Filtering the Good, the Bad, and the Ugly

An IT innovation  from Brazil that reliably filters out unsolicited pornography  videos from web and social-media.

An artificial intelligence application developed by Brazilian computer scientists has begun to combat the plague of pornographic junk mail and obscene videos besieging billions of mobile phone users around the world.

Samsung, one of the world’s top mobile phone producers, in 2012 commissioned scientists to develop an effective system that would block unsolicited messages with offensive screen images. This April, scientists at the University of Campinas’s Computer Science Institute (IC-UNICAMP) announced they’d created an efficient filter that identifies over 97% of the pornographic content available.

Filtering unwanted video pornography benefits billions of  mobile phone users.

Filtering unwanted video pornography benefits billions of mobile phone users.

The research is a spinoff from a wider project on social media forensics and interpreting criminal events, which was supported by the São Paulo Research Foundation. The project – named “Déjà Vu” – investigates how social media may be aggregated and filtered to support forensic investigation of sophisticated crimes and terrorist threats. Researchers are teamed with the University of Notre Dame in the US.

To date, porn filters have been hampered one crucial defect. Just like those irritating spam filters that devour our eagerly-awaited emails containing airline boarding passes, theatre tickets or vital receipts, porn filters simply couldn’t separate “good” naked flesh from bad. Spouses in bikinis on the beach, children in paddling pools, wrestling fights and innocuous family photos, all often ended up in the “sin-bin” alongside the hard-core porn.

A partnership between the Samsung Research Institute of Brazil and IC-UNICAMP led to a system based on machine learning (or artificial intelligence). It’s capable of filtering out over 90% of the pornographic content available on any device. The new technology was co-patented by Samsung and UNICAMP.

This is a smart IT solution that can “be installed on smartphones, smart TVs and computers so that consumers with children could block access to sensitive content – either upon buying the device or at any time before use by a child,” said Anderson Rocha, a professor at IC-UNICAMP and principal investigator for the project.

Rocha and other researchers affiliated with IC-UNICAMP and Samsung recently published an article in the journal Neurocomputing that describes in detail how the new technology was developed. Rocha (currently working at Singapore’s Nanyang Technological University) claims “we have a very efficient filter that identifies over 97% of the pornographic content available.” Currently, says Rocha, state of the art commercial systems can expect to deliver no more than 94% problem free image selection.

The new technology developed in Brazil relies on what’s known as “deep learning.” While the earliest filter systems attempted to detect pornography  by identifying bare flesh and then using a static benchmark of nakedness to decide what was unacceptable, these ended up “arresting” harmless images of toddlers in paddling pools, or even medical videos.

The next development was to filter out pornographic content based on a list of words that might accompany illicit images. But the Computer Science Institute team’s real breakthrough was video. By sampling extracts at a slowed-down one frame per second from each video, and then analysing the movements of objects and people present in each scene using an algorithm containing “learned” descriptions of permitted/pornographic scenes., they reached a long-sought benchmark in reliability.

According to Rocha, this method has been tested using a dataset containing approximately 140 hours of video, including 1,000 pornographic videos and 1,000 non-pornographic videos, each lasting between 6 seconds and 33 minutes.

Filtering unwanted video pornography benefits billions of  mobile phone users.

Sampling of  video content yields a 97% accurate result for filtering out obscene images.

The researchers say their future efforts in artificial intelligence will be in other types of image analysis, including extreme violence, and, of course, child pornography. Today, the system is only 90% accurate when it comes to child pornography and 80% accurate for scenes of unacceptable violence.

You can find an article by reporter Peter Moon on this link:

If you want to pay publishers Elsevier for details of the Brazilian team’s work click here. The article is entitled “Video pornography detection through deep learning techniques and motion information,” (doi: It is by Mauricio Perez, Sandra Avila, Daniel Moreira, Daniel Moraes, Vanessa Testoni, Eduardo Valle, Siome Goldenstein and Anderson Rocha.

Researching in Brazil

Brazil’s best-known scientist, the veteran nuclear physicist  José Goldemberg,  in September 2015 took office as the president of  the São…

Read more