Reduction of genuine and imposter score overlapping based on intra-class variations and deviation of scores in a multimodal biometrie system

Many score normalization and transformation techniques [2, 4, 13] have been introduced in score level fusion of multimodal biometric traits for enhanced authentication and security. Such normalization techniques help in fusing heterogeneous scores at matching score level. This paper represents a novel approach which produces distributed scores without application of normalization and quantization techniques thus eliminating the overhead of such algorithms. The proposed method considers real time face vibrations and intra-class variations in face modality. It has been observed that standard deviation of genuine scores and standard deviation of imposter scores for continuous stream of face inputs distributes the scores and eliminates the problem of overlapping. Standard deviation of genuine scores and standard deviation of imposter scores are sufficient for face modality and can be scaled to a new score by scalar weight multiplication for further fusion with other biometric trait scores. The proposed algorithm of standard deviation algorithm improves GAR significantly and reduces FAR to an extremely mitigated value as desired.