The dramatic growth in network bandwidth and storage capacities, as well as the increase of the sheer CPU power in desktop, laptop and hand-held devices have lead to the offline and online generalization of multimedia content. People share more and more image and video content as a matter of course in their everyday life. Nearly two billion photos are posted daily on social networks1 and one hundred hours of video content are uploaded on YouTube2 every minute.
This outburst of content is strongly hampering moderation strategies for social network, sharing platform and cloud storage providers, who face an increasingly difficult task of timely and properly sifting through submitted content.
By the same token, traditional filtering solutions from specialized software vendors are increasingly out of their depth, with a growing part of the content left unchecked, on one hand, and wholesale site and domain blocking becoming less and less suitable to block shared inappropriate content, on the other hand.
(1 KPCB – Internet Trends 2014. 2 YouTube)
Unsurprisingly, porn industry turned this situation into an unprecedented opportunity. As a result, underage are increasingly exposed to online inappropriate content available only one click away, and impact on productivity in businesses is increasing. Online life as a whole is calling for a means to analyze image and video in order to tell inappropriate content from the rest.
For the time being, the search for nudity and
pornography in video content is mostly manual. While this method can still
lead to blacklisting of sources in some cases, it is clearly fighting an
uphill battle against the recent multiplication of sharing and broadcasting
platforms. With each and every blog becoming a potential source for video
content, manual methods are quickly becoming irrelevant.
Computer-driven analysis of online text around video content can still do an excellent job in some cases, but inevitably misses the mark on text that is either unrelated to the video content or is lacking altogether.
For these reasons, Profil Technology took the challenge of developing a groundbreaking technology to analyze image and video content – the VISIA Technology.
VISIA Technology is a content recognition technology in a league of its own with respect to the existing ones, on par with nowadays forms of communication and consumption of information.
It performs detection, analysis and classification of pornographic image
and video content, either stored on physical storage or transferred as an
online stream. Profil Technology's R&D team developed their own computer
vision technology, based on visual pattern recognition. It identifies and
extracts the most significant, yet manageable features in images and videos.
The data thus extracted is then fed into Profil Technology's own artificial
intelligence layer for sharp analysis and detection of inappropriate content.
The technology remains effective no matter the scaling factor, position and orientation of the analyzed image or video. A far cry from known simplistic approaches like the search for skin tones, VISIA Technology builds on machine learning, data fusion and visual pattern isolation and recognition.
VISIA Technology is available for integration as a software development kit (SDK), and as such can be deployed according to nowadays' major access modes for image and video content:
As inappropriate content detection based on text semantic analysis is becoming irrelevant, VISIA Technology targets the data stream as a whole instead, categorizing content in real time. Brief and/or late inappropriate sequences won't bluff their way through a given stream – VISIA will single them out just the same.
Detection is performed while the movie is watched online, without slowdown or lagging.
The only time offset in user experience takes place at the very beginning of
the loading process (5 to 10 seconds).
No particular, top-of-the-shelf hardware capabilities are required.
VISIA Technology allows analysis of any type of digital storage in search of explicit pictures and videos. Storage structures are scanned in both breadth and depth, beyond file type and into actual file structure. As a result, an actual picture or video compressed in an archive or concealed as a .xlsx, .docx or .pdf file will nevertheless be analyzed.
VISIA algorithms went through several steps of optimization changes and are now able to work on-the-fly on mainstream consumer hardware.
VISIA algorithms secure over 95% detection rate all the while keeping false positive at the lowest possible level.
The asymmetric setup allows for a custom detection level, to strike the best balance between detection and false positive rate in a given environment. For example, a household setup will aim at total child protection through conservative porn filtering, at the expense of a few over-blocking cases. At the other end of the spectrum, a business environment will prefer minimum false positive over detection, thus securing unhampered access to legitimate content.
The core of VISIA Technology is available for all major operating systems in use: Linux, MacOS X and Windows.
Analysis applies to 100+ file formats, including JPG, PNG, GIF, BMP, AVI, MP4, FLV, to name but a few.
The VISIA SDK is meant for quick development of one's own applications as well as for seamless integration of detection functions in third party products and services.
Integration is designed to allow for custom behavior in preventing display of explicit sequences: block, skip, blur, etc.
The SDK is available as a library of native code targeting the chosen operating system. As such, it merges smoothly into applications written in C, C++ and C#, and so forth.
The VISIA SDK is complete with API reference documentation and includes several application samples as source code, making for a quick, hands-on tutorial.
VISIA Technology Application Examples