Combining multiagent negotiation and an interacting verification process to enhance biometric-based identification

Designing a biometrics-based system poses many challenges, such as how many and which modalities to use (multimodal configurations being widely adopted), which classification methods are appropriate, user acceptability issues, and so on. Machine learning techniques need as much information as possible to maximise accuracy, but biometric samples are not necessarily straightforward to acquire and usability factors can be very influential. This paper presents a new recognition structure for biometric systems design, using an agent-based approach which maximises the value of the available information. Using handwritten signature as an illustrative modality, we present results which show that carefully structured unimodal systems can deliver excellent performance.