Individual Tensorface Subspaces for Efficient and Robust Face Recognition that do not Require Factorization

Facial images change appearance due to multiple factors such as poses, lighting variations, facial expressions, etc. Tensor approach, an extension of the conventional 2D matrix, is appropriate to analyze facial factors since tensors make it possible to construct multilinear models using multiple factor structures. However, tensor algebra provides some difficulties in practical usage. First, it is difficult to decompose the multiple factors (e.g. pose, illumination, expression) of a test image, especially when the factor parameters are unknown or are not in the training set. Second, for face recognition, as the number of factors is larger, it becomes more difficult to construct reliable multilinear models and it requires more memory and computation to build a global model. In this paper, we propose a novel Individual TensorFaces which does not require tensor factorization, a step which was necessary in previous tensorface research for face recognition. Another advantage of this individual subspace approach is that it makes the face recognition tasks computationally and analytically simpler. Based on various experiments, we demonstrate the proposed Individual TensorFaces bring better discriminant power for classification.