Linearized Alternating Direction Method with Adaptive Penalty for Low-Rank Representation