Quality-Aware Estimation of Facial Landmarks in Video Sequences

Abstract

Face alignment in video is a primitive step for facial<br/>image analysis. The accuracy of the alignment greatly<br/>depends on the quality of the face image in the video<br/>frames and low quality faces are proven to cause<br/>erroneous alignment. Thus, this paper proposes a system<br/>for quality aware face alignment by using a Supervised<br/>Decent Method (SDM) along with a motion based forward<br/>extrapolation method. The proposed system first extracts<br/>faces from video frames. Then, it employs a face quality<br/>assessment technique to measure the face quality. If the<br/>face quality is high, the proposed system uses SDM for<br/>facial landmark detection. If the face quality is low the<br/>proposed system corrects the facial landmarks that are<br/>detected by SDM. Depending upon the face velocity in<br/>consecutive video frames and face quality measure, two<br/>algorithms are proposed for correction of landmarks in<br/>low quality faces by using an extrapolation polynomial.<br/>Experimental results illustrate the competency of the<br/>proposed method while comparing with the state-of-theart<br/>methods including an SDM-based method (from<br/>CVPR-2013) and a very recent method (from CVPR-2014)<br/>that uses parallel cascade of linear regression (Par-CLR).