Forensic identification using siamese, transfer learning and custom deep learning models

Forensic identification from human skulls plays a crucial role in forensic anthropology, providing critical insights into the determination of sex, ancestry, and other important characteristics of unidentified human remains. In recent years, deep learning techniques have emerged as powerful tools for automated analysis and classification tasks in various fields, including computer vision and pattern recognition. This study explores the potential of various deep learning models, including our proposed Siamese Neural Network's and other deep learning (DL) models such as VGG-16, custom convolutional neural network (CNN), ResNet50, DenseNet, MobileNet, InceptionV3, EfficientNet, and AlexNet learning for forensic identification from human skulls. These models were applied to skull images obtained from Digital Imaging and Communications in Medicine (DICOM) files on the New Mexico Decedent Image Database (NMDID) dataset. The findings reveal that the proposed siamese neural network method can effectively detect forensic identification of people with a high accuracy rate as 85.33% and that it outperforms the state-of-the-art methods in the literature.