Abstract
One of the tasks of law enforcement agencies is to find evidence of criminalactivity in the Darknet. However, visiting thousands of domains to locatevisual information containing illegal acts manually requires a considerableamount of time and resources. Furthermore, the background of the images canpose a challenge when performing classification. To solve this problem, in thispaper, we explore the automatic classification Tor Darknet images usingSemantic Attention Keypoint Filtering, a strategy that filters non-significantfeatures at a pixel level that do not belong to the object of interest, bycombining saliency maps with Bag of Visual Words (BoVW). We evaluated SAKF on acustom Tor image dataset against CNN features: MobileNet v1 and Resnet50, andBoVW using dense SIFT descriptors, achieving a result of 87.98% accuracy andoutperforming all other approaches.