Content area
Full Text
Int J Comput Vis (2014) 109:94109 DOI 10.1007/s11263-013-0693-1
Domain Adaptation for Face Recognition: Targetize Source Domain Bridged by Common Subspace
Meina Kan Junting Wu Shiguang Shan Xilin Chen
Received: 15 March 2013 / Accepted: 7 December 2013 / Published online: 31 December 2013 Springer Science+Business Media New York 2013
Abstract In many applications, a face recognition model learned on a source domain but applied to a novel target domain degenerates even signicantly due to the mismatch between the two domains. Aiming at learning a better face recognition model for the target domain, this paper proposes a simple but effective domain adaptation approach that transfers the supervision knowledge from a labeled source domain to the unlabeled target domain. Our basic idea is to convert the source domain images to target domain (termed as targetize the source domain hereinafter), and at the same time keep its supervision information. For this purpose, each source domain image is simply represented as a linear combination of sparse target domain neighbors in the image space, with the combination coefcients however learnt in a common subspace. The principle behind this strategy is that, the common knowledge is only favorable for accurate cross-domain reconstruction, but for the classication in the target domain, the specic knowledge of the target domain is also essential and thus should be mostly preserved (through targetization in the image space in this work). To discover the common knowledge, specically, a common subspace is learnt, in which the structures of both domains are preserved and meanwhile the disparity of source and target domains is reduced. The proposed method is extensively evaluated
M. Kan J. Wu S. Shan (B) X. Chen
Key Lab of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology (ICT), CAS, Beijing 100190, Chinae-mail: [email protected]
M. Kane-mail: [email protected]
J. Wue-mail: [email protected]
X. Chene-mail: [email protected]
under three face recognition scenarios, i.e., domain adaptation across view angle, domain adaptation across ethnicity and domain adaptation across imaging condition. The experimental results illustrate the superiority of our method over those competitive ones.
Keywords Face recognition Domain adaptation
Common subspace learning Targetize the sourece domain
1 Introduction
Machine learning has been widely used for various vision tasks, such as image classication, multimedia retrieval, and object recognition, etc. Typically,...