energy in low frequency area, it is good for image compression because of orthogonalization. However, for face identification, what we care about is how to pick up the unique information of a individual, the inverse transform is not important any more. Therefore, a non-orthogonal transform is developed in this study. The proposed method inherits the priorities of previous one. Those are little registration data, and no retraining problem, and a faster calculation.