Incremental Tensor by Face Synthesis Estimating for Face Recognition

Hua-Chun Tan,Hao Chen,Wu-Hong Wang,Jian-Wei Shi
DOI: https://doi.org/10.1109/icmlc.2009.5212704
2009-01-01
Abstract:When a new person faces before a tensor-based face recognition system, this person is unable to be recognized, since this person's identity subspaces is not contained in the training data. Although PCA method can figure out this problem by adding new image to the training data, but it cannot maintain the original tensor framework and the merit of multi-factor analysis. In this paper, incremental tensor data by facial synthesis estimating is proposed for face recognition. To make full use of the information of new input person in the tensor framework, facial expression synthesis method is used to estimate the missing tensor data. Then the new tensor is constructed, and the subspace of the new person could be constructed based on the new tensor. Thus, the tensor framework can be used to carry on face analysis of the new person, including face recognition. The experimental results show that the proposed method has average 20.1% higher rate for face recognition compared with batch PCA method.
What problem does this paper attempt to address?