How does the normal pixel based recognition done by us differ from the kind of recognition our brain does? Definitely our brain can recognise a pixel based image and in fact the sensors in our retina should be digital in nature. SO where does the difference really come. Take a look at the picture below and try to recognise whatever is there in it.

It appears to be some kind of a random noise. I don't think it is possible for any brain whowever highly developed it is to recognise something like this, just this! Once I present u with the context you will be able to identify it correctly, if not recognise. It was actually cropped out of another image where it becomes obvious as to what it is. Infact it becomes so obvious that the question turns silly. Take a look at the context, I mean from where it was cropped out, below.

It is obvious that they are people watching a race, I mean the audience for the race. Even though our brain does not recognise our surrounding on an individual basis, it can successfully do so through context. The performance of our brain depends on not what it sees, but in what way it is presented to it. Computer based recognition systems differ in a major sense with this aspect of the brain. Computer based systems can recognise an object irrespective of the context, which is why they have a poor track record when compaired to the brain.

The recognition of each of the items in this image strongly depends on the successful recognition of the other. I mean, if I present you with a part of the road, side wall, people or vehicle individually, ur brain will definitely fail to identify them, but when presented altogether it becomes obvious. So where and how does the process of recognition start, with the man, vehicle, road? Does ur brain read each and every pixel, threshold them, group them together and say that it is a road, or for that matter anything else? Is there any process of thresholding taking place at all? Does size really matter to our brain? Let me put it with an example.

Even though our programs can measure the thickness of the two letters and compare them exactly, our brain can't do it, I mean exactly. Both of them (computer and brain) show this property of very exactness and fuzziness respectively in each and every one of their processes and still fuzziness takes an upperhand over exactness most of the times in recognition.

This is where the concept of reading a pixel comes in with which I wanted to start my discussion. By reading a pixel and tracing the curves we are bound to measure them. Our main roots of recognition are based on these concepts and this is what we need to avoid.