Android-based multimodal biometric identification system using feature level fusion

Nowadays designated as one of the final frontiers, biometrie identification is widely researched and rapidly developed. However, when being adopted in real world, single modality is found to have numerous limitations such as insecure and unreliable. To overcome these weaknesses and enhance the convenience when person identification is in urgent needs in outside world, we designed a novel method to identify an already registered user or grant authorization using multimodal biometrics with face and voice traits on android devices. Both face feature vector and voice feature vector are extracted independently using haar-wavelet transform and then are fused at feature level. At last, we use support vector machine (SVM) to perform the binary classification. The experiments results indicate that our system can obtain a satisfactory performance giving identification accuracy of 93.6% and can be used in financial field, where information security is foremost. Further comparison on experiments results also show that our proposed system is more reliable than other similar multimodal identification system.