ICA based face recognition robust to partial occlusions and local distortions

We propose a novel method using a perfectly local facial representation based on ICA. We named our method "LS-ICA method". In the LS-ICA method, basis images are made from their corresponding ICA basis images simply by removing non-salient regions. The LS-ICA basis images display perfectly local characteristics because they contain only locally salient feature regions. This enables us to effectively realize the idea of "recognition by parts" for face recognition. Experimental results using AT&T, Harvard, FERET and AR databases show that the recognition performance of the LS-ICA method outperforms that of PCA and ICA methods especially in the cases of facial images that have partial occlusions and local distortions such as changes in facial expression and at low dimensions.