Computer Vision News - October 2022

21 Xi Fang few millimeters this way, and I set back the lower jaw a few millimeters that way, your face is going to look like this, but there is no scientific way to predict how the face is going to react after the bony surgery.” This project was initiated around a decade ago, and James recently worked with Daeseung on a paper exploring a novel approach using the finite-element method (FEM) to simulate facial changes. FEM is a numerical biomechanics-based method reported to be the most common and accurate for simulating or analyzing mechanical changes in body structure. However, its prediction accuracy in clinically critical regions, such as the lips, was below the acceptable range. “We proposed to improve the accuracy along this clinically critical area with a method called incremental facial change simulation with realistic lip sliding effect,” Daeseung tells us. “We focused on the sliding effect of the soft tissue on the bone. We needed to accurately simulate the interaction between the bony structure and soft tissue to predict the facial changes.” Most previous works used a simple simulation condition, assuming bony structure and soft tissue were attached and moved together. The reality is that when bone moves, soft tissue freely slides over its surface, so their paper’s method applied mucosa and lip sliding effects. “Previous methods did not consider the opening of the lips,” Daeseung points out. “In those models, the upper and lower lips were connected, so you could not accurately simulate individual lipmovements following bony movements. We separated the upper and lower lips in our model and and could accurately simulate their movements following each upper and lower jaw movement.” Daeseung found that balancing the accuracy and complexity of the simulation condition was the most challenging aspect. If it was too complex, it used too many computational resources. The realistic lip sliding effect was the perfect balance between accuracy and efficiency. Facial changesweresimulated littleby little,which made it more natural, and it demonstrated a significant improvement over previous methods, improving prediction accuracy around the lips in both quantitative and qualitative measures. Despite these positive results, FEM still presented some challenges, particularly OF AI