Automated craniofacial landmarks detection on 3D image using geometry characteristics information.

Area of Science:

• Medical Imaging
• Computer Vision
• Anthropometry

Background:

• Indirect anthropometry (IA) relies on manual landmark plotting on 3D facial images, which is time-consuming and prone to human bias.
• This variability in manual measurements can lead to significant errors, especially in large datasets.
• Automating landmark detection is crucial for improving measurement accuracy and efficiency.

Purpose of the Study:

• To develop and validate an automated craniofacial landmarks (ACL) system for 3D facial images.
• To compare the accuracy and efficiency of the ACL system against manual IA methods.
• To assess the system's suitability for automated facial recognition.

Main Methods:

• Developed an automated system (ACL) using geometric characteristics to detect eight craniofacial landmarks (nasion, pronasale, subnasale, alare, labiale superius, stomion, labiale inferius, chelion) on .obj 3D facial models.
• Performed manual IA using Mirror software for comparison.
• Extracted eight linear measurements from both methods and conducted paired t-tests on data from 60 subjects to assess validity between subjects and between methods.

Main Results:

• The ACL system demonstrated accurate detection for nasion, subnasale, pronasale, stomion, labiale superius, and labiale inferius landmarks.
• Seven linear measurements showed statistical significance (p < 0.05) when comparing ACL and IA.
• The ACL method proved more accurate than IA, with a p-value of approximately 0.03.

Conclusions:

• The automated craniofacial landmarks (ACL) system is validated for eight key facial landmarks, proving suitable for automated facial recognition.
• ACL offers a practical and valid alternative to indirect anthropometry (IA).
• The automated system is time-saving and eliminates human bias, enhancing measurement reliability.