Modeling Skull-Face Anatomical/Morphological Correspondence for Craniofacial Superimposition-Based Identification
IEEE Transactions on Information Forensics and Security
Craniofacial superimposition (CFS) is a forensic identification technique which studies the anatomical and morphological correspondence between a skull and a face. It involves the process of overlaying a variable number of facial images with the skull. This technique has great potential since nowadays the wide majority of the people have photographs where their faces are clearly visible. In addition, the skull is a bone that hardly degrades under the effect of fire, humidity, temperature
... , etc. Three consecutive stages for the CFS process have been distinguished: the acquisition and processing of the materials; the skull-face overlay; and the decision making. This final stage consists of determining the degree of support for a match based on the previous overlays. The final decision is guided by different criteria depending on the anatomical relations between the skull and the face. In previous approaches, we proposed a framework for automating this stage at different levels taking into consideration all the information and uncertainty sources involved. In this study, we model new anatomical skull-face regions and we tackle the last level of the hierarchical decision support system. For the first time, we present a complete system which provides a final degree of craniofacial correspondence. Furthermore, we validate our system as an automatic identification tool analyzing its capabilities in closed (known information or a potential list of those involved) and open lists (little or no idea at first who may be involved) and comparing its performance with the manual results achieved by experts, obtaining a remarkable performance. The proposed system has been demonstrated to be valid for sortlisting a given data set of initial candidates (in 62,5% of the cases the positive one is ranked in the first position) and to serve as an exclusion method (97,4% and 96% of true negatives in training and test, respectively).