Automatic facial synthesis from single frontal face image for model based video coding

Abstract

In this paper, we propose an algorithm for automatic facial synthesis of a 3D face from a single frontal image. A 3D wire frame face model has been developed by using 2758 triangular surfaces. Complex Wavelet Transform (CWT) based facial feature detection procedure has been employed to automatically locate 20 facial feature points. The automatically detected features are used to map the face image onto the 3D face model. During the mapping process the 3D face model adapts itself to the facial feature locations of the individual. The quality of the generated 3D faces has been evaluated by using qualitative and quantitative measures. Peak Signal to Noise Ratio (PSNR) and Mean Observer Score (MOS) metrics are used to indicate the quality of the synthesized 3D faces.