基于时延深度神经网络的耳纹身份识别OA

An individual identification method is proposed based on earprint to make up for the deficiency of traditional biomet-ric features in terms of recognition security.This approach employs headphones to capture echo signals from the external auditory canal and then takes this auditory canal environment as an attenuation filter,serving for earprint recognition through analyzing the attenuation characteristics of the echoes across different frequency bands.Firstly,a comprehensive statistical analysis of the spectral characteristics about earprint is conducted to identify discriminative frequency bands that exhibit significant inter-individual variation.Consequently,their corresponding Mel-frequency cepstral coefficient filters are designed and aim at enhanc-ing the extraction efficiency of earprint features.Then,the X-vector model based on the time-delay deep neural network is im-proved to adapt to the identity recognition task in the earprint scenario.The experimental results show that the accuracy of indi-vidual identification utilizing earprint alone can achieve a remarkable 98.7％,thereby substantiating the efficacy of earprint fea-tures in identity verification.To further bolster individual identification security,this paper performs experiments involving joint identification that integrates both earprint and voiceprint features.The experimental results show that it can increase the accuracy of traditional voiceprint recognition from 96.3％to 98.9％for combined earprint soundprint recognition.This indicates the signifi-cance of supplementing earprint information in the recognition system and also demonstrates that earprints identification are more reliable than voiceprints.