A Non-template based Automatic Landmarking on 3-Dimensional (3D) Faces
Face landmarking is an important task in many face image processing approaches. Detecting and localising landmarks from face data are often performed manually by trained and experienced experts. The appearance of facial landmarks may vary tremendously due to facial expressions (such as opened/closed...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2022
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/38326/1/Pui%20Suk%20Ting%20ft.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Face landmarking is an important task in many face image processing approaches. Detecting and localising landmarks from face data are often performed manually by trained and experienced experts. The appearance of facial landmarks may vary tremendously due to facial expressions (such as opened/closed eyes and mouth) and pose variations. Therefore, this process is not only laborious but also prone to inaccuracies. The developed automatic landmarking model is currently not robust in such unconditional environments. A novel non-template based automatic landmarking model on 3D face data is presented. This can work robustly on faces with variants of facial expressions at different intensities. This model is referred as Mean-Bound-Cluster Automatic Face Landmarking in 3D, MBC-AFL3D. It consists of two main processes which are detection and localisation in the model. The detection process involves mean (H) surface curvature and 3D bounding box segmentation to extract distinct features from face, while the localisation process implies K-means Clustering to search for the best landmark of the face. Experiments show that landmarks found with MBC-AFL3D are on average between 3.17 mm to 9.78 mm as compared to manual landmarking. It is also demonstrated robustness of its method on landmarking faces with high expression variations even with mouth opened. The MBC-AFL3D has improved the accuracy and performance on automatic 3D face landmarking with face expressions. |
|---|