Learning Based Approach for Facial Fiducial Points Prediction of Non Frontal Faces

Facial features points (FFP) are used for localization and representation of the prominent features of the face that are traced on tips, corners, and mid-points of facial segments. The Proposed algorithm to detect facial landmarks on frontal and non-frontal faces on video sequences. It syndicates a regression-based approach with an advanced block matching technique. In the novel methodology, it will be shown that how trees are used to take the estimated position of facial landmarks, straight from a scant subsection of pixel intensities with less time and by handling missing labeled data, especially for non-frontal images. The Proposed algorithm, encompasses the regression-based model to provide a quality measure of each prediction and use the shape model to restrict and correct the sampling region. Our approach is an extension of the existing LEAR that combines the low computational cost with the selection of important features based on 22 facial points. The proposed algorithm is tested on five datasets. Results presented significant enhancement over the existing approach.