A novel approach for classifying faces of the upper and lower Himalayan region using CNN- implications in forensic...

With the advent of advanced computational methods, facial identification has evolved from traditional anthropometric techniques to machine-learning and deep-learning models capable of recognizing the distinctive morphological features of the face. The human face is one of the most distinctive features of an individual's identity, providing investigators with a non-invasive means of narrowing down the suspects and identifying the victims. The present study explores the application of Convolutional Neural Networks (CNNs) for the classification and identification of Himalayan populations using facial features. Human faces from the upper and lower Himalayan regions of Himachal Pradesh State in the northern India were studied. A novel AI model has been developed to enhance facial recognition capabilities. The developed AI model particularly focuses on classifying individuals from the upper and lower Himalayan regions based on their facial features. Morphologically, the upper Himalayan population groups can be distinguished from the lower Himalayan population groups in their overall facial shapes, shape and size of the nose and eyes, and so on. The phenotypic differences observed in a population may vary due to both intra-observer and inter-observer assessments. The CNN models have the potential to extract peculiar facial features with a high level of accuracy. In the present study, the designed model yielded a training accuracy of 89.32% and the validation accuracy of 86.00%. Moreover, the highest validation accuracy of 88.02% was achieved at epoch 109, indicating its reliability and effectiveness in facial classification. In addition, the individual and average Grad-CAM analysis was performed to visually confirm that which region is contributing more to the CNN classification. The customized AI model shows its potential applications for individual identification, ethnicity classification, facial identification research work, security and surveillance, forensic examinations, border control, and crime scene investigations.