The performance of traditional face recognition approaches is sharply reduced when encountered with a low-resolution (LR) probe face image. The basic idea of a face super-resolution (SR) is to desire a high-resolution (HR) face image from an observed LR one with the help of a set of training examples. In this paper, we propose a locality-constrained iterative matrix regression (LCIMR) model for face hallucination task and use the alternating direction method of multipliers to solve it. LCIMR attempts to directly use the image matrix to compute the representation coefficients to maintain the essential structural information. A locality constraint is also enforced to preserve the locality and the sparsity simultaneously. Moreover, LCIMR iteratively updates the locality similarities and reconstruction weights based on the result (the hallucinated HR patch) from previous iteration, giving rise to improved performance. Experimental results on the benchmark FEI face database show the superiority of the proposed method over some state-of-the-art algorithms.
CITATION STYLE
Gao, G., Pang, H., Wang, C., Li, Z., & Yue, D. (2017). Locality-Constrained Iterative Matrix Regression for Robust Face Hallucination. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10636 LNCS, pp. 613–621). Springer Verlag. https://doi.org/10.1007/978-3-319-70090-8_62
Mendeley helps you to discover research relevant for your work.