Searching for the image: Picitup’s founders, Dan Atsmon (left), Alon Atsmon (right).There are billions of bits of text on the Internet – and without text search engines, chief among them Google, we’d be lost without direction in a sea of un-navigable information. And what of the billions of images now on the Web – thanks to sites like Facebook, Flickr, and the many other photo sharing sites? For that, we have Israeli startup Picitup, which aims to do for photo search what Google did for text search.
It sounds like a tall order, but Alon Atsmon, CEO of Tel Aviv based Picitup, feels he’s got things covered. “When you search for a picture or photo using Google’s image searching function, you’re essentially searching for text tags attached to an image. That’s not what Picitup does,” Atsmon tells ISRAEL21c. “We search for an image’s physical characteristics, enabling you to track down the closest match to what you’re really searching for.”
A demonstration, comparing a search for the term Picasso on Google Images and Picitup, drives the point home. While Picasso can refer to the legendary painter, it’s also the name of a car by France’s Citroen. A Google Image search for the term Picasso yields page after page of Web sites with on-line versions of Picasso paintings, with nary a car in sight. On Picitup, the Citroen is right there among the top choices. When you click on that image in search mode, more Citroen Picassos – as well as similar looking cars from other manufacturers – show up.
In order to enable a search for similar attributes in photos (the system works for videos, as well) based on image recognition Picitup requires a great deal of back-office programming – with each component of an image being described in programming language.
Getting the images you want
How do you explain the color “blue” to a search engine that has no vision? With a lot of bits and bytes. Even the simplest image has hundreds or even thousands of attributes, all of which need to be describable for a search engine to be effective. “An image has a million times more information than a text string,” Atsmon says, describing how much work goes into building an engine like Picitup.
Another issue is ranking search results, so that the ones you really need are at the top of the list – on the first page or two of search results.
“People expect a search engine to give them what they want on the first couple of pages – they’re not going to click beyond that,” Atsmon says, adding that any search engine specializing in photos has to be able to compare the searched results and present the user with the most likely choice.
“As a result, an image search that would usually require 10 page views will be reduced five times to two page views. No other image search engine can provide image ranking,” – unlike Picitup, Atsmon says, which is able to search and compare those attributes in a moment – and on a moment’s notice.
“There’s a reason why image search has been limited to keywords until now. We have a huge server farm with thousands of computers, giving us several Teraflops of computing power, equivalent to thousands of Intel Pentium processors,” Atsmon says, adding that the company has developed some unique innovations to keep the cost of processing cheap enough to make Picitup a worthwhile venture.
And Picitup has already “pre-analyzed” tens of millions of images, in order to keep things speedy.
A million page views a month
The proof of a search engine’s success is in the numbers, and Picitup is getting about a million page views a month, Atsmon says. “In fact, our tests have shown that we’re far ahead of the ball on image search than anyone else – including Google,” he says, citing the results of scientifically conducted lab tests that showed user preference for Picitup’s method of delivering photo results over that of Google’s – or any other search engine’s.
And while Picitup is great as a standalone search engine, image search and comparison technology can go far beyond just allowing individuals to seek photos. Imagine this scenario: A discount online department store lets customers input images of the latest Parisian “haute couture” – and turns up the closest knockoff available, for a fraction of the price.
Perhaps what’s most miraculous about Picitup has been its ability to build such a sophisticated search engine with a small staff. “We’re lean, but we have big plans,” says Atsmon, describing several directions he wants to see the company go towards.
The company is putting the final touches on the shopping engine, and there is already an application called Celebrity Match, where you can search for stars that look like your uploaded photo. And Picitup works with videos, as well as photos. While Google may have a lock on the text search market, a picture is worth a thousand words – and those images belong to Picitup.