Using Gradient Features from Scale-invariant Keypoints on Face Recognition

As the individual identification, access control and security appliance issues attract much attention, face recognition applications are more and more popular. The challenge of face recognition is that the performance is mainly constrained by the variations of illumination, expression, pose and accessory. And most algorithms which were proposed in recent years focused on how to conquest these constraints. In this paper, an algorithm which combines Principal Component Analysis (PCA), Scale Invariant Feature Transform (SIFT) and gradient features to face recognition is proposed. The feature vectors invariant to image scaling and rotation are firstly extracted by SIFT with a different local gradient descriptor. And PCA is applied to the dimension reduction of the local descriptors for saving the computation time. Then the K-means algorithm is introduced to cluster the local descriptors, and the local and global informations of images are combined to classify human faces. Simulation results demonstrate that PCA-SIFT local descriptors are robust to accessory and expression variations and that these descriptors have better performance than other comparative methods. In addition, PCA-SIFT local descriptors have better computation efficiency than standard SIFT local descriptors because of the dimension reduction of the PCA projection.