Accelerating computer vision-based human identification through the integration of deep learning-based age estimation...

Area of Science:

• Forensic Science
• Artificial Intelligence
• Medical Imaging

Background:

• Computer Vision (CV)-based human identification using orthopantomograms (OPGs) aids in identifying unknown deceased individuals by comparing postmortem OPGs with antemortem CV databases.
• Increasing CV database sizes lead to extended processing times, necessitating more efficient identification methods.

Purpose of the Study:

• To develop a standardized and reliable Convolutional Neural Network (CNN) for age estimation from OPGs.
• To integrate this CNN into the CV-based human identification workflow to enhance processing efficiency.

Main Methods:

• A CNN was trained on 50,000 OPGs with ages ranging from 2 to 89 years.
• The CNN was tested on postmortem OPGs, a large set of antemortem OPGs, and an additional 70 OPGs within a CV identification context.

Main Results:

• Integration of the CNN for age estimation reduced processing time by up to 96% in a CV database of 105,251 entries.
• Age estimation accuracy showed a mean absolute error (MAE) of 2.76 ± 2.67 years for postmortem OPGs and 3.26 ± 3.06 years for antemortem OPGs.

Conclusions:

• Incorporating CNN-based age estimation into CV identification processes significantly decreases processing time.
• The developed CNN provides reliable age estimation results, improving the efficiency of human identification from OPGs.