Enhancing the Accuracy of Bimodal Feature Recognition using Deep Learning Techniques

摘要

Human recognition through multimodal biometrics has become an emerging trend. Multimodal biometric advancements achieves high authentication and hence are being attracted by the researchers to ameliorate the performance and accuracy of human recognition. Human face can well exhibit the emotions, behavior, age, expressions and more precisely, it replicates the human brain. Iris biometrics has a high stability rate and cannot be compromised easily. In this paper, a new approach is proposed to enhance the accuracy of human recognition based on the match score fusion of iris and face biometric traits through deep learning feature extractor, Autoencoder and Extreme Learning Machine (ELM) matcher. ELM is executed on the extracted features of face and iris traits to generate the individual match scores. The scores are then fused to amplify the recognition accuracy. The experiments are conducted on CASIA and FEI datasets. The results demonstrated a high level of accuracy and effectiveness of the proposed system.