I am a big fan of tagging as a way to organize content, save my bookmarks on del.icio.us, and optimize content for consumption on social media sites. Yet I have also written before about the inherent difficulties of tagging (suggesting that a TagWiki might be the solution). The biggest issue is that multiple people can use many tags to indicate the same thing. This creates a difficulty in aggregating similar content, because you cannot rely on a single tag or keyword to show all relevant content. Other problems stem from differences in syntax. For example, right now, "publicrelations" is a different tag to "public_relations" or "public+relations." Chances are, most people would mean the same thing by using any one of these tags, but there is currently no way to aggregate them. Spelling and plurals also cause issues, as someone might use the tag "socialmediaoptimization" or "socialmediaoptimisation" depending on where they are from. When it comes to images, the issue is more basic … most images online are lacking any tagging or semantic data to describe them. They are therefore effectively invisible to search.
With all of these minor issues with tagging, I came across Michael Arrington’s post yesterday on TechCrunch about Ookles, a soon to be launched online photo site. One of the intriguing features of this site is the ability to set up autotagging with facial recognition. Taking the work of Riya in visual search one step further, Ookles is extending this idea into autotagging. After beta testing the site, Arringon describes a three step process to train Ookles to recognize a particular face – and then every photo uploaded from then onwards will recognize the same face. Today on Techcrunch, Arrington posts about Polar Rose, another such offering taking the different tact of creating a plugin to let users generate semantic data around images. Looking at both of these services, it is clear that the uses go far beyond simply organizing your holiday photos.
In this age of consumer co-creation, brand logos are being photoshopped left and right. The problem is, it’s not easy to know when and where this is happening. Training a technology like Ookles to recognize versions of a brand’s logo and autotag real uses, as well as derivative uses with specific tags could help brands to monitor logo usage and spot potential brand issues. In my view, the ideal use of autotagging for images would be similar to how the DDB music database is integrated into music sharing sites such as iTunes. Currently, when you import music, you can automatically get a list of song tracks, album data and even cover artwork downloaded automatically. These new technologies being introduced by both Polar Rose and Ookles offer the same potential – to allow people to upload images of people, places or things, and get automatic suggestions for tags that describe those images. Aside from being a great time saver, its one of the more powerful ideas I have seen to help organize the growing billions of photos that are being posted online every year.
