Artificial Neural Network Based Multimodal Biometrics Recognition System

Biometric recognition involves measuring unique physiological or behavioural traits of human being. Unimodal biometric system involves measuring single trait but it has several limitations like noisy data, lack of performance, spoofing, etc. To overcome the limitations of unimodal biometric system, this paper proposes a multimodal biometric system consisting of a combination of face, ear (physical traits) and gait biometric (behavioural traits) modalities. The ear has an advantage since it is co-located with the face and hence it can be captured with the same or similar sensor. The Gait recognition has unique advantages over traditional biometric. Advances in sensor technology like miniaturized accelerometers in smart phones and Kinect camera have provided the means to record and analyze gait data from a new point of view. In this work we employ a wavelet transform for feature extraction, which describes the ratio between dark and bright areas. In the recognition stage, we use artificial neural networks to achieve good recognition rate in the presence of wide facial variations. Samples of Face, Ear and Gait datasets from GAID, CASIA, USTB, AR, UWA and ORL database were used to evaluate the performance of the system. The samples are normalized using z-score method for better fusion results. Further, match score fusion approaches were used for fusing the face, ear and gait.