Multimodal Trusted Identity Recognition Based On Differences Between Natural And Human Behaviors

Under the environment of identity anti-counterfeiting,individual can impersonate others easily by using masks,altered fingerprints and other fake attributes to conceal their real identity,and without advanced technical skills,which brings enormous challenges to physical identity recognition.Moreover,massive fake attributes seriously threaten identity retrieval and management,and the security of recognition makes people worried,which further limits the widespread use of identification systems.Therefore,solving fake attributes is an urgent problem in the development of identity recognition.Aiming at the existence of fake attributes during identification,this thesis summarizes the existing methods of domestic and foreign fake attributes detection,and fully explores the differences of data distribution formed by natural and human behaviors,proposes to detect fake attributes with data analysis,and obtains real identity.The major work includes the following four aspects:(1)We analyze the essential reasons caused wrong recognition results,and attribute to natural and human behaviors.For wrong recognition results,we explore the data distribution caused by two behaviors,and propose that the wrong results caused by natural behavior have proximity relationship,whereas caused by human behavior do not have proximity relationship.Several experiments are designed to verify and show the separability of two behaviors.(2)For the condition that the boundary of fake attribute is less than 49%,a method based on order-of-consensus-calculation is proposed for trusted identity computing.The proposed method first obtains consensus identity,then utilizes the differences of the rank of consensus identity in recognition results to judge the authenticity of attributes,finally acquires real identity and realizes identity computing with fake attributes.Experimental results on dataset contained three attributes demonstrate that the method can detect fake attributes effectively,and compute real identity accurately.(3)We research the peak value of boundary of fake attributes,which is that calculation can be realized,what conditions that the number of fake attributes should meet.By analyzing the recognition distribution between real and fake attributes,we find that the calculation of consensus identity affects the detection accuracy.Therefore,we discuss the relationship between the maximum number of attributes formed with identity gathering in results and the number of real and fake attributes,combined with set theory,the conclusion of boundary peak can be obtained in three aspects,and the research can better guide trusted computing.(4)For the condition that the boundary of fake attributes is less than peak value,a method based on adaptive order-of-consensus-calculation is presented for trusted identity computing.With the increase of fake attributes,the detection effect of original method has decreased.To deal with fake attributes in complex situations,we further explore the differences of data distribution,based on the original method,and update the calculation process of consensus identity.Experimental results on dataset contained five attributes show that the method can detect fake attributes effectively during recognition and obtain trusted identity,which has a better performance.As said above,for the shortcomings of existing detection methods,this thesis studies from the perspective of natural and human behaviors,analyzes the differences of data distribution,and proposes method to solve fake attributes during recognition,and compute real identity.The results on different datasets demonstrate that the proposed method can acquire better performance.Besides,the research method in this paper can deal with different types of fake attributes,even with unknown attacks,and it has significant value to improving the security of identification and anti-attack capability of systems.