Face Pose Estimation with Ensemble Multi-scale Model and Deep Learning

Abstract

Face pose estimation plays important roles in broad applications, including visual based surveillance, face authentication, expression automatic understanding, human-computer intelligent interactions, etc. However, face pose estimation is also a challenge issue, especially under complicated real application environments. In this paper, a novel face pose estimation approach with ensemble multi-scale model and deep learning is presented. First, using VGG-Face CNN [1] as backbone three middle scale layer outputs are extracted and go through additional transfer learning adapting to pose estimation task, then a multi-layer ensemble sub-model can be formed. Similarly, a multi-scale Curvelet features are extracted and then another ensemble sub-model can also be formed. These two sub ensemble model predictions can be integrated as a weighting combination, and a final ensemble model for face pose estimation is built. The experiment results testing on public large face databases CAS-PEAL show that the proposed approach achieved estimation accuracy above 99.5% for yaw angle estimation and above 99.7% for pitch angle estimation.