Component based representation for face recognition

In this paper we propose a component based face recognition method. In the framework, the facial landmarks is detected by using a view invariant AAM, and the components including eyes, nose, and mouth are extracted. To improve recognition rate, the components are aligned by using the procruetes analysis. After aligning, the component is insensitive to translation, scale, and rotation. The face features are extracted by applying a random measurement matrix on the input component image. The obtained features are in a low dimensional space which saves computation time. For face recognition, Gradient projection is used to resolve the convex problem in norm minimization and determine the class to which the test face image belong. Finally, the presented method is demonstrated on the Extended Yalu face database. The results demonstrate that the method performs well in face recognition.