Addressing privacy constraints for efficient monitoring of network traffic for illicit images
Amin Ibrahim, Miguel Vargas Martín · 2009
The sexual exploitation of children remains a very serious problem and is rapidly increasing globally through the use of the Internet. This paper focuses on the privacy issues involved in design and implementation of a system capable of image classification at the network layer. In this paper, we examined two learning algorithms, namely the Maximum Likelihood Estimator (MLE), and the Stochastic Learning Weak Estimator (SLWE) as well as six distance measures including the Euclidian Distance (ED), the Weighted Euclidian Distance (WED), and the Cosine Distance (CosD). Our experiments indicate that the SLWE algorithm has slightly better classification accuracy than MLE and as a result the SLWE algorithm combined with a Linear Classifier can be used to actively filter illicit pornographic images as they are transmitted over the network layer.