Product

Technology

Videntifier™ Forensic builds on a huge database of visual fingerprints, so it can identify most commonly available video material. A visual fingerprint is extracted around specific interest points in the individual frames of a video. It encodes the visual signal around these points into a sequence of numbers, that is then stored in Videntifier Technologies' patent protected database. As each video contains many thousand frames and each frame up to a few hundred points, Videntifier™ Forensic extracts about 150,000 visual fingerprints from each hour of video.

Currently, Videntifier™ Forensic contains over 6 billion video fingerprints in its patented database, representing about 70,000 hours of video content. The database has been growing since the official start of the service in December 2009. Our first customer, the Icelandic police, did extensive testing and practical evaluations on the system before full deployment. The chart below shows the identification rate in tests of three actual cases. As the database grew, the identification rate also went up. Therefore Videntifier Technologies continues collecting fingerprints, enabling Videntifier™ Forensic to remember more than 100,000 hours of video by the end of 2011.

Robustness

The fingerprints used by Videntifier™ Forensic have been designed to be robust against the most common distortions of visual content. They only encode the coarse shapes of the structures inside an image, such they are very robust against compression, rescaling, brightness and contrast, aspect-ratio and color changes. Many points per image are taken into account, so Videntifier™ Forensic is also very robust against cropping and clipping of the video content. Furthermore, rotated and mirrored versions are easily identified. As our central database does not rely on the exact same fingerprint sequences, it is very flexible and can also identify fingerprints that are only similar – the system is extremely robust even in the context of many billion fingerprints.

Security

Videntifier Technologies has taken several measures to make the Videntifier™ Forensic service very secure. First of all no image data is sent over the internet or reaches the Videntifier™ Forensic server. Only fingerprints are sent as a query or insertion request to the NV-Tree database server, and only over a secure, isolated connection.

Central database

The central database holds the fingerprints of all video material. During the identification process the fingerprints of each video frame are queried against the database. In case several frames are recognized the videos name, classification and description is sent back to the client. If the sequence was not identified (e.g. because Videntifier™ Forensic had never seen it before), the client software moves the video file to a separate folder, where it can be manually evaluated later. In order to keep the service up-to-date, Videntifier Technologies regularly adds new video fingerprints to the database. Furthermore we appreciate the work of all investigators who actively add unidentified video fingerprints into the database.

Videntifer's Forensic main strengths

• Remembers thousands hours of content
• Recognizes video 80 times faster than real-time
• Tolerates strong modifications (e.g. compression, mirroring, cropping, aspect ratio changes etc.)
• Identifies clips down to a few seconds long