Improving Google image search (theory)

by subacati

The 'holy grail' of a mad scientist is to be able to build an electronic 'brain' that has real intelligence rather than artificial intelligence! :left:.

I stumbled across this interesting article about an artificial neural network, simulated on a Commodore 64! :hat:. …

The Commodore 64 is my favourite computer. But even so, I didn't expect this to be possible on such a simple 8bit computer! :yes:.

So what does this have to do with Google Image search? Well, if simple pattern recognition is possible on the single tasking, 8bit Commodore 64, then full blown image recognition must be a practical reality on a modern multitasking, multimedia system. :idea:.

So this is my idea, Take this basic algorithm, parallelize it, and load it onto a Beowulf Cluster. Now, create a database of thumbnails, google already has one, and allow visitors to search those thumbnails with a copy of an image they already have! :happy:.
The results will be pretty close to the image submitted. This will make it possible, for example, to find out what model aircraft that is that you took a photo of when you were twelve! :p. And what about an easy way to identify that wierd looking insect you saw yesterday? You took a photo with your cellphone's camera, but you don't have an entomologist at hand? Google Image to the rescue! :yes:.

So I hope Google considers adding this feature soon! :hat:.