Unsupervised Feature Extraction Inspired by Latent Low-Rank Representation

Latent Low-Rank Representation (LatLRR) has the em-pirical capability of identifying “salient ” features. How-ever, the reason behind this feature extraction effect is still not understood. Its optimization leads to non-unique so-lutions and has high computational complexity, limiting its potential in practice. We show that LatLRR learns a transformation matrix which suppresses the most significant principal components corresponding to the largest singular values while preserving the details captured by the compo-nents with relatively smaller singular values. Based on this, we propose a novel feature extraction method which directly designs the transformation matrix and has similar behav-ior to LatLRR. Our method has a simple analytical solu-tion and can achieve better performance with little compu-tational cost. The effectiveness and efficiency of our method are validated on two face recognition datasets