(This is continuation to the first part of computer vision)
Sets of neurons excite one another if there is contrast along a
line at a certain direction. Higher-level networks aggregate these patterns
into meta-patterns: a circle moving upwards. Another network chimes in: the
circle is white, with red lines.
Another: it is growing in size. A picture begins to emerge from
these crude but completely descriptions.
For a few objects in controlled situations, this worked well, but
imagine trying to describe every object around you, from every angle, with
variations for lighting and motion and a hundred other things. It became clear
that to achieve even toddler like levels of recognition would require
impractically large sets of data.
A bottom-up approach mimicking what is found in the brain is more
promising. A computer can apply a series of transformations to an image and
discover edges, the objects they imply, perspective and movement when presented
with multiple pictures, and so on.
The process involves a great deal of math and statistics, but they
amount to the computer trying to match the shapes it’s is which shapes it has
been trained to recognise-trained on other images, the way our brains were.
Of course, you could build a system that recognises every variety
of apple, from every angle, in any situation, at rest or in motion, with bites
taken out of it, anything -and it would not be able to recognise an orange .For
that matter, it could not even tell what an apple is, whether it is edible, how
big it is or what they are used for.
The problem is that even good hardware and software are not much
without an operation system.
For us, that is the rest of our minds: short and long turn memory,
input from our other senses, attention and cognition, a billion lessons learned
from a trillion interactions with the world, written with methods we barely
understand to a network of interconnected neurons more complex than anything we
have ever encountered.
This is where the frontiers of computers science and more
artificial intelligence converges-and where we are currently spinning our
wheels between computer scientists, engineers, psychologists, philosophers and
neuro scientists, we can barely come up with a working definition of how minds
work, much less how to simulate it.
That does not mean we are at a dead end. The future of computer
vision is in integrating the powerful but specific systems we have created with
broader ones that are focused on concepts that are a bit harder to pin down: context,
attention, intention.
That said, computer vision even in its nascent stage is still
incredibly useful. It is in our cameras, self-driving cars. It is in factory
robots, monitoring for problems and navigating around human co-workers.
There is still a long way to go before they see like we do-if it
is even possible-but considering the scale of the task at hand ,it is amazing
that they see at all.
Posted By - Guru Charan
