Searching Rich Media at TechCrunch50
Video search was a key topic at the TechCrunch50 show in San Francisco this week. Lots of newly launched companies have presented (kind of like DEMO minus the payola). In a “rich media” grouping of companies, the proverbial challenge of making images and video searchable kept coming up.
Presenting companies were
• VideoSurf: video search
• GazoPa: image search
• Fotonauts: image search
• Bojam: online music collaboration (like garage band in the cloud)
On the panel of judges was Bradley Horowitz who was a key part of Yahoo’s acquisition of Flickr back in 2005. Now he in charge of Google Gadgets and recently oversaw the integration of facial recognition software in Picassa.
“For a number of years I worked at MIT on image recognition technology and the first time I saw Flickr, I wanted to rip up my diploma,” he said from the stage. “It turns out humans are very good at this. You don’t need a hyper-technical solution. It can sometimes be the activity around the video, not the content in the video that defines relevance.
This point underscores many of the companies walking different lines between algorithmic solutions and user interaction/tagging in order to solve this longstanding indexing issue.
For the sake brevity, I’ll just talk about the best one I saw — San Mateo, Calif based video search engine VideoSurf. CEO Lior Delgo agrees that users should be brought into the mix but asserts that relying on this alone is flawed.
The company still got Horowitz’ vote for its underlying technology, which includes a great deal of intellectual property around visual recognition. This is paired with some social features that together make it a viable choice for video search over and above existing engines like Truveo and Blinkx.
The technology essentially goes beyond the voice recognition, meta data, and surrounding content that many existing sites rely on, and instead tags each frame with more granular information. This includes character names in popular shows and other things that are more contextually relevant and likely as search terms. Relevancy is weighed on frequency of these keywords, click throughs and a few other factors.
This could also be the right time for this technology, given the expanding corpus of video being brought online. This isn’t just the long tail YouTube ranks but head content that will get more search queries such as network shows on sites like Hulu. Lots more to the nuts and bolts, and it’s clear video search will continue to be a tough nut to crack. But this will be one to watch.
I hope to dive deeper in a column later this month on this and the many other search related companies at the show — including a social network for bird watchers (seriously).