A fast 3D face reconstruction method from a single image using adjustable model

Abstract

In this paper, we propose a fast and robust method which uses only a single frontal face image as input to reconstruct a plausible 3D face. Our method mainly consists of three stages: feature point detection, model adaptation in X-Y plane and model adjustment on Z-axis direction. At first stage, we detect some face regions such as face contour and facial components automatically. In these regions, we extract several feature points which can generally describe the structure of face. Subsequently, we apply several deformation processes and optimization procedures on an adjustable 3D face model in the X-Y plane based on these feature points. Finally, we present a method of insertion to obtain a dense and smooth model. Experimental results demonstrate the effectiveness and efficiency of our method as well as the robust adaptation to the complex imaging condition.