Show profile
Link to this post
Subject: Cognitive Neuroscience and Artificial Systems: Vision
While I was living in Japan I was hashing out the fundamental mechanics of the brain and how it could be replicated in an artificial system, an AI if you will. I had several blinding epiphanies, the sort that nearly make you drive off the road as you suddenly start assembling a more complete view of a problem that's been troubling you for a while.
The problem I was tackling this day, c.2002, was the idea of image recognition. It's hard for machines, because the images they're presented are, basically, a bunch of colourful pixels. As I was driving across the bridge it suddenly dawned on my that one of the things that makes visual analysis so easy for us humans is an awareness of ourselves: we are mentally oriented and can recognize things around us because of this context.
Consider the place I was in at the time: Driving across a bridge that spanned a railway line. I knew I was in a car, I knew I was on a bridge. I knew from experience the height of my eyes from the ground, I knew how things get smaller in the distance. I was not trying to tear a two-dimensional image into its component pieces, instead I was using my contextual awareness to more quickly slot things into place.
Grey, flat, below me, white lines on its sides. That'd probably be the road then, and the thing blocking my central forward vision is another vehicle, recognized by its location (on a road), and its other features: matching my speed, red taillights, windows.
Knowing where the image comes from and where the viewer is makes a massive difference in our ability to comprehend the things we see.
This brought me to the next epiphany rather rapidly: how we learn about ourselves, our bodies, and start to place ourselves in our world. The thing is, in order to have this context, you need to build it, and how does that happen?
That's for the next post. =)
(This was also posted to MyOpera blog)
The problem I was tackling this day, c.2002, was the idea of image recognition. It's hard for machines, because the images they're presented are, basically, a bunch of colourful pixels. As I was driving across the bridge it suddenly dawned on my that one of the things that makes visual analysis so easy for us humans is an awareness of ourselves: we are mentally oriented and can recognize things around us because of this context.
Consider the place I was in at the time: Driving across a bridge that spanned a railway line. I knew I was in a car, I knew I was on a bridge. I knew from experience the height of my eyes from the ground, I knew how things get smaller in the distance. I was not trying to tear a two-dimensional image into its component pieces, instead I was using my contextual awareness to more quickly slot things into place.
Grey, flat, below me, white lines on its sides. That'd probably be the road then, and the thing blocking my central forward vision is another vehicle, recognized by its location (on a road), and its other features: matching my speed, red taillights, windows.
Knowing where the image comes from and where the viewer is makes a massive difference in our ability to comprehend the things we see.
This brought me to the next epiphany rather rapidly: how we learn about ourselves, our bodies, and start to place ourselves in our world. The thing is, in order to have this context, you need to build it, and how does that happen?
That's for the next post. =)
(This was also posted to MyOpera blog)
BLEARGH
This post was edited on 2007-12-07, 11:53 by