Related Experiment Video
Updated: Jan 9, 2026
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
2.4K
Forensic identification using siamese, transfer learning and custom deep learning models
Omer Sevinc1, Mehrube Mehrubeoglu2, Abdullah Asım Yılmaz3
1Computer Programming Department, Ondokuz Mayıs University, 55900, Samsun, Türkiye.
Scientific Reports
|November 29, 2025
Summary
This study introduces a Siamese Neural Network for forensic identification from human skulls, achieving 85.33% accuracy. This deep learning approach surpasses existing methods for analyzing unidentified human remains.
Area of Science:
- Forensic Anthropology
- Computer Vision
- Machine Learning
Background:
- Forensic identification from human skulls is vital for determining characteristics of unidentified remains.
- Deep learning (DL) models offer advanced capabilities for automated analysis and pattern recognition.
Purpose of the Study:
- To evaluate the efficacy of various deep learning models for forensic identification using skull images.
- To introduce and assess a novel Siamese Neural Network for this task.
Main Methods:
- Applied multiple DL models (VGG-16, CNN, ResNet50, DenseNet, MobileNet, InceptionV3, EfficientNet, AlexNet, and a proposed Siamese Neural Network) to skull images.
- Utilized skull images from the New Mexico Decedent Image Database (NMDID) in DICOM format.
Main Results:
- The proposed Siamese Neural Network achieved a high accuracy rate of 85.33% for forensic identification.
- The Siamese Neural Network demonstrated superior performance compared to other state-of-the-art DL methods in the literature.
Conclusions:
- Deep learning, particularly the Siamese Neural Network, shows significant potential for accurate automated forensic identification from human skulls.
- This approach can enhance the analysis of unidentified human remains in forensic anthropology.
