Ensemble recognition model with optimal training for multimodal biometric authentication

Biometric security methods deploy any type of biometric recognition for safety purposes. Different from unimodal biometric system that uses single trait for security, the multimodal biometrical authentication (MBA) system deploys two or additional diverse biometric traits for authentication. In addition, MBA systems incurs high cost and have high level of security, as 2 biometric traits are complex for intruders to counterfeit more than a single trait. A new Ensemble Recognition Model with Optimal Training for Multimodal Biometric Authentication (ERMOTMBA) model is proposed in this work using Self Upgraded NBO (SU-NBO) algorithm. At first, median filtering is deployed to pre-process the palm print image, finger vein image, finger knuckle image, iris image and finger print image. The features extracted for palm print, finger vein, finger knuckle, iris and finger print are as follows: Palm print - extracting statistical features and line features, Finger vein - extracting bifurcation point and improved LGBP, Finger knuckle- Global feature and local feature, Iris- extracting 2D Gabor kernel and polynomial filtering, Finger print- Improved Minutiae feature (Binarization and thinning). After the extraction of feature set, enhanced feature level fusion (FLF) takes place under correlation basis. Finally, EC (with the combination of DBN, CNN and Bi-LSTM) is exploited, whose outputs are averaged to finalize the recognition outcome. Especially, proposing a training model for tuning the EC weights, and that is termed as SU-NBO algorithm.