Abstract
Abstract. A technique for automated face detection and its pose estimation using single image is developed. The algorithm includes: face detection, facial features localization, face/background segmentation, face pose estimation, image transformation to frontal view. Automatic face/background segmentation is performed by original graph-cut technique based on detected feature points. The precision of face orientation estimation based on monocular digital imagery is addressed. The approach for precision estimation is developed based on comparison of synthesized facial 2D images and scanned face 3D model. The software for modelling and measurement is developed. The special system for non-contact measurements is created. Required set of 3D real face models and colour facial textures is obtained using this system. The precision estimation results demonstrate the precision of face pose estimation enough for further successful face recognition.
Highlights
A problem of object orientation determination based on single image often arises in image analysis field
There are two main approaches to face pose account in face recognition algorithms: 1) Account of face pose at the stage of biometrical template calculation (Jie et al, 2010, Zhang et al, 2013). 2) 3D-model-based face pose estimation and artificial frontal view generation before the biometrical template calculation (Choi et al, 2010, Kemelmacher-Shlizerman and Basri, 2010, Kemelmacher-Shlizerman et al, 2013)
Two 3D models were used for the accuracy of 2D pose estimation: a mannequin head and a real person head
Summary
A problem of object orientation determination based on single image often arises in image analysis field. The second approach allows using the existing 2D frontal face recognition algorithms for processing of non-frontal face images. It is very popular and useful approach. The photogrammetric approach for precision of face pose estimation is developed too based on comparison of synthesized facial 2D images and scanned face 3D model. This approach allows separating the influence of pose correction errors from the influence of other factors to the final results of biometric face recognition
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
