VentureBeat

Visual search technology is still in its infancy, but a number of new start-ups are pushing the possibilities forward — the latest being Xcavator.

Riya’s Like launched recently (see our coverage), giving users a way of highlighting a shoe, or part of a shoe, and then look for similar shoes within its database of retail items — and so is helping with comparison shopping. You’ll notice, though, that Like operates in a controlled environment, allowing search within limited categories (bags, watches, shoes). Similarly, Swedish company, Polar Rose, just raised $5.1 million to help it launch its facial recognition, apparently due to be unveiled within the next couple of weeks.

Enter San Francisco’s Xcavator, which attacks visual search differently. It’s not open for general use yet, but VentureBeat got an early look. It will launch next year. Compared to other players, Xcavator lets you drill down further into a picture to find similar features. It claims it is getting closer to what the human eye/brain does. For example, if you’re looking for pictures with attributes similar to the Asian woman’s face (see above), Xcavator lets you circle it, and ignore her white shirt, arms and white background. See the YouTube demo below a demo of the technology. It finds pictures with other faces of Asian women (homing in on skin tone, hair color, and dark eyes and hair (it works for blondes too).

It also lets you click on unique points on a picture, like a butterfly’s wings, or petals of a daisy to find those specific features (see image at bottom, and related tour, which explains more).

The big question, though, is who will want to use this technology. We all have a favorite shirt or blouse that is getting old, and so we might upload a picture of it in an effort to find something that looks exactly the same. That seems like a lot of effort and its probably rare. It’s early days for this company, and its unclear how it can become a stand-alone company earning real revenue. Unlike Like, Xcavator is not crawling the entire Web to index pictures with the goal of becoming a destination site. Rather, it intends to liscense its technology to other companies, even to competitors, such as retail and fashion sites, but also to Riya, Yahoo, Google, or to companies with image software like Apple or Adobe. It’s also working with military agencies for intelligence work. It may work with professional photography sites like Getty Images to scan ad placements across the Web to check for copyright infringement.

The company has been self-funded for three years, and is based on technology built by Russian Lenny Kontsevich, PhD, who has researched his area for a decade. The total team is 10 people, split between SF and Moscow. It is looking to raise a round of venture capital.

The strength of this company is its ability to search vectors within a picture. If you select the entire picture of the Asian girl, for example, it will look for pictures with the white background, with dark hair, arms in similar positions. But it also allow for rotation of such vectors — finding picures with arms in slightly different angles, for example.

Xcavator’s parent company, Cognisign, has various projects up its sleeve. One site will focus on a younger demographic (girls aged 14 to 25) than Riya, because of the viral marketing possible with that age group. [Update: Xcavator has a placeholder name for site, but wants us to keep it confidential for now, for competitive reasons; we've removed screenshot too]. It will let users search, rank, and comment on images, and do it from their mobile phone. The seed/A round would be used to launch this project, says chief executive Bryan Calkins. We aren’t quite sure what would attract girls to this particular offering, but who are we to know.

Updated

So which Silicon Valley venture firm will dare touch Ucloo?

Ucloo is a year-old Chinese people search engine, and there are many things not to like about this site. First, check out the logo. What does it remind you of? At least they switched the colors of the first three letters (it is red/yellow/blue, instead of blue/red/yellow). See the page comparisons below, with Google, and with Baidu.

Chief executive Randy Ding tells VentureBeat that Ucloo has already indexed 90 million personal profiles for its database, making it the largest people search engine so far. Spock, the ambitious people search engine we profiled here plans to launch with 100 million. But Ucloo is adding between 10,000 and 100,000 profiles a day, Ding tells us, so it will zoom past Spock within a few weeks. It is focused on China now, but is expanding to cover the Chinese community abroad, including in Taiwan, US and Canada. It wants to launch an English version after it raises venture capital — and is now beginning to reach out to Sand Hill Road.

It’s eerily comprehensive. It wants to list everything it can find about you, including your name, birthday, contact information, education, work background, pictures, online activity and reputation — even court records. Check out this profile below, of “LoveGirl,” which shows her Gmail and MSN accounts, among other things. So what happens when it crawls your information, finds out what sites you’ve watched online, who you’ve corresponded with, etc, and you can’t appeal to U.S courts to take it down? Yikes.

It wants to 1) let people search for information on other people, 2) let companies check employee background and references, 3) let companies check other companies’ reputation and employees, and 4) lets a user check a company’s background. There’s the additional unanswered question about what China’s government would use it for.

Update: VentureBeat contacted three plugged-in sources in the Chinese tech community, and heard back from each of them that they hadn’t heard of Ucloo until we brought it to their attention. One of them tried it out, and found it useful. She did some research and reports that the current version of Ucloo launched in Sept. after settling some legal issues concerning its data sources, including allegations of theft. In addition to the material we mentioned above, Ucloo even collects your IMs and posts on online forums (so-called bulletin boards in China). Our correspondent found that most media coverage in China about Ucloo is negative, so the commenters below are off-base when they suggest VentureBeat is anti-Chinese. Ucloo’s Ding attended the recent Web 2.0 conference in SF.

Ding, meanwhile, has since responded to some of our criticisms on privacy. We will not edit his remarks. He is Chinese, and so applaud his effort to correspond with us in English.

Ding:

We have the function of “It’s me”, after we verify (by the email address which we send from our database or ask them upload the ID) the person who is listing on Ucloo, the user can edit his/her profile and add/remove some of information.

VentureBeat:

So I can remove anything I want, so that Ucloo won’t have anything about me?

Ding:

1. User can remove SOME of information which is included home address, home phone number or some of other directly contact information and can ask system to transfer all connection request through ucloo.
2. Personal education and work experience can be added new information by the owner, but can’t be removed.
3. Web reference is an hyperlink depend on source website, Ucloo cached to check.
4. Business membership can see all information whatever it has been edited or not.

Also, next version which will be launched the end of this year, we are going to change our logo NOT like Google.

Like has finally launched its visual search engine company, and it’s going to appeal to the shopping set, especially women.

Users select products from images, and then Like’s search engine will find comparable items to buy. Like is owned by Riya, which we have covered before, including here.

We find it compelling. There are other players in the market, but aren’t nearly as sophisticated as Like. For example, there is Pixsta, of London, which has built a shoe search at www.chezimelda.com with about $200,000. But it is elementary when compared to Like’s technology. Like is a classic Silicon Valley play, heavy on engineers, and stoked with about $19 million in venture capital from Bay Partners, Leapfrog and Bluerun. We’re still waiting for Google to release its competing visual search, which some say is better (see our story here).

Take for example, the image here of Tyra Banks. You can select her boot by drawing a box over it with your mouse, and then check a box to have Like search for similar boots. You will get a page like the one below (see bottom). Further, you can ask Like to search for things with a similar pattern, color, and shape — using a slider for each of the variables. So if you’re looking for a handbag with buttons on it, you can emphasize pattern, and maybe color, but deemphasize shape — and have Like generate results. Here’s a tour of how it works. And here are more details of the launch.

It is not perfect. It can get confused by focusing on a color or image in the background of a picture, for example, instead of the object you’re focused on. But it’s better than what’s out there. If you search for “red strappy shoes,” Like will produce a page of results with — shocker — red strappy shoes. While this may be obvious, if you try this at Shopping.com, you’ll get only four shoes that are red and strappy. That’s because “pixel values” of an image rank highly in Like, whereas Shopping and most other search engines don’t factor this in.

For now, Like is focused on jewelry, handbags, shoes and watches. Soon, it will add clothing, and other things like fabrics and garden flowers. Everyone has their favorite shirt pattern that’s wearing out at the seams, and can relate to not being able to find another one just like it. Like will let you find a similar one easier. You just upload a photo, select it, and search.

It will also be releasing an upload toolbar, so that you can select anything you see online, hit the toolbar and have Like search for it. It will also launch a mobile SMS product.

Jewelry, shoes and clothing command $15 billion in online sales, and merchants spend about 10 percent of their revenue on customer acquisition — meaning there is about $3 billion floating around to pay Web sites and other sources for lead generation. If Like can just get one percent of that, or $30 million in revenue, chief executive Munjal Shah says he’ll have a nice company on his hands. (He’d better hurry, because Google’s product will probably take some wind out of his sails).