We are glad to announce that our journal paper “Pornography classification: The hidden clues in video space–time” has been accepted at Forensic Science International Journal, Volume 268 , 46-61, 2016.
In the paper, the authors deal with an important issue nowadays: the many contexts in which pornographic content is unwelcome, especially where underage viewers are concerned. The reseach proposes an end-to-end BoVW-based framework of video-porno-graphy classification, allowing to incorporate temporal information in different ways. To perform experiments and validation, it introduces the Pornography-2k dataset, a new 65 challenging pornographic benchmark that comprises 2,000 web videos and 140 hours of video footage, available upon request. The paper also introduces Temporal Robust Features (TRoF), a novel space-temporal interest point detector and descriptor, which provides a speed compatible with real-time video processing and presents low-memory footprint.
The obtained results confirmed that the incorporation of space-temporal information leads to more effective video-pornography classifiers and observed that the dense low-level video descriptions increase the system effectiveness (accuracy), but at prohibitive reductions in efficiency (computational time and memory footprint).