AI-based Media Analysis has always been a big part of the triage functionality of LEAP, since the early days of its development. And while it has improved over time with the growth of the platform, the ever-growing demands have led T3K to a point, where it was necessary to redesign the whole Media Analysis engine and elevate it to a new level.

The first major change came with the introduction of key-frame detection of contents in videos. It provides a streamlined but accurate method of detecting inappropriate or otherwise relevant content by splitting videos up into frames, wherever the scene changes enough to be considered a new scene. The frames acquired by this technique are then screened just like regular images. This leads to a quick, robust detection of contents in videos.

Another big step to make the LEAP Media Analysis more versatile and powerful is the ability to upload folders of images and videos from external sources directly into LEAP.

Where it was previously only possible to analyze images and videos that were in the extractions from mobile devices, this will give law enforcement new possibilities to get a more comprehensive look at seized data like external drives – or even internal storage locations of old case data.

Shortly after, the general revamp of the Media Analysis started and with the launch of LEAP 2.5, we are happy to introduce the new Object Recognition Backend into LEAP.

The most important changes resulting from this improvement are:

  • 5 times faster – Simultaneous object recognition in minutes, automatically locating CSA material, weapons, ID documents, Vehicle license plates and more.
  • Multi-step CSA detection with enhanced machine learning models – Pornography detection followed by CSA determination reduces False Positive rate
  • Improved Facial Recognition – Simply uploading one or more reference faces to LEAP allows automated facial matching from media inside mobile device extractions or other media source

“Figuring out how to squeeze your CPUs and GPUs like lemons was not the hardest part. Managing to squeeze them all at the same time without “clogging the strainer” however was.” says David Weichselbaum, CSO and Head of Machine Learning Dept. at T3K, jokingly.

We look forward to seeing the new Media Analysis in action and the improved performance it brings to our customers.

We are available for any questions or trial requests at office@t3k.ai.
For more information on LEAP, you can also see www.t3k.ai/leap