3D Facial Biometric Verification Using a DNA Sample for Law Enforcement Applications.

A large majority of violent crimes such as homicides, sexual assaults, and missing person cases are not solved within a reasonable timeframe and become cold cases. The ability to predict a person's facial appearance from a DNA sample may generate important investigative leads and provide an unprecedented advancement in criminal investigations. To achieve the above goal, it is first essential to substantiate, model and measure the intrinsic relationship between the genomic markers and phenotypic features. In the first step, we have standardized the 3D face scans using a widely used 3D data format - CoMA. The standardization was followed by its projection into a low-dimensional latent embedding space. The second step was to reduce the dimensionality of the genetic space. The dimensionality reduction was achieved by performing Principal Component Analysis on the genomic markers to generate compact genomic properties. A simple multi-layer perceptron was trained to classify an ensemble of facial embeddings and genomic properties into genuine and imposter pairings. The classification model could match the DNA with the given 3D face with an average Area Under the Curve score of 0.73. The introduction of hand-picked genomic markers was an important contribution toward improving the final AUC score. Furthermore, results indicated that incorporating additional phenotypical properties such as sex and age leads to better verification. Thus, this study represents an important milestone toward building a functional machine learning pipeline capable of predicting facial appearance and other visible traits from a DNA sample.