On the heels of Apple this morning inking a $600 million deal to acquire IP, talent and licenses from Dialog Semiconductor in Europe, it has also confirmed another acquisition of a smaller startup in the region.
Apple has purchased Spektral, a computer vision company based out of Denmark that has worked on segmentation technology, a more efficient way to “cut out” figures from their backgrounds in digital images and videos, reportedly for over $30 million.
This type of technology can be used, for example, to make quicker and more accurate/realistic cut-out images in augmented reality environments, but also for more standard applications like school photos. That was actually the first market the startup targeted, in 2015, although it appeared to shift strategy after that to build up IP and make deeper inroads into video. You can see a demo of how its technology works at the bottom of this post.
Rumors of the deal started to surface yesterday, first in Danish financial newspaper Børsen, without confirmation from Apple. We reached out, and Apple has today finally confirmed the deal with its standard statement: “Apple buys smaller technology companies from time to time, and we generally do not discuss our purpose or plans.”
From what we understand, the acquisition happened a while back — which lines up with a LinkedIn profile for Toke Jansen, who had been a co-founder of Spektral but now notes that as of December 2017 he has been a manager of computational imaging at Apple.
Others associated with the company — including the other co-founder, Henrik Paltoft — have not updated their profiles, so it’s unclear how many others have joined. Børsen reports that the deal includes the company’s engineers and was in the region of 200 million Danish kroner, which is equivalent to around $31 million.
Spektral href=”https://techcrunch.com/2015/08/10/cloudcutout-applies-machine-learning-to-image-editing/”>started life as CloudCutout, built on algorithms from Jansen’s PhD. The startup initially pitched its product as a cheaper and more efficient “green screen” technology, to remove primary images from their plain (typically green) or standard-pattern backgrounds, with an early iteration of the product built by training the system on over 100,000 professional cutouts.
Spectral’s first application may have been the fairly retro world of school pictures, but what’s most notable here is what Spektral might contribute to Apple’s imaging business. That could be for applications that Apple has yet to launch, but also to improve the quality of those that are already in the market, from legacy products like PhotoBooth through to ARKit, the company’s platform for mobile development.
Segmentation could help add live filters to human figures in a photo but can also be effective in occluding AR environments behind figures to make digital AR content appear to interact with the position of humans.
Spektral’s segmentation technology is also able to run on mobile phones, making it potentially a quicker and more efficient way of processing AR images directly on devices. And one of its main selling points is how it copes with very fine details, such as hair.
“To provide high quality cutouts, the core of our engine exploits recent advances in spectral graph theory and neural networks. The computation of pixel transparencies (the alpha channel) for a single image involves solving multiple large-scale equation systems, as well as carrying out multiple feed-forward passes in our neural networks,” we reported the founders saying when the startup raised a seed round. (It raised more funding in 2017, $2.8 million from Litecap and Amp Ventures, to diver deeper into video.)
“We pose the problem of determining an alpha channel of an image as a machine learning task. Compared to usual chroma keying, this allows us to consider a much broader range of backgrounds since the model will learn, i.e., texture representations from existing training data.”
Computer vision has been a focus (sorry!) for Apple for a while now. The iPhone and Mac giant has made more than 40 acquisitions in Europe in the last 10 years — I guess we may still have some hunting to do — and a number of its acquisitions both in Europe and elsewhere have been in the area of computer vision. They have included Akonia Holographics, InVisage Technologies, Regaind, Vrvana, SensoMotoric Instruments, Indoor.io, Flyby Media, Emotient, Faceshift, Metaio, Polar Rose and more.
Additional reporting Natasha Lomas.
Powered by WPeMatico