Face Recognition from One Sample per Person

As one of the most visible applications in computer vision communication, face recognition (FR) has become significant role in the community. In the past decade, researchers have been devoting themselves to addressing the various problems emerging in practical FR applications in uncontrolled or less controlled environment. In many practical applications of FR (e.g., law enforcement, e-passport, ID card verification, etc.), there is only one sample per person. Face recognition (FR) with a One Sample per Person is a very challenging problem due to the lack of information to predict the variations in the query image. The number of training samples per person will greatly affect the performance of face recognition. In the case of One Sample per Person, existing manifold learning algorithms cannot be directly applied, since those algorithms need multiple training samples per person to represent the query face. To address the problem of One Sample per Person, we propose in this paper a novel multi manifold analysis method by learning discriminative features from each manifold. In this technique, each recorded face image is first detected. Detected face image is then partitioned into several non overlapping patches to form an image set where each set contains images belonging to the same subject. By modelling each image set as a manifold, we formulate the One Sample per Person face recognition problem as the computation of the distance between two manifolds and learn multiple feature spaces to maximize the manifold margins of different persons. Finally, we present a reconstruction-based manifold-manifold distance where identification is achieved by seeking the minimum manifold-manifold distance (MMD) from the probe to the gallery of image sets.