Abstract: Efficient face alignment is the key problem for the face applications on the mobile platform which has limited computing and storage capacity.We studied the problem of fast face alignment on the mobile platform.To reduce the computing and storage requirements for face alignment,sparse constrained cascade regression model was proposed in this paper.Sparse constraint was introduced to learn the regression matrix,which can not only select the robust features,but also compress the model size to about 5% compared to the original model.We further constructed the fast face alignment algorithm on mobile platform based on sparse cascade regression model.First,the facial landmarks on the tip of the nose,the corners of the mouth and eyes are quickly located by binary features after face detection,and face pose is estimated.Face image is rotated to frontal view according to the face pose.Then,the corresponding model (frontal model or profile model) is selected according to the face pose,and cascade regression with sparse constraint is used to face alignment.Extensive experiments show that the alignment method proposed in this paper is effective and efficient with compact model size.On the Samsung smart phone of Note3,the alignment time for each face image is about 10ms,and the size of whole apk is only 4MB,which is suitable for face applications on mobile platform.

Key words: Mobile platform,Fast face alignment,Cascade regression,Sparse constraint