Human identification is crucial in forensics for the investigation of large-scale disasters such as fires, epidemics, earthquakes, and tsunamis. Even though biometric identification using panoramic dental radiography (PDR) has been the subject of several studies in the literature, further study remains a necessary and challenging issue. In this research, a human identification system was developed based on a convolutional neural network (CNN) and contour transform (CT). The proposed system was implemented on a total of 1540 PDR from 302 individuals. The preprocessing applied to PDRs for enhancing and taking the Region of Interest (ROI). The features were extracted using CT transform. These features were fused with features extracted from the CNN to perform identification. Various models with different numbers of layers were applied as a try and test for the proposed system. Data augmentation is used to enhance the system's results. The experimental results illustrate that the best accuracy is 98.9% achieved by implementing the PDRs of size 224*224 using 16 layers (VGG16) with data augmentation of batch size 64*64 and 200 epochs. The prediction time of the proposed system to test PDR was just 3.1 sec. per image. The proposed system can be used to generate candidate images for critical issues. It will likely help with criminal investigations and people identification in large-scale disasters.
-norm Least Mean Square (