Roy Schestowitz wrote:
__/ [Ajay] on Tuesday 13 September 2005 15:22 \__
Dear Martin and Roy,
...
I don't seem to have much choice than to hang in with whatever is here.
The worst you can do is complain about it. Complaints lead to bitterness,
which in turn leads to your motivation being shattered.
If it helps, think about it as going to prison for
four years. To survive your sentence it is important
to maintain a positive attitude.
Of late I thought about an idea related to "content based image
segmentation". However during literature reading I have been surprised
to find that no one is concerned about the fundamental question, What
can we define "the content" in terms of its mathematical explanation?
This would be difficult because "content" can be a variety of different
things. In photography it can be a tree, a bicycle. In biology it can be
chromosomes or fingerprints. You would often end up with ad-hoc solutions.
I believe it is good to think in terms of definitions.
Define what you mean by "content". As Roy points out,
this is obvious to humans but non-trivial for
computers.
So much work on this is being done in the big places like Berkeley and
Stanford, still no answer. If there is please let me know.In any case,
if there is any ,defining the content of an image prefectly, I think we
have solved the major problem of pattern recognition.
I personally like the AI/machine learning approach whereby you feed (train)
a classifier and then assign labels to objects. Someone is already
exploiting 'free labour' for training purposes:
http://www.schestowitz.com/Vision/modules.php?name=News&file=article&sid=35
Here, Roy is suggesting a definition of "I know it
when I see it". This is nice in that it uses humans
to do what they are good at, and computers for the
rest.
If that is not so, why not people first address this question and then
should jump into content based image processing?
Please help me with your knowledge in the area. Even suggestion to the
links and reference to the people will help me retrieve further
information..
I am afraid that my main field of exploration is registration and models.
Segmentation is something I know from afar.
My field is/was enhancement of satellite images of the
earth's surface, particularly for human visual
interpretation. I am of even less use than Roy.
Help me do a better PhD
-Ajay
The best of knowledge often lies in the literature. Quite frankly, however,
I don't believe much in books as they are old-stylish. While they are
rather reliable, they are too sequential, and flat. Ask yourself a question
and then do a Web search to find an answer which you can merely trust. This
works well for me. Remember to quickly hop between citations (links) where
appropriate and form your own hypotheses.
Roy
Think of it this way. PhD research must be of a
sufficient quality to be publishable, and to
publish research is to push back the frontiers of
science. But to push back the frontier you first
have to reach the frontier. You do this in steps. A
textbook or two is the first step, followed by journal
papers (use the Science Citation Index to find these),
followed by conference papers, followed by doing
actual research. This takes you to the frontier,
from where you can start to form a research project
and to do research.
--
Regards,
Martin Leese
E-mail: please@xxxxxxxxxxxxxxxxxxxxxxxxxx
Web: http://members.tripod.com/martin_leese/
|
