Dayong Wang , Nanyang Technological University, Singapore
Steven C.H. Hoi , Nanyang Technological University, Singapore
Ying He , Nanyang Technological University, Singapore
Jianke Zhu , Zhejiang University, Hangzhou
Tao Mei , Microsoft Research Asia, Beijing
Jiebo Luo , University of Rochester, Rochester
Retrieval-based face annotation is a promising paradigm of mining massive web facial images for automated face annotation. This paper addresses a critical problem of such paradigm, i.e., how to effectively perform annotation by exploiting the similar facial images and their weak labels which are often noisy and incomplete. In particular, we propose an effective Weak Label Regularized Local Coordinate Coding (WLRLCC) technique, which exploits the principle of local coordinate coding in learning sparse features, and employs the idea of graph-based weak label regularization to enhance the weak labels of the similar facial images. We present an efficient optimization algorithm to solve the WLRLCC task. We conduct extensive empirical studies on two large-scale web facial image databases: (i) a Western celebrity database with a total of $6,025$ persons and $714,454$ web facial images, and (ii)an Asian celebrity database with $1,200$ persons and $126,070$ web facial images. The encouraging results validate the efficacy of the proposed WLRLCC algorithm. To further improve the efficiency and scalability, we also propose a PCA-based approximation scheme and an offline approximation scheme (AWLRLCC), which generally maintains comparable results but significantly saves much time cost. Finally, we show that WLRLCC can also tackle two existing face annotation tasks with promising performance.
Face, Encoding, Optimization, Vectors, Sparse matrices, Image databases, Image coding, Image/video retrieval, Information Search and Retrieval, Image databases, Mining methods and algorithms, Multimedia databases, Systems, Database Management, Information Technology and Systems
T. Mei, J. Zhu, Y. He, S. C. Hoi, D. Wang and J. Luo, "Retrieval-based Face Annotation by Weak Label Regularized Local Coordinate Coding," in IEEE Transactions on Pattern Analysis & Machine Intelligence.