Ear biometrics studies using electroacoustic transducers in headsets

Abstract

This master’s thesis studies the potential of ear biometrics using acoustical waves. The goal is to know whether it is possible and accurate to authenticate someone with a system using headphones and microphones on ears. The idea is to send a signal with the headphones in the ears of the subject and get the reflections with microphones. These reflections depend on the shape of the ear and therefore could be used to authenticate someone. For this research, different handcrafted prototypes have been manufactured. By placing microphones in different headphones. Some physical expectations of the measurable data have been studied allowing to see the differences that could occur between people. Then some databases with different users have been established, a pre-processing scheme of the data has been discussed and two recognition algorithms have been implemented and tested on these databases. The different parameters for these algorithms have been studied in order to determine the best way to recognize people correctly and get optimal parameters. The results were quite satisfying and in accordance with previous studies. The efficiency depends on the headset but the best obtained EER value was equal to 98.99 %. As an additional work, the detection of the correct positioning of the headset has also been briefly studied. This detection was really good for some people while for others it was not. Finally, further work and ideas for future studies are proposed and presented.