A View-Invariant Feature Learning Model for Cross-View Security Authentication in Mobile Smart Devices

Multi-view data is common in the real-world applications ever since many widely used mobile smart devices could capture and recording data from multi-perspective. One of the applications using multi-view data is for security authentication. Existing security authentication methods used by smart device show unacceptable recognition performance in most of the situations. To overcome the problem, this paper propose a view-invariant feature learning model, which learns a feature subspace to map the two-view data into a common space. Our model exploits two regularization terms constructed with low-rank constraint to dig the global class-structure and the view-invariant structure from different views. Therefore, the complementary information is preserved while redundant information can be removed from the two views. Besides, we develop a numerical iterative scheme to optimize the novel model. Experimental results prove that the proposed method outperforms other methods.