@InProceedings{PerezTestRoch:2017:ViPoDe,
author = "Perez, Mauricio Lisboa and Testoni, Vanessa and Rocha, Anderson",
affiliation = "{EEE - NTU} and {Samsung Research Institute Brazil} and {IC -
UNICAMP}",
title = "Video pornography detection through deep learning techniques and
motion information",
booktitle = "Proceedings...",
year = "2017",
editor = "Torchelsen, Rafael Piccin and Nascimento, Erickson Rangel do and
Panozzo, Daniele and Liu, Zicheng and Farias, Myl{\`e}ne and
Viera, Thales and Sacht, Leonardo and Ferreira, Nivan and Comba,
Jo{\~a}o Luiz Dihl and Hirata, Nina and Schiavon Porto, Marcelo
and Vital, Creto and Pagot, Christian Azambuja and Petronetto,
Fabiano and Clua, Esteban and Cardeal, Fl{\'a}vio",
organization = "Conference on Graphics, Patterns and Images, 30. (SIBGRAPI)",
publisher = "Sociedade Brasileira de Computa{\c{c}}{\~a}o",
address = "Porto Alegre",
keywords = "Pornography classification, Deep learning and motion information,
Optical flow, MPEG motion vectors, Sensitive video
classification.",
abstract = "Recent literature has explored automated pornographic detection -
a bold move to replace humans in the tedious task of moderating
online content. Unfortunately, on scenes with high skin exposure,
such as people sunbathing and wrestling, the state of the art can
have many false alarms. This paper is based on the premise that
incorporating motion information in the models can alleviate the
problem of mapping skin exposure to pornographic content, and
advances the bar on automated pornography detection with the use
of motion information and deep learning architectures. Deep
Learning, especially in the form of Convolutional Neural Networks,
have striking results on computer vision, but their potential for
pornography detection is yet to be fully explored through the use
of motion information. We propose novel ways for combining static
(picture) and dynamic (motion) information using optical flow and
MPEG motion vectors. We show that both methods provide equivalent
accuracies, but that MPEG motion vectors allow a more efficient
implementation. The best proposed method yields a classification
accuracy of 97.9% - an error reduction of 64.4% when compared to
the state of the art - on a dataset of 800 challenging test cases.
Finally, we present and discuss results on a larger, and more
challenging, dataset.",
conference-location = "Niter{\'o}i, RJ, Brazil",
conference-year = "17-20 Oct. 2017",
language = "en",
ibi = "8JMKD3MGPAW/3PJFC5L",
url = "http://urlib.net/ibi/8JMKD3MGPAW/3PJFC5L",
targetfile = "wtd-sibgrapi.pdf",
urlaccessdate = "2024, May 02"
}