Robust feature extraction for face recognition based on ultrasonic sensing

Abstract

Ultrasonic sensing could be used to recognize objects with complex shapes like human faces. The high resolution range profile (HRRP), also called ultrasonic sonar image, is obtained by demodulating the faces' scatter (echo signal) of detection ultrasonic signal. Though HRRP feature represents the geometry of the human face, it's unstable because of the aspect sensitivity. In this paper, the HRRP and the robust feature extraction approach of ultrasonic sensing are investigated. The cause of HRRP sensitivity is analyzed based on the multi scatter centers model. And an average power feature extraction method is proposed. Experimental results show that the averaged HRRP feature is stable with the change of facial expressions and could be used in face recognition application. The feature extraction approach could also be used in other ultrasonic sonar detection and recognition applications.