Rapid determination of age classification
Law enforcement agencies that employ Computer Forensics teams often find themselves involved in cases of child pornography. Officers who investigate these cases are required to sort through thousands and sometimes tens of thousands of digital images in order to find the evidence they need. This manual search is time consuming and not an optimal use of the investigator's time. In order to alleviate some of this burden, law enforcement has expressed an interest in an automated tool that can identify potential contraband images. This dissertation outlines the research and implementation of a method that will aid in the identification of images of interest to computer forensics investigators. This project uses open source libraries to find images with faces, and from those faces, extract the main facial features. To achieve the highest rate of success in finding the face and the facial features, it is necessary for the subject's full face to be clearly present. It is desirable for there to be very little horizontal and vertical rotation of the head, however, the feature detection method allows for some slight rotation. Calculations are performed on the position of the features resulting in an attribute set which is used by machine learning to determine if the face is that of a child or an adult. This study shows that the ideal separation of these classes is with children 12 years of age and under and adults 18 and over. The resulting methodology described in this paper achieves an accuracy rate of 70% separation of images into the child and adult classes, and it processes images at a rate of approximately one second per image.
Timothy H Ball,
"Rapid determination of age classification"
Dissertations and Master's Theses (Campus Access).