Performance evaluation of Transform Domain Diagonal Principal Component Analysis for facial recognition employing different pre-processing spatial domain approaches

Facial recognition using spatial domain Diagonal Principal Component Analysis (DiaPCA) algorithm produces better accuracy compared to the Two Dimensional PCA (2DPCA). Transform Domain - 2DPCA (TD2DPCA) retains the high recognition accuracy of the 2DPCA while considerably reducing storage requirements and computational complexity. In this work, the Transform Domain PCA implementation of the DiaPCA (TDDiaPCA) is presented. All the test results, for noise free and noisy images, consistently confirm the considerable storage and computational savings for different spatial domain pre-processing scenarios while retaining the high recognition rate. The performance is evaluated using ORL, Yale and FERET databases. Sample results are given.