Multivariate Analytic Technique for Forensic Human Identification based on Dual Cross Patterns of Hand Radiographs

Abstract

In recent times, the necessity for personal identification systems has increased due to several accidents. Under circumstances of human-made and natural disasters, it is not possible to employ a traditional biometric system. Hence, biometric radiographs of the skull, hands, and teeth are good replacement methods to identify victims. The fundamental intent of the research is to acquire a novel approach for identifying missing and anonymous individuals based on Dual Cross Pattern (DCP) features of hand radiographs. The suggested technique has contains two major steps: feature extraction and classification of the feature vectors. In this paper, an effort is made to find the most adequate classifier between the Classification Tree (CT), Feed forward Neural Network (FNN), Multiclass Support Vector Machine (m-SVM), and k-Nearest Neighbor (k-NN) based on the accuracy of retrieval of 10 adult subjects from the dataset of 300 right-hand radiographs. The classification results attained from simulation and discriminant analysis on a small primary database are encouraging.