Way, way back in the early days of my career (‘01 or so), I started an incredibly small little startup focused on building custom software for researchers. For two years I wrote software that would track the movement of lab rats in cages for Parkinson’s researchers, which led into an opportunity to work on an insanely awesome project called the Direct Brain Interface (DBI).
The premise of the DBI was to detect patterns in the electrical activity in the brain, and use those signals to control a simple video game. There are a few catches, though. First, instead of using rats, the team was working with real live humans. And second, as opposed to the Indirect Brain Interface, the signals we were reading came from probes that were inserted directly into the brain.
Like I said: insanely awesome.
Of course we didn’t insert the probes for the sake of the experiment, we worked with patients who already had the probes inserted for other medical reasons (mostly as treatment for severe epilepsy).
Anyway, I spent the summer writing all sorts of crazy signal processing algorithms for two main purposes. The first was to look at the signals coming in from the 30 or so probes to try to discover the location of the specific “thought” we were trying to capture (brains are crazy flexible, no two are the same). Then, once we found the right location, we then analyzed that one channel to pick out the “thought” as it occurred.
I had never done much signal processing work before, but the amazing thing was that once you got all the wickedly complex math right, it would actually do a pretty good job of picking up a signal out of this incredibly noisy input from the probe.
This is leading somewhere, I promise.
There are millions of candid photos floating around the Internets, especially as more and more people opt-in to share their Facebook data with the public. And just like that tiny signal that could be pulled from the noisy brain probe, I believe there are lots of interesting things that could be pulled from these pictures using similar signal processing techniques.
The one that I’m focusing on today is a site that looks at pictures of people and tries to determine the most popular clothing and fashion choices. For example, if the algorithm found 100 pictures of people wearing the same purple-and-pink tank top from the Gap in the past 24 hours, it would show up on the site in the “Hot Now” section. Users could go to the site to see what clothes, shoes, purses, etc.. are popular today, and click through to purchase the items online from their favorite stores.
Along with just pictures, we could also throw in a little metadata for more accuracy. For example, if the photo is from a public Facebook feed, we can cut the data by age, location, and even popularity (measured by number of friends). You could also use that same data to apply weighting (i.e. a purse that is worn by a college student in New York City with over 1,000 friends would get a high weighting)
The end result would be a constantly-updated site showing the hottest fashion styles in real time. Think of it as a Digg for fashion, but one that doesn’t have the wicked-hard chicken/egg problem of getting users to contribute content to an unknown site. Along with generating money via lead-gens to online stores, the data could be used to generate real-time reports for fashion designers.
Like this idea? If so, you should follow me on Twitter at @astartupaday.
If not, you should probably follow someone else.