Skip to Main Content
A Transform Domain Two-Dimensional Principal Component Analysis algorithm (TD2DPCA) applied to facial recognition in the presence of noise is presented. The new algorithm maintains high recognition accuracy in the presence of noise. In addition, the TD2DPCA has attractive properties with respect to storage and computational requirements. As the storage requirements are reduced by more than 90 percent, and the computational speed is reduced by a factor of two, compared with the spatial 2DPCA method. The new algorithm is applied to the ORL and Yale datasets, in the presence of salt and pepper as well as gray scale white Gaussian noise, where the Discrete Cosine transform is used. The results are given which confirm the excellent recognition accuracy of noisy facial images employing the proposed technique.