Efficient face verification in mobile environment using component-based PCA

While face verification technology is proving its value on the security of smartphones, it finds a more suitable environment of implementation than on desktop computers. Targeting to the “close-range frontal” photos taken by the front camera of smartphones, an efficient face verification approach is proposed in this paper. A dedicated rule-based algorithm is first implemented to detect four facial feature points which are used to align the input face images and partition the face region into four components. Based on each group of facial components obtained from the training dataset, an eigen subspace is constructed through principal component analysis(PCA). Finally the weighted sum of correlations between each input face component and its back-projection onto the subspace is calculated to measure the similarity of the input person against that in the dataset. Experiments are conducted on a dataset with 464 face images taken from 9 persons with variable illumination, background and expression. The Experimental results prove a 98.2% of accuracy on feature detection, a 8.5% of EER on face verification and the computational time being less than 0.8 seconds on a personal computer.