Multiview Active Shape Models with SIFT Descriptors for the 300-W Face Landmark Challenge

Abstract

The 2013 iBUG 300-W Challenge tests the ability of participants' software to locate landmarks in unseen images of faces. We approach the challenge with Active Shape Models (ASMs) that incorporate a modified version of SIFT descriptors. We make use of multiple ASMs, searching for landmarks with the ASM that best matches the face's estimated yaw. This simple approach is shown to be effective on a variety of datasets.