Riya heading into Web search
Currently, Riya allows users to upload photos to its site and users can train the system to recognize faces. It also recognizes text in images, and can slice photos by time taken (not uploaded), location (with a Google Maps mashup) and found text. It also provides tagging, photo sharing and public and private albums. I have over a thousand photos on Riya, and the recognition technology is impressive.
So far, Riya users have uploaded over 7.23 million photos (it launched on March 21), and users have 'trained' about 60 percent of the faces uploaded, Shah told me.
He expects to hit 10 million photos in the next six to eight weeks. "As a photo site, we have exceeded the initial growth rate of Flickr," Shah said. Of course, there is more broadband connectivity than when Flickr came on the scene, but there are also many more photos uploading sites. Flickr was acquired by Yahoo in March 2005, and has over 100,000 million photos.But Shah isn't planning to focus on photo organization, sharing and hosting. His goal is to build in capabilities to search across billions of publicly available images. "We are still focused on people uploading [images]," Shah said. "And, we have been a search engine the whole time. We are better at finding people, and our [searching] is growing faster than uploading. You will see us do more crawling and searching of public images in addition to what is uploaded."
Based on some recent insights, Shah said that Riya is building a new type of photo search engine, one that finds similarities among images. "Riya started with facial recognition, but we will deliver public image search and do similarity, such as finding images similar to your location or girl friend," Shah said. "Click on any photo of any person and get similar results. You could have a picture of a car and find things that look like cars."
Similarity searches could be applied to dating sites or for finding similar images, such as a particular landmark, without explicit tagging. "The only technique today is linguistic query, and many things are hard to describe linguistically, such as a person. That's the challenge. It's easier to say 'more like this,' 'less like this,' and that's the direction we are heading," Shah said.
He doesn't expect to have his crawler and public search available any time soon, however. "There is core science to crack to actually run similarity across billions of photos and have any semblance of accuracy and speed," Shah said. But he doesn't expect his company's technology (he has filed some patents) to reach a level of accuracy that would be appealing to national security agencies. "The level of accuracy and scoping will only be useful to consumers," he said.
It's an ambitious undertaking, one that seems more on the scale of what Google, Microsoft or Yahoo might do. Shah has collected $19 million in funding to keep the fires burning. I am sure that the big engines are keeping a watch as Munjal Shah adds fuel to the fire. In the meantime, Riya should find a way (a bridge with a business model) so that users of photo sites, such as Flickr, Webshots and Yahoo Photos, can very easily take advantage of its unique facial and text recognition services.