Kuznech ready to help Internet users enjoy Web with ‘grasshopper’s eyes’
17 May '13
Oleg Kouzbit, Online News Managing Editor
Russian IT developers have created what might be identified as an image-focused search alternative to Google or Yandex. Inspired by Mather Nature’s sophistication in giving the grasshopper its uniquely rich eyesight, St. Petersburg-based Kuznech is said to have applied similar finesse to developing a technology that enables users to index and compare a host of images online. It’s not just for fun; the Kuznech technology has already brought about solutions that help shoppers find like goods in stores. With a few million dollars invested in the two-year effort, and with about $1.5m in external funding currently under its belt, the start-up focuses primarily on the vast US markets.
The Kuznech.com project is a brainchild of three very different people who now run the company. These are Mikhail Pogrebnyak, a software developer and the CEO of Kuznech; Alexander Valencia-Campo, a mathematician and the Chief Scientist in the project; and Pavel Cherkashin, a serial angel investor with a strong IT background working as Kuznech’s VP Sales.
The start-up’s core technical team of 12 people is based in St. Petersburg, on Kuznech’s home turf, and another three staff supervise business matters in Moscow and Boston.
Kuznech.com has taken two years of work and “a few million dollars” that the partners have invested out of their own pockets. External investors gave the project seed capital of a reported $700,000; that was followed last June by a $750,000 grant from Russia’s Skolkovo Foundation and by a $20,000 prize at the IDCEE 2012 competition in Kiev, in Ukraine, last October.
The St. Pete company estimates the global pent-up demand for image-based search technology to be around $200m, and expects that in a near future, the image-focused advertising market alone may top $6bn.
In the eye of a grasshopper
At the heart of the Kuznech.com project is a technology that draws on what might be referred to as knowledge of ‘grasshopper optics.’
Just as the eye of a grasshopper contains a multitude of hexahedral cellules, each generating its own version of an external object and enabling the insect to see the world as a kaleidoscope of thousands of different images, so too Kuznech’s automated visual search system can provide a user with a new ‘set of eyes,’ thus giving him a hand in sifting through tons of images and videos on the Internet the same way we all search through text.
Coloring web search, facilitating shopping
One could liken the innovation to the well-known search engines, such as Google or Yandex, or some lesser systems that exist. Unlike the predecessors, however, which are designed to only operate with text associated with an image, leaving therefore virtually unsearchable between 96 and 99 percent of an estimated 3.5 trillion images uploaded onto the web without any text description, Kuznech.com is said to work with the image itself.
The developers claim that their technology enables users to index and compare “billions of images” by as many as 150 different parameters, including composition. “The core technology can find similar images in seconds without doing any object recognition – just based on statistical comparison of images to existing reference database,” Kuznech says.
To enable the comparison, the system has been ‘taught’ to extract image signatures by composition, colors and textures, and then offer its ‘guesswork’ on what’s on the image after the signature was compared to Kuznech’s database.
“This doesn’t provide extreme accuracy, but allows very high speed of processing and low cost,” the start-up claims.
The company also says its solution can help consumers shop more resourcefully. It doesn’t require that in doing search for products, a user lose images to having to convert his or her request into a text query. Kuznech claims it offers a “new visual way” for shopping, in which people “can use existing product image as a query to find similar products on sale.”
An array of new solutions can be developed based on their technology, Kuznech.com’s ‘founding fathers’ say. The service already can help users find like goods and place tags on images instead of text descriptions; and more is expected to come.
Kuznech has been test-running its service in collaboration with Sapato.ru, an online shoe store, and with RIA Novosti, a leading Russian news agency. The former is reported to be tapping into the St. Pete technology to look for similar items in the store, and the latter seems to have found it useful in spotting instances of unauthorized copying of its proprietary images.
Kuznech has reportedly been in talks with Facebook since last fall over using the Russian system to help put in order the social network’s enormous archive of user photos.
According to Mr. Cherkashin, “our key potential [and a few existing] customers are in the U.S.”