Sparse Learning for Face Recognition with Social Context

摘要

Face recognition in uncontrolled environments, such as large pose variations, and extreme ambient illumination and expressions is a challenging task for traditional face recognition methods. Some recent works show that context information such as clothes and social relationships is very important for solving the problem. Furthermore, sparse representation-based method is very robust for face occlusion and pixel contamination. In this paper, the authors propose a sparse learning framework for face recognition with social context. First, sparse representation based dimensionality reduction method is used to find the low dimension representation of the face images. Second, sparse representation based classification is utilized to classify test face images as known label and unknown label. Third, for test images classified as unknown label, social context descriptor is constructed according to co-existence information. Finally, based on social context descriptor and the low dimension representation of the test images classified as unknown label, sparse representation based clustering is adopted to perform face recognition.