Improving online signature verification by user-specific likelihood ratio score normalization

Online handwritten signature is a behavioral biometric trait with several practical applications. Examples of these applications include access control to personal devices and validation of online transactions. Several research work have been done to improve the performance of online signature verification systems. This paper presents an improvement of a recently proposed online signature verification system by introducing a new user-specific score normalization strategy. This new normalization strategy relies on user-specific log likelihood ratio resulting from the Maximum a Posteriori Adaptation (MAP) of both client and impostor scores modeled a priori by Gaussian mixture distributions. Experimental results on the SUSIG database demonstrate the effectiveness of the proposed strategy. The EER is reduced from 6.2 to 2.8%.