Exploring Facial Asymmetry Using Optical Flow

Human faces are highly but not precisely bilaterally symmetrical. We present in this letter a measurement of facial asymmetry based on the optical flow field between a face image and its bilaterally mirrored counterpart. We revisit the problem of facial asymmetry quantification and confirm some conclusions in previous research by applying the proposed asymmetry measurement on a dataset containing 4000 subjects. Moreover, the proposed measurement also contains information on facial asymmetry compensation and thus can be used to facilitate various face processing tasks such as face image beautification, 3D face reconstruction and face recognition. Experimental results show the flexibility and effectiveness of our proposal.