Multilinear-CLAFIC methods for tensor data in image recognition

In this paper, we propose M-CLAFIC (multilinear class-featuring information compression) and M-CLAFIC-mu methods for image recognition problems in which data samples are represented by high order image tensors. Operating directly on the tensor data preserves the natural data representation form, and it may yield better classification accuracies. Compared to the classical subspace methods, CLAFIC and CLAFIC-mu, the proposed methods are more robust to the small sample size problem which is widely encountered in image recognition applications. Experimental results on the AR and COIL100 databases show that M-CLAFIC and M-CLAFIC-mu methods can produce successful classification accuracies.