User individuality based cost-sensitive learning: A case study in finger vein recognition

State-of-the-art cost-sensitive learning based techniques in biometrics ignore cost difference between users and determine the loss only based on the misrecognition category. In practice, this may not always hold and the user individuality may also affect the loss of misrecognition. For example, misrecognizing an imposter as an administrator can cause a much more serious loss than misrecognizing it as a normal user. At the same time, two administrators/normal users may have different probability to accept imposter. To confidently prevent the high-probability error, the cost of false acceptance for one user with a high probability should be larger than it for the other users. To make cost definition more reasonable and further lower misrecognition cost of a recognition system, we propose to incorporate the user individuality, i.e., user role and user gullibility, into the traditional cost-sensitive learning model through defining an improved object function. By employing the new model, we further develop a user role and gullibility based mckNN (rg-mckNN). Experimental results on finger vein databases demonstrate the effectiveness of the proposed method.