Adaptive face verification based on visual condition estimation

• Published in: Ji suan ji ying yong yan jiu Vol. 32; no. 12; pp. 3805 - 3809
• Main Authors: Huang, Renjie, Li, Tao, Dou, Yumin, Liu, Feng
• Format: Journal Article
• Language: Chinese
• Published: 01.12.2015
• Subjects: Adaptive algorithms; Algorithms; Face recognition; Facial; Learning; Support vector machines; Visual
• ISSN: 1001-3695
• DOI: 10.3969/j.issn.1001-3695.2015.12.066

For reducing the effects of interferences in face verification in unconstrained environments, this paper proposed an algorithm of adaptive face verification based on visual condition estimation. It firstly cropped a pair of input faces into multiple facial regions. Then it adopted support vector regression to estimate the visual conditions of image pair on each region and select such reliable regions. Finally, by combining multiple technologies of fusion of multi-feature, metric learning and support vector machine, it implemented the classifications to verify the pair of faces on the selected facial regions. The experiment results show the effects of the complicated interferences can be effectively reduced by adaptively selecting facial regions according to the visual conditions between face images. It significantly improves the accuracy of face verification in unconstrained environments.