Facial Model Reconstruction for Plastic Surgery Simulation

Abstract

In facial surgery simulation, it usually requires the patient’s CT (Computed Tomography) data for building the facial model. However, CT imaging is expensive, time consuming, and the patients have to be under radiation exposure as well. Thus, it is undesirable to undertake CT examination for minor surgeries; instead, we propose a new three-dimensional (3D) facial model reconstruction algorithm based on color stereovision techniques. The proposed technique provides a non-radioactive alternative for facial model reconstruction. In the proposed system, a color-coded grid pattern is first adopted to simplify the establishment of stereo correspondence. Because human facial surfaces are usually with few image features, the generated grid greatly facilitates the correspondence detection and is also robust due to its fault-tolerance capability. In addition, we setup four cameras around the patient and image the subject simultaneously to enlarge the field of view covering the whole facial areas. The resulting models from all three camera pairs were registered and integrated by the least square method. Experimental results have shown that the proposed algorithm successfully reconstructs the full facial model and recovers the 3D information with efficient computation.