Quick Identification of Child Pornography in Digital Videos

Abstract

Computer forensics has the main objective to find digital evidences of crimes. One of the most researched digital crimes is the sexual abuse of children, including the production, sharing and possession of child pornographic files. Aiming to quickly detect files of child pornography at crime scenes, the NuDetective Forensic Tool was previously developed and it uses techniques like nudity detection in images and videos, among others. To automatically detect child pornography videos, a prior approach was developed, based on extraction and sampling a fixed number of frames of each video files, independently of the video duration. This work proposes a new adaptive sampling approach, considering the video duration, with the objective to increase the detection rate and/or reduce the runtime. Several experiments were performed and the results proved that this new approach is more appropriate to be used in the automatic detection of child pornographic videos at crime scenes, with detection rates around 87% and a reduction about 45% of the runtime over previous experiments. An experiment with a real forensic case was also performed and proved that the new approach can be used to quickly identify such illegal files at crime scenes.