Human head modeling based on fast-automatic mesh completion

Abstract

The need to rapidly create 3D human head models is still an important issue in game and film production. Blanz et al have developed a morphable model which can semi-automatically reconstruct the facial appearance (3D shape and texture) and simulated hairstyles of new faces (faces not yet scanned into an existing database) using photographs taken from the front or other angles [Blanz et al. 2004]. However, this method still requires manual marker specification and approximately 4 minutes of computational time. Moreover, the facial reconstruction produced by this system is not accurate unless a database containing a large variety of facial models is available. We have developed a system that can rapidly generate human head models using only frontal facial range scan data. Where it is impossible to measure the 3D geometry accurately (as with hair regions) the missing data is complemented using the 3D geometry of the template mesh (TM). Our main contribution is to achieve the fast mesh completion for the head modeling based on the Automatic Marker Setting and the Optimized Local Affine Transform (OLAT). The proposed system generates a head model in approximately 8 seconds. Therefore, if users utilize a range scanner which can quickly produce range data, it is possible to generate a complete 3D head model in one minute using our system on a PC.