Fake identity attributes detection based on analysis of natural and human behaviors

Abstract

Under confrontational environment, individual can impersonate others by wearing masks or other skills to conceal real identity, which brings enormous challenges to physical identity recognition. Moreover, massive fake attributes seriously threaten identity management. Aiming at the existence of fake attributes during multimodal identification, we propose a novel method to detect fake attributes and compute real identity for security of systems. Most previous methods focused on the differences about features between fake and normal attributes, but each method generally targeted one type of fake attribute, and the results decreased a lot with unknown attacks. In this paper, we first explore the essential differences of data distribution caused by natural and human behaviors, then with order-of-consensus-calculation based on the differences, fake attributes are detected by analyzing the rank of consensus identity in recognition results, finally maximum-consensus-calculation is applied to compute real identity for evaluating detection performance. Experimental results on face, fingerprint, and voiceprint demonstrate that the proposed method can detect fake attributes effectively, which has a higher accuracy, and the accuracy of identity recognition is increased obviously by about 13.20% with forgery detection. The additional experiments further confirm the feasibility of the proposed method with increase of fake attributes. Furthermore, the proposed method can deal with different kinds of attacks, even with unknown attacks, and it is also significant to improve the security of identification with complex environment.