In the last word of the previous page, I wrote "cquence" for "sequence". Now,
for a recognition software "cquence" doesn't mean anything because it has no dictionary meaning, but for us it is still "sequence".
A software will be able to recognise the individual letters of the word, but will not be able to interpret it. When we
read the word it gives us the same sound as "sequence" and hence the interpretation. The robustness of our brain is due to
such a parallel network. Just pattern matching would not have produced this effect.
In the previous page I said that our brain initially starts with exactness
and then transforms itself to a fuzzy based recognition. I mean to say that it starts to use knowledge to recognize whatever
it sees. One easy way to observe this is by allowing our brain to reconstruct noisy characters of a language it is unaware
of, say for example in my case Chinese. The only thing my brain knows is that Chinese is a complicated language at least in
terms of characters with lots of lines and curves. The knowledge about Chinese characters is so meagre that I wouldn't be
able to distinguish Chinese from a similar language. Now since ur brain knows nothing about Chinese it cannot use knowledge
and so has to go with the normal segmentation process, that we generally do in computers. Let the noisy as well as the actual
character be accessible to you, but do not look at the actual character before you threshold the noisy one. See how you perform.
Now ask a person who knows Chinese to do the same thing and see how knowledge improves the thresholding process.
Since I somewhat know what a Chinese character looks like, atleast I can recognize
its presence if not the character exactly. This is one more thing that I observed in a small child. My sister's son had come
to my house and we were watching a News channel. He is still learning English characters and knows nothing about Kannada
(which is my mother tongue and the News language that we were watching). There was some information in Kannada running from
one side of the TV to the other at the bottom. He started asking us as to what was being written in that running message.
I wanted to know how much knowledge he had about the different languages and so started asking him if it was written in Kannada
or English. He couldn't figure out as to what I meant. He didn't know the meaning of a language, so couldn't distinguish
between different languages. All he knew was that something was written and that we could read it to get some information.
He was only curious to know what that message was!
So how does one identify a character even though he doesn't recognise it?
How will a person who has never seen any character or heard of it respond to a character image?
The question is that, can our computers threshold an image properly
without having any knowledge about what it contains? If we are expected to know what it contains before the very process
of thresholding, then why in the first place do we need to thresholding it? So does thresholding have any role to play
in the entire visual process? How does this process start when we first open our eyes.
