Integrated estimation of facial scale and position

Abstract

Face detection incurs the highest computational cost in the process of automatic face recognition. To localize a face having scale variations, there needs to be a trade-off between accuracy and efficiency. In this paper, we integrate estimation of facial position and scale, and we propose a method that estimates facial position in parallel with facial scale. The method is composed of four states: position estimation by global scanning, position estimation by local scanning, scale conversion, and verification. The scale conversion estimates facial scale efficiently. Facial position and scale are estimated by changing these states, and are updated by using a beam search. We demonstrate the advantages of the proposed method through face localization experiments using images taken under various conditions. The proposed method can accurately localize the face having scale variations at a small computational cost.