There is a tendency to quote large numbers to impress these days. If it is dollars then it’s trillions, if it’s data then it is Exabytes or even Zettabytes and if it’s CPU cycles it’s usually MIPs. This week, video recognition specialist idenTV said it passed 50 trillion image comparisons an hour and talks once more of sorting through Zettabytes of data.
What idenTV does comes under Automated Content Recognition (ACR) but it augments this in a number of ways, claiming it uses Artificial Intelligence. It uses video fingerprinting and compares it with a database of cloud based image and video searches, but uses proprietary algorithms to do this exceptionally fast. It takes just a few seconds to check if a piece of video content it finds belongs to a copyright owner, and the same few seconds to identify a person’s face for national security purposes, though this announcement says little about that.
idenTV says that its technology has the power to revolutionize how a range of industries use video to do business, including facial recognition or tracking products to monitor ad spends. Key magic terms that it used to describe it are machine learning, computer vision and advanced content recognition technologies – no more detail than that.
The product only came out in November last year after four years in stealth, run by CEO and founder Mohammad Shihadah, after long spells at Net-to-Voice and integrator SAIC.
At the launch, he said, “iDenTV is to video what Google is to text. It has become practically impossible to index, search or analyze massive video in a consistently intelligent, comprehensive, or scalable way, until now. This problem is shared across a broad range of industries, whether you’re in the national security space or whether you sit on top of large video repositories and streams such as Snap, Facebook or Google.”
Our rudimentary understanding of image recognition suggests that if you can get away with fewer pixels which need to be compared, by applying machine learning, then you can process comparisons faster and this seems like its core competence. While it makes sense to use this technology for finding rogue video on the internet, the company has targeted advertising applications first and foremost so far.
Its Intelligent Video-Fingerprinting Platform (IVP) claims to be able to extract visual data from thousands of streams simultaneously and is already deployed in support of the military and intelligence communities and is now being targeted at what Shihadah says are expensive legacy ACR systems.
iDenTV is really competing with Google on indexing of video files and Google would be a likely early acquirer of this type of technology. Analysts have put the ACR market at over $3.5 billion by 2021, which, if anything, seems ludicrously low to us.
Last month iDenTV said it had achieved a new milestone when it released into beta test its Neural Upscaling, whereby before it tries to recognize an image, it upscales it, rather like modern 4K TVs do to HD video, so that it is in a better position to make a comparison and said it is especially effective for object and facial recognition. This automates the process of enhancing quality of a degraded image and amplifies the original with 400% more detail enabling faster detection and matching of objects.